❌

Reading view

There are new articles available, click to refresh the page.

Scottish Web Folk 2026: UX team reflections

With others from the Website and Communications team, Nick and Mel headed up to Dundee to attend the Scottish Web Folk conference. In this post, they share their reflections on the day.

Nick’s reflections 

This was my fourth trip to Dundee for the Scottish Web Folk conference and my third time presenting. Scottish Web Folk is a great chance to hear from people working in digital roles in higher education, and catch up with how people across the sector are approaching similar challenges. 

We presented about optimising content for AI summaries 

This year, Mel Batcharj, Katie Spearman and I presented about AI summaries, such as the ‘AI Overview’ that Google now presents at the top of a search results page. These summaries take information from our websites and repackage it for users, who in some cases no longer click through to the websites themselves. 

In our presentation, we talked about how to create content that is optimised for these summaries, and how we created a Content Improvement Club session to address this topic. 

We emphasised that much of the guidance for creating content optimised for search engines hasn’t completely changed in the age of AI. It’s still important to write clear, structured content that addresses user needs and presents people with the information they need to take their next step. It’s also still the case that following good  digital accessibility practices will help AI tools and search engines to process the information included in your content. 

Read Katie’s write-up of our  Content Improvement Club training session ‘Is your content AI ready?’ 

Speakers discussed the theme of ‘Human versus machine’ 

There were various talks looking at how we can use AI to support and inform our work with websites in higher education. 

In the panel discussion, Marta Krukowska (Web and Digital Manager at Bangor University) described her experiments with using AI tools to interview colleagues to gather information for course pages. If asked to describe their course, colleagues sometimes provided vague information. For example, they might talk about how the university provides access to “state-of-the-art facilities”. But the AI interviewer could push back and ask for the names of specific facilities, and what their availability means for students. This turned out to be a more efficient way to get the information needed to write marketing copy for a course. 

The panel also addressed questions of how prospective students are using AI tools to support their search for a university course. The discussion made me reflect on how difficult it can be to identify a starting point for a complex user journey like this. It’s easy to fall into the trap of thinking that a user journey is a linear process or a series of steps, but this frequently isn’t the case. When people are thinking about what they want to study and where they want to study it, there are typically a wide range of information sources feeding into that decision. The information is coming in from these disparate sources over a broad period of time. So it’s probably not accurate to say that the journey starts with an AI search, even if that forms part of the information environment that a prospective student is using. 

The Prospective Student Web team presented about their content playbook and how they’re appraising degree finder 

Two presentations I particularly enjoyed during the day were from our Edinburgh colleagues Lauren Tormey and Pete Watson. 

Lauren was presenting about how her team had put together a content playbook: a set of patterns that staff can follow when working with content. The playbook helps content designers to take a consistent approach when using page elements such as feature boxes, pull quotes and page summaries. It was interesting to hear Lauren’s account of how they found alignment within their team to produce the guidance in the playbook. I’d sat in one of the alignment sessions myself and liked how everyone had had a chance to put forward their views by editing the same page of content. In her talk, Lauren described how meeting up to discuss a piece of content helped to clarify shared standards and supported a more consistent approach to editorial work. 

Pete’s talk was about how he has been appraising the performance of Edinburgh’s degree finder. Pete used a scorecard similar to the one Lou Rosenfeld describes in his article ‘Stop redesigning and start tuning your site instead’. The scorecard lays out your most important audiences, and for each one, a prioritised list of their tasks and information needs. The final column on the scorecard is a grade for how easy it is to complete those tasks or find that information on your website. 

The challenge then was to find a robust way to give these grades. To do this, Pete created a set of marking criteria of the type you might see associated with educational assessments. With this in hand, he ran a series of usability tests to see how easily people could complete tasks on Degree Finder. The result was a table showing how effectively we’re supporting prospective students to complete their key tasks, and highlighting where further development effort is needed. 

Read Pete’s blog post about appraising the new postgraduate study website 

Thanks to the Dundee team for hosting the event 

As always, the day was brilliantly organised, with a wide range of interesting talks. We’re still learning about how AI is changing our use of digital tools, and this event was a good chance to reflect on how this is having an impact on those of us who work on websites in the higher education sector. 

Mel’s reflections 

This was my second time attending Scottish Web Folk in Dundee, but my first time presenting. I’d been looking forward to returning, as it’s always a welcoming room and a great opportunity to hear what others across the sector are working on. 

We took our AI research into a wider conversation 

Our session, ‘AI search summaries: a new lens on old content problems’, built on a Content Improvement Club training session we’d run recently. Preparing that training was the first time I’d really explored how AI relates to the content we produce, and it involved a lot of research. So it felt like a nice full-circle moment to bring some of that learning to Scottish Web Folk and share it as part of a wider discussion about AI. 

Our presentation focused on AI and content design, while some of the other sessions took a more technical angle. It was interesting (and reassuring!) to hear that our points resonated with others, particularly the idea that well-structured, maintained and accessible content matters for AI as well as for people. 

It was also a lovely surprise to find Claire Gregory from the University of Dundee in the room. Her UCISA talk, ‘Feeding the bot: making content work beyond the webpage’, had been a really useful source when preparing our training. 

Watch Claire Gregory’s talk: ‘Feeding the bot’ 

We also had questions about the training itself and how we get people involved. Our sessions have developed quite a bit over the past year, so it was nice to have a moment to reflect on that and chat about the process with a room of peers. 

Andrew Millar’s opening talk put AI in a wider context 

Beginning the day, Andrew Millar’s got us thinking about how we approach new technology. He took us through a timeline of major technological turning points, from the printing press to the web, and asked what became less scarce, what became more valuable and who lost out. With the printing press, for example, copying text became easier, but judgement and trust became more important. 

We’d actually considered using a timeline in our own presentation, so it was interesting to see a similar approach being used elsewhere. It felt like a useful way to tackle the AI elephant in the room, putting the changes we’re seeing into a wider context rather than treating AI as something entirely without precedent. 

A new acronym has really stuck with me from Andrew’s talk: GIGO, or ‘garbage in, garbage out’. It’s a simple idea, but an important one in the context of AI. If the information we’re putting out there isn’t good in the first place, we can’t expect the systems using it to produce reliable answers. Definitely relevant to what we were talking about! 

Andrew also talked about how, more than ever, we need to advocate for users who aren’t in the room. That really resonated with me. I don’t think the rise of AI search tools means we should start rewriting everything for bots. Rather, it’s another ‘audience’ to consider, while keeping our human audiences at the centre. The things we already think about as content designers (understanding user needs, using plain language and making content accessible) still matter.  

The morning panel discussed making university information work harder for prospective students 

The panel discussion was another highlight for me. I enjoyed the conversational format, which offered something a little different from listening to a series of presentations. 

A key part of the chat was around how prospective students’ questions go well beyond course details. They’re also thinking about the cost of living, student welfare and the practicalities of studying, including whether they’ll be able to work alongside their course. These priorities are changing, and universities need to be upfront about the answers rather than burying the lead. 

This is particularly important as AI search tools may draw on information from outside university websites, such as forums and online discussions. If universities don’t provide clear answers themselves, AI may turn to other sources that aren’t always accurate or complete. There’s an opportunity here for universities to answer people’s questions directly and reliably. 

I also enjoyed hearing about Marta Krukowska’s (Bangor University) example of using AI to introduce a bit of friction when gathering course information from colleagues. Rather than accepting vague descriptions, AI could push people to dig deeper and provide more specific answers. I liked this idea of using AI not simply to speed things up, but to challenge people to provide the useful detail that prospective students actually need. It was another example of finding a suitable use for AI that wasn’t about creating content. 

A day of shared learning 

Overall, it was a really good day and well worth the early start. The ‘Human vs machine’ theme could easily have made the day all about the latest AI tools and technologies. What I took away instead was how much of the conversation came back to the human side. 

Big thanks again to the Dundee team for hosting.

How to create consistent content using the editorial style guide: What we covered in October’s Content Improvement Club

We’ve been refreshing the editorial style guide to make it more useful for colleagues across the University. In October’s Content Improvement Club, we explored why consistency matters, what’s changing and how you can use the guide in your own work.

You can read more about the reasons behind the project in Nick Daniels earlier blog post:

Why we’re refreshing the editorial style guide

Using the editorial style guide to make consistent decisions

The editorial style guide helps us make consistent decisions about how we write and present information. It’s designed to help us create content that’s more accessible, readable and inclusive.

Editorial style guide

It covers a range of everyday editorial decisions, including:

  • when to capitalise something and when not to
  • how to spell words that have multiple possible spellings
  • how to explain abbreviations and acronyms

Exploring consistency through examples

In the session, we started by looking at some everyday editorial decisions to explore how the style guide can help us create consistent content.

How we write the University’s name

We looked at the different ways we might write the University’s name. It’s surprising how many variations we can come up with for something so familiar.

But using different versions can be confusing, particularly for people who don’t know the University as well as we do. A name is an important part of our identity, so we need a shared approach rather than everyone deciding what feels right.

Luckily, we don’t have to guess. The editorial style guide provides clear guidance (the correct wording is the University of Edinburgh).

University terminology style guide page

A slide with the question 'What are we called?' in the centre. A group of blue squares surrounds this, containing a variety of ways people have referred to the University of Edinburgh. These include 'UoE', 'Edinburgh Uni', 'Edinburgh University' and the correct version, 'The University of Edinburgh'.

How we write dates

We also looked at some of the smaller editorial decisions we make every day, using dates as an example. We showed four date formats in a poll and asked attendees which they would naturally use when writing University content.

Our options were:

  • A) 27/10/2026​
  • B) 27 October 2026​
  • C) 27th October 2026 ​
  • D) 27th Oct 2026  ​

Looking at the options, we discussed how it’s not that one format is universally right and all the others are wrong. Different formats can be appropriate in different contexts. For University content, though, we’ve agreed to use option B) 27 October 2026. Following the same convention helps us keep our content consistent, wherever it appears.

Dates and days editorial style guide page

Why consistency matters

Having looked at some practical examples, we asked attendees why they thought consistency was important when creating content. Using sticky notes, people shared their ideas. There was a lot of overlap between the reasons they identified and the ones we’d thought about beforehand.

One reason is brand reputation. We want people to recognise us as one University, wherever they come across our content. Using consistent terminology and conventions helps build that recognition and trust.

Consistency also supports readability and accessibility. Familiar, predictable formats make content easier to scan, understand and use.

It contributes to clarity and accuracy, too. Having an agreed approach helps avoid confusion and makes it easier for people to understand the information we’re sharing.

Finally, consistency helps create a joined-up user experience. Whether someone is visiting our website, reading an email or looking at social media, we want their experience of the University to feel connected.

Putting the style guide into practice

The session was a little different from some of our previous Content Improvement Clubs. Rather than spending most of the time presenting guidance, we wanted to give attendees the opportunity to explore the style guide for themselves.

We included two practical activities to help people get familiar with the guide.

Exploring the guide through example pages

We looked at two mocked-up webpages we’d created. Both contained deliberate editorial inconsistencies and mistakes. Attendees explored the guide for themselves, find the relevant sections and use the guidance to identify possible improvements.

They used sticky notes to highlight things they’d spotted and consider what they might change. This included things like flagging unexplained acronyms, link text that needed to be placed on a new line and incorrectly formatted bullet lists.

A page titled 'Tuition Fees and Loans for Students from Scotland' with eight sticky notes annotating issues with the content. These include comments like 'Use sentence case for page titles', 'Do not use all caps for emphasis' and 'Explain acronyms on first mention'.

The first example of a mocked-up page used in the session

A page titled 'Sports clubs' with 11 sticky notes annotating issues with the content. These include comments like 'Use -ise endings instead of -ize endings', 'Avoid ampersands' and 'Bullet list items do not end in punctuation'.

Second example of a mocked-up annotated page from the session

This gave everyone a chance to become more familiar with the guide and see how it can support everyday decisions when creating and editing content.

Where should you start with the style guide?

One thing we hear from colleagues is that there’s a lot to remember in the style guide. It can be difficult to keep every convention in mind when you’re busy creating and maintaining content.

Our advice was not to try to remember everything. Instead, we suggested starting with four priority areas:

  • Acronyms and abbreviations
  • Headings
  • Links
  • Dates and numbers

These are useful areas to focus on when reviewing your content. It’s also worth asking someone else to read it over. A fresh pair of eyes can often spot inconsistencies or things that you might have missed.

Reviewing our own content with other publishers

For our second activity, we used these four priority areas as a starting point. Attendees worked in breakout rooms to look at content from their own websites. People took turns sharing their screens and reviewing each other’s content.

The breakout rooms generated some interesting discussions about the practical challenges of applying editorial guidance. For example, we talked about using links within bullet lists and how much information to divide into separate headings and sections.

These decisions don’t always have a straightforward answer. The important thing is to think about what works best for the people using your content and use the guidance to help inform your decisions.

The activity also gave publishers a chance to get feedback from colleagues they might not usually work alongside and look at their content from a different perspective.

Experimenting with ELM to check content against the style guide

In the session, we also talked briefly about whether ELM (the University of Edinburgh’s AI innovation platform) could help publishers check their content against the editorial style guide.

We haven’t always had reliable results when using ELM for this purpose

As John Wilson explored in his February 2025 blog post, ELM can sometimes identify errors and correctly explain why they don’t follow our guidance, but it can also miss other errors or refer to guidance that doesn’t exist.

Testing ELM’s ability to return useful results and prompts about the editorial style guide

A text-based Markdown version of the style guide is what we’ve tested more recently

We’ve been experimenting with giving ELM the style guide as a text-based Markdown file. This means it can use the actual guidance to check a webpage against, rather than relying on its existing knowledge.

We tried this approach with the two mocked-up pages used in the session. ELM was able to identify the majority of the deliberate inconsistencies, although it still missed some issues, including an example relating to capitalising ‘University’ when referring to the University of Edinburgh.

It’s an interesting experiment, and one we’re continuing to explore. While tools like ELM may be able to help identify potential issues, they should not replace checking the guidance yourself and using your own editorial judgement.

We shared a couple of examples of what we’ve been working on in the guide

It’s been a big undertaking working through each section of the guide. We wanted to give attendees a flavour of the work by looking at a couple of examples.

Bringing University-specific terminology together

We’ve expanded the previous ‘Referring to the University’ page into a broader University terminology page. You can read more about the development of this page in our blog post:

Creating a new style guide page on University terminology

Restructuring the punctuation guidance

We’ve also restructured the punctuation guidance into separate pages.

Punctuation section of the editorial style guide

Previously, much of this guidance was grouped together using accordions. We wanted to move away from this format, which can make content less accessible and harder to navigate.

We’ve reviewed how the different topics are grouped, too. For example, colons and semicolons now sit together, making it easier to compare the two and understand when to use each.

Separating the guidance gives us more flexibility to provide the right level of explanation and include examples where they’re most useful, without simpler guidance getting lost among more detailed sections.

Keeping the guide up to date

We want the editorial style guide to remain a useful, living document rather than something we publish and leave untouched.

Alongside those mentioned in this blog post, some of the changes we’ve made to the guide have been much smaller tweaks or clarifications. We’ve been logging changes on the guide’s updates page, so if you’re interested in following how it continues to develop, you can see what’s changed there.

Updates page (editorial style guide)

How to hear about future Content Improvement Club sessions

We promote these sessions via our mailing list. If you’re interested, please sign up:

Join the UX and Content Design mailing list (University login required)

Suggest a topic for a future session

We’re keen to continue covering topics that colleagues across the University would find useful. It would be really helpful if you could let us know any ideas you have using this form:

Suggest a topic for Content Improvement Club (University login required)

Other training that we offer

More training is listed on the User Experience Service website:

Training | User Experience Service

 

Just do it for me? Balancing human and AI input in content improvement

Working with student interns and contributing to the Drupal community, I’ve experimented with AI in various  tools designed to support content improvement. I reflect on what I’ve learned about peoples’ preferences for engaging with AI tools and consider wider design challenges for web content to support co-browsing.

Writing good quality, user-centred content has always been a challenge, and accordingly, a suite of tools evolved to address this; Hemingway, Grammarly and many more. With their ability to learn patterns and predict word sequences, the advent of LLMs provided a new direction for content improvement. Where previous tools had used fixed rules, algorithms and formulas to flag content problems and score sentences, LLMs changed the playing field as they were able to be prompted to rewrite and create text and to adapt words and sentences not only in line with fixed rules, but also aligned to different interpretations of style.

The UX Service’s content design course, Effective Digital Content, and our ongoing Content Improvement Club meet-ups are both designed to help staff publishing digital content for the University. It’s been encouraging to see a positive uptake for both training formats, however, we realise we can do more to help staff publish the best quality content they can, and to this end, for the past few years it’s been a privilege to support a series of interns in the UX Service to look at how we can innovate and apply AI to the persistent challenge of writing good quality, user-centred content.

Thinking beyond – embracing the spirit (or grasping the nettle, depending on how you look at it) of AI innovation, it’s sensible to note that, more broadly, the browsing landscape of the web is increasingly shifting to include AI agents as well as humans, giving cause to consider what using AI to design content to be consumed by AI might look like. That’s the subject of another blog post (or several).

For now, as I begin another phase of work with a new student in the role of AI and UX Innovation intern (Viktor Sharha), it felt timely to pull together what I’ve learned so far.

Designing, creating and editing content happens in various ways

When designing any user-centred assistive features, a good place to start is the current workflow or user journey people adopt to achieve the task in question. In other words, to look at the stages content editors currently go through to make sure their content is as good as it possibly can be.

One of my earliest research tasks when I joined the University was to usability test the Editorial Style Guide, to find out how publishers used it to keep their content consistent in EdWeb. This work gave me insight into the type of workflows publishers followed to get compliant content published. Some publishers worked in Microsoft Word, using the in-built spelling checker for typos, and cross-checking against a printed-out copy of the style guide for University-specific errors before pasting content into the EdWeb body text. Others went straight to EdWeb, checking their content line-by-line with a PDF version of the guide. Others relied on colleagues giving their work the once-over before pressing publish. Some used Grammarly and Hemingway, some didn’t.

Overall, there wasn’t a dominant content-publishing workflow, with unified friction points waiting to be alleviated through a standard user-centred design process. Instead, there were varied approaches, habits and ways of working which people had adapted over time to suit their specific circumstances and to address the areas they found difficult as individuals. Tools were used at different stages of the process, and to a greater or lesser extent depending on personal preferences.

Therefore, designing content improvement tools isn’t one-size-fits-all

It follows that, since people follow different pathways to create, design and edit content, tools to assist this process need to be developed in different ways to suit varied preferences. Since 2025, interns in the UX team have prototyped our own different content improvement tools to test with publishers, and we’ve also tested open-source tools that exist already, developed within the Drupal community. From our testing and research, we found pros, cons and nuances associated with different tools. The summary below pulls together detail of the tools and how they compared.

Tool 1: Drupal Editoria11y – no AI, flags errors in the editing interface

Editoria11y is a Drupal module built by John Jameson from Princeton University that automatically checks content inside the editorial interface. It was originally designed to catch accessibility issues, but it can be configured to check content against other rules too, such as institutional style guide conventions.

What it does and how it works: Editoria11y checks content as authors type, and also on previews and published pages. It includes an indicator bar to show how many issues are on the page and has inline markers to show where each issue is, as well as a dashboard summarising site-wide errors. When an issue is opened, a pop-up modal explains the problem and suggests a fix.

What worked well: In tests, publishers were able to make sense of the inline markers and understand how these related to the indicator bar for the whole page. They liked the tracked change visualisation in the pop-up modals, and were able to work through successive modals, accepting or ignoring the suggested changes.

What needed improvement:  Some publishers found the number of markers overwhelming, and since the pop-ups covered the text, it was difficult to judge each recommendation in context. In the pop-ups, some of the rule descriptions used formal grammatical terms (such as ‘ordinal suffixes’) which were unfamiliar to some publishers meaning they were confused about the problem to be fixed. Since rules were applied mechanically, there were instances where errors were missed, or where awkward fixes were suggested (such as swapping ‘1st floor’ for ‘1 floor’ due to straight removal of the ordinal suffix) – these degraded publishers’ trust in the tool.

Read more about what we found testing Drupal Editoria11y

Embedding the rules: What we learned UX testing a style guide helper tool built with Drupal Editori11y

Tool 2: ELM-powered content design chatbot – with conversational advice

This tool was developed by student Mostafa Ebid during a 2025 summer internship. It connected ELM, the University’s AI platform, to EdWeb2, so publishers could get help with their content by conversing with the University’s suite of LLMs without leaving the Drupal editing interface.

What it does and how it works: This tool sits in a chat sidebar next to the main Drupal editorial interface and offers three modes: Write Content, Design Content and Proofread. In response to a user prompt to improve content, it activates an orchestration of agents (each with specific tasks such as checking the style guide, structuring pages, profiling audiences) which read the page content and act on it accordingly, publishing the revised content in the chatbot window.

What worked well: In tests, publishers understood how the tool could be used as a writing partner, and were impressed with the way it restructured content with headings, bullet points and calls to action and hyperlinks.

What needed improvement: When faced with the empty chatbot interface, some publishers weren’t sure where to start, and they didn’t notice the three modes to work on content. Some were also unsure of the sources the chatbot was using for its revised content and skeptical that the revised content contained hallucinogenic content. They would have appreciated tracked changes to be able to pinpoint what had changed and found the long thin sidebar chat interface awkward to review the updated text. The chatbot was also limited to reviewing body text content, whereas publishers needed it to review content in other page components used in Drupal.

Read more about what we found developing and testing the ELM-powered content design chatbot:

Initial insights from UX testing our Drupal AI content assistant tool

Tool 3: QuillMark v1 and v2 – some AI, clusters errors with correction suggestions

QuillMark was an AI-enhanced tool which worked in Drupal to check content against the University’s Editorial Style Guide. Shlok Gupta built and tested two iterations of Quillmark during his 2026 summer placement as AI and UX Innovation Intern.

What it does and how it works: QuillMark sits in a side panel of the Drupal editorial interface and scans content on the page in two stages. The first stage uses code-based pattern matching (without AI) to catch straightforward errors. The second stage brings in AI to find deeper issues (such as passive instead of active voice) and to suggest fixes. Findings from both stages are displayed in the side panel – each finding describes the problem, links to the relevant rule in the style guide and offers an ‘Apply fix’ button (which can be reverted). Findings are grouped by category, and this panel also includes an indicator of the environmental impact and cost of the scan.

What worked well: In tests, publishers understood that the tool was there to help them correct content on the page, they liked the environmental and cost indicator, the suggested fixes and the links to the style guide reassured them that the tool was applying University-specific rules.

What needed improvement: Publishers reported feeling overwhelmed by the number of findings on a single page and they wanted a way to deal with related findings and fixes in bulk, to avoid going through them one-by-one. Related to this, they wanted

Read more about what we found developing and testing the QuillMark tool:

QuillMark, iterated. How user research helped us evolve an AI-enhanced Drupal style guide assistant

Tool 4: Drupal AI Content Review – full AI, with a plan provided to improve content

As part of the Drupal AI Initiative, I’ve contributed to shaping the self-explanatory Drupal AI Content Review module. This module plugs into a Drupal site, references content rules and good practice from context items in Drupal Context Control Center (within the same site) and uses the configured AI provider to review site content

What it does and how it works: The feature reads the page content and checks it against criteria set by the site owner, such as tone and voice guidelines, inclusive language rules and target readability levels. These rules are stored in the Drupal Context Control Center, along with examples of good and bad content. The review gives the page a score for each criterion and recommends fixes to improve it. Accepted recommendations become an improvement plan, which is applied in the editorial interface. Re-reviewing the page shows the updated scores, with the option to keep making improvements. Over time, the rules can be refined by adding more examples of good and bad content to the Context Control Center.

Read more about Drupal AI Content Review on Drupal.org

Read more about the Drupal Context Control Center on Drupal.org

View a demonstration of AI Content Review in a mocked-up university website as part of the Drupal AI demo on the Drupal Forge platform

What worked well:  In tests of this module within the AI demo, people liked seeing the red, amber and green indicators against page content for an easy visual highlighting areas to be worked on, and they liked being able to select a checklist of recommended improvements to apply to improve their content.

What needs improvement: The screens explaining the recommended improvements were difficult to scan read, and it was also difficult to compare the original text and the text with the recommended improvements side-by-side (despite colour shading to denote which was which). The stage where the recommendations were turned into an improvement plan felt like an unnecessary extra step, it would have been better to have implemented the recommendations directly. Also, it was presumed that implementing the plan would have corrected all errors resulting in a score of 100%, without the need for further work and re-review. Finally, it wasn’t immediately clear how to use the Drupal Context Control Center and the AI Content Review together.

When designing AI-enhanced tools, consider three AI design maxims

Taking on board variations in blending human effort and automated operations, there are three approaches to designing AI features and AI-assisted workflows worth considering. These are defined below, with ideas for how they may apply to content improvement.

1.      Human-in-the-loop

In this set-up, humans typically share the task-load with AI – actively working alongside an automated operator to supervise or influence actions, and to approve AI decisions before they happen (ensuring AI outputs are not released without the green-light from a person).

Human-in-the-loop tends to be adopted when:

  • there is a need for human judgement to get something correct
  • the consequences of an error could be high-profile or damaging
  • trust in an automated tool is still being built, and the boundaries of its capabilities are still being realised
  • the goal is for the system to facilitate shared learning between the person and the automation

For content improvement, human-in-the-loop is appropriate in these scenarios:

  • ensuring correctness of important factual data – for example, prices, dates, contact details, terms and conditions
  • checking the tone of sentences in line with inclusive language guidance
  • verifying accuracy of high-stakes published content such as legal information, policy wording or written instructional content
  • rewriting complex topics in plainer language
  • writing meta descriptions or page summaries to be accessed by search mechanisms
  • translating content into other languages

2.      Human-on-the-loop

These systems involve the AI or automator acting on its own, with the human reviewing and spot-checking at intervals, monitoring output quality over time, and intervening to correct failures periodically, where required.

Human-on-the-loop tends to be adopted when:

  • there is a high volume of low-risk tasks to be completed, which would be tedious for humans to do themselves
  • consequences of errors are minimal
  • there’s a need for tasks to be completed quickly, and including human-input at each task would be too time-consuming
  • consistency is required across tasks, and there’s a risk of human error introducing variability

For content improvement, human-on-the-loop is appropriate in these scenarios:

  • fixing straightforward style rule violations such as correcting spelling errors
  • categorising and tagging text content (for example, to facilitate its discovery in search)
  • scanning content for readability and accessibility, to flag or report areas to be addressed

3.      Human-out-of-the-loop

In this arrangement, the human sets up the AI to act autonomously, and leaves it to proceed, with no routine review.

Human-out-of-the-loop tends to be adopted when:

  • there are very low-risk, behind-the-scenes tasks to be completed where errors have little or no consequence and can be fully reversed

For content improvement, human-out-of-the-loop is appropriate in these scenarios:

  • autocorrecting typing errors
  • detecting content duplication
  • auto-optimising media content to defined parameters

Choice, pacing and transparency are constant design requirements

Testing four different tools has showed that options always need to be available, whatever balance of human and AI the tool includes. People approach AI differently based on preferences and past experiences. Some love nothing more than a good grappling of a sticky sentence, some just want an agent to do the work for them, and many sit somewhere in between, happy to have some AI help as long as they have the final say.

Alongside humans in, on and out of the loop, it’s important to remember the fourth option: human only, where people can choose to do their content unaided, using the methods they relied on before AI came along. AI shouldn’t be forced on anyone, so all tools need to come with a ‘no thanks to AI’ option.

AI is a fast-evolving and non-deterministic technology, so at any point in time it may work differently to the way it did before.  Responding to challenge from a design perspective, it’s important to factor in pacing. People need to be taken through how a tool works, and the options available. They need time and space to experiment and try it out, so they can learn for themselves how it can work for them. Pacing not only matters in the way tools are introduced, but also in the way their outputs are presented, to avoid the overwhelm that comes when AI does the heavy lifting all at once, and so people know what changed and how, and they have a way to accept, decline or edit what it produced.

If people don’t trust an AI tool, they won’t use it. Earning trust requires transparency, explainability, relatability and reliability – so AI tools need to be upfront about what they can and can’t do, with clear, intuitive interfaces and labels. Reliability will always be hard to promise given AI is probabilistic, so instead of accuracy guarantees which may fall short, a better approach is to design for experimentation, so people can find their feet and make their own judgements.

Designing for humans and AI together – metadata, markdown and MCPs

Looking at the broader digital landscape, the challenge widens from designing tools that help humans make content to designing websites with content ready for both humans and agents to use.

In co-browsing, people are increasingly arriving at a website with an agent alongside them, handing off some tasks while keeping control of others and wanting to see what their agent is doing. That means firstly deciding which tasks belong to humans and which one to AI, and secondly designing for each. Humans need intuitive interfaces. Agents need data they can read and act on without ambiguity:

  • metadata, such as schema markup that tells an agent what a page is about, who it’s for and how it connects to other content
  • markdown versions of pages and llms.txt files, which give agents clean, structured text without layout and navigation getting in the way
  • MCPs (Model Context Protocol servers), which let sites expose what they can do as well as what they say, from simple actions declared directly in HTML to more complex ones written in JavaScript – in preparation for use by the emerging WebMCP browser standard.

Labels matter for humans and agents alike, because words are becoming a currency that machines read as well as people. Alignment on terminology facilitates the creation of a single source of truth, to prevent the human experience from drifting from the agentic experience. On another level, humans need a way to consistently communicate with AI – to give AI the rules and instruct it and to train it with new insight. ‘Rosetta stone’ mechanisms like the Drupal Context Control Center will play an important part in designing for human and AI interaction and collaboration and I am excited to keep working and researching in this area to address these evolving design challenges.

 

Show me your sources! What students make of content provenance data, AI signifiers and sparkles

The line between human-made and AI-generated content is blurring fast. I added a disclosure indicator to an AI-powered search tool and tested it with students, which taught me about perceptions of AI-influenced content and what signals authenticity. I reflected on the need for a shared language around provenance and disclosure.

Estimations of the proportion of AI-generated content on the web vary and are becoming progressively difficult to measure. New legislation advocates for disclosure of AI use, but the nuances of communicating how AI has been used within a feature, and detailing the finer points of the provenance of AI-generated content both remain a challenge.

Earlier in 2026, I took part in a study to design a C2PA provenance indicator into a prototype AI searchbot. I revisited my designs to test them with students, to assess if this authenticity data made sense and to understand its impact on students’ potential trust of the AI searchbot. In this blog post, I describe what I tested and what I learned, and make broader considerations of designing with provenance data, the impact of AI legislation and what makes people trust AI features.

People’s trust in technology depends on explicit and implicit factors

The ISO 9241-210 standard for UX defines user experiences as ‘a person’s perceptions and responses that result from the use of anticipated use of a product, system or service’, which sums up the influence trust can have on the way a person experiences a technology.

I was interested to explore more about the relationships between human trust and AI, so I read a selection of papers on the topic and extrapolated four factors that commonly affect how people trust AI technologies:

  1. Transparency (surface and presumed credibility)
  2. Reliability (earned and reputed credibility, how consistently it works and how dependable it is)
  3. Explainability (how clear the purpose of the AI is and how clear it is how it works)
  4. Relatability (how well the AI supports the person’s interests, how easily they can use it and how it resonates with them)

When it comes to designing AI features people trust, and communicating how AI has been used within those features, it’s important to try and convey each of these factors in ways that make sense to the user in context. In other words, striking a delicate balance of providing sufficient trust indicators to allow people to make reasoned judgements, without going overboard and running the risk of confusion or skepticism.

The EU AI Act has set out obligations to disclose AI usage

From an ethical perspective, advising people that they’re using an AI feature has always been the right thing to do, in order to foster a sense of user trust. Legislation published in August 2026 made the requirement to disclose AI usage more explicit and provided more detail. Article 50 of the EU AI Act details Transparency Obligations for Providers and Deployers of Certain AI systems and includes recitals which cover areas including the following:

  • Designing AI systems to ensure users know they are talking to AI
  • Marking content from AI systems (text, images, videos, audio) so it is detectable as such
  • Informing people of AI systems that include emotion recognition or biometric categorisation
  • Disclosing deepfakes (audio/video/image content that has been manipulated as well as public-interest text, unless human-reviewed with editorial insight) as AI-made
  • Clarity, timing and accessibility of AI disclosures – to be provided at the first interaction or exposure and to be clear and accessible.

With this legislation now in place, it’s anticipated that a growing number of ways to disclose AI usage will emerge, making AI disclosure as mainstream as AI features.

I adopted the C2PA digital content provenance standard into an AI searchbot prototype

When it comes to icons that disclose AI usage, sparkles, neurons, and robots are all widely used, but they only signify that AI is present or available, they don’t provide any detail of how it has been used or how it has shaped content. C2PA is an established provenance standard which has been designed to convey progressive levels of content provenance data. I incorporated C2PA to disclose how AI was used to shape synthetic chatbot responses in a search feature and wrote about my process in an earlier blog post:

How can people trust AI-generated content? Designing provenance data into our prototype AI searchbot

I concluded this design work with a resolve to test my designs with users to understand the following:

  • if they cared about seeing provenance data about the AI searchbot response
  • if they grasped the concept of provenance data
  • if they understood what the C2PA Content Credentials icon signified
  • if the names for the different categories of data made sense to them

Drawing on my recent reading into trust factors, and taking into account the new EU AI disclosure guidance I also wanted to use my designs to understand more about how users make sense of AI features, and what affects their trust of them in context.

Testing revealed students’ responses to progressive levels of AI disclosure in the searchbot

In a simple test, I asked the student participants to imagine they were in a scenario where they had used the AI searchbot to find information from a website. I then showed them mock-up screens with the indicator in progressive levels of disclosure, going from the C2PA ‘CR’ logo, to ‘content credentials’, ‘About this response’, ‘Certification data’, ‘Digital source data’ and ‘Transformation processing data’. At each stage, I asked them to speak aloud to describe what they saw, what they thought it meant, and how it resonated with them.

Screenshot showing the AI searchbot interface showing all the provenance data available

Screenshot showing the AI searchbot interface with progressive levels of provenance data on display

The C2PA icon was unfamiliar but content credentials made more sense

Students’ first exposure to the provenance data was as the C2PA ‘cr’ icon which was displayed at the opening interface of the searchbot (when the searchbot was ready to receive a user query). On its own, the icon was pretty meaningless to them. One felt it could be a watermark, others thought it was a button to make the chat window bigger, others said they would Google ‘cr’ to find out what it was. The next level of disclosure was a signifier that read ‘content credentials’, which appeared once the searchbot had received a query and produced a response. Students found this indicator more meaningful – they felt it was a way to verify where the response came from, and said they would be inclined to press it to find out more.

The level 1 disclosure ‘About this response’ was too vague to be useful

When asked what they thought they would find in the section ‘About this response’, students were largely unsure. Some thought it would contain a hallucination warning, others thought it would contain sources. When they read the generic descriptor about the C2PA provenance data, which explains that the response was AI generated and had content credentials and assertions to demonstrate authencity, they felt it was too vague to be useful to them, and didn’t feel it was telling them anything new. They were unfamiliar with ‘content credentials’ as a term and by the language, they reasoned that this message was aimed at a legal team rather than someone reading the output of the searchbot.

‘Certification data’ wasn’t a term that resonated with student participants

Reading the label ‘certification data’ students had mixed ideas of what this section would contain. Some felt it would reveal intellectual property details, some thought it would show licensing agreements relating to data entered in the searchbot, and others thought it would explain how the searchbot handled data. There was a further expectation that this section would contain the sources behind the searchbot response.

When they expanded this section to see the identity data (name of the organisation certifying the content) and the associated timestamp of when the content was certified, they were unclear what was being certified (the output of the searchbot or the AI model), and how they would use this information. Some reasoned that the responses from the AI searchbot were stored in a database and that this certification data could help keep track of a bank of these responses.

‘Digital source data’ made more sense to them

Of all the data categories displayed, ‘Digital source data’ was the one students gravitated towards. This is where they assumed they would find the links to the sources used to create the AI searchbot response, and their assumptions were correct. For many students, seeing the links to the sources proved the authenticity of the AI content more strongly than any other sections of the data they had seen. As well as the source links, this section included the C2PA Actions assertion (which recorded the stages used to create the digital content (or asset) including the AI model used). Students recognised the AI model but questioned the terminology ‘Actions assertion’, thinking it would be more straightforward to just list the AI model used.

In the ‘Transformation processing data’, detail of the AI prompts was of particular interest

Preconceptions of what they would find in the ‘Transformation processing data’ section varied from a logical set of steps to show how the responses were formed from source data, to details of how data inputs would be used to train AI models, to sustainability data showing energy expenditure in the formation of the AI searchbot responses. When they saw information labelled ‘Search mechanisms’ which detailed how embeddings were created and a vector database was used in semantic matching, most students felt this was too much technical detail. Some who were familiar with the process felt this section would be more appropriately labelled as a technical stack, rather than search mechanisms. The ‘Transformation processing data’ section also contained the AI prompt, and most students responded positively to this, saying they felt it gave them a fuller picture, it helped them understand ‘the personification of the AI’ and made the process more transparent.

The effectiveness of my C2PA designs largely depended on students’ existing knowledge

Reviewing the findings together, I concluded that my designs partly succeeded in communicating details of how AI had formed the searchbot responses, but this depended on students’ perceptions and what they knew already, as well as what they wanted to know. Overall, my designs fell short when it came to instilling a sense of trust in the students, largely due to unfamiliarity with the C2PA terminology, the challenges of targeting implicit trust factors and the partic.

Students familiar with AI responded positively, but others wanted a non-AI option

The C2PA indicators focused on the provenance of the AI-generated searchbot responses, but for some students this was irrelevant as they wanted a choice to be able to not use the AI feature at all. Several students were skeptical and negatively pre-disposed to AI features based on their previous experiences and concerns about the environmental cost, and therefore upon seeing the sparkle icon realising the searchbot was AI-powered they were sought for a way to turn it off, or to make a choice about whether to use it or not. This reinforced the need to ensure AI features are a choice, and in the case of the AI searchbot example, an alternative way to achieve the same task is provided.

The C2PA terminology was off-putting to students

As they read the data in the different categories, students felt alienated by some of the C2PA language, particularly terms like ‘content credentials’, ‘provenance’ and ‘assertion’. When they were able to see contextual detail associated with these terms, most were able to fathom what the terms meant, but I concluded that including these terms in the designs and in the labels of the different data categories was giving students an extra unnecessary hurdle of jargon to overcome before they could reach an understanding of the data presented to them.

Some C2PA data was clear and useful, other data felt unnecessary

The four categories of data resonated differently with the students, and, based on their reactions, some of these data sources felt unnecessary to show data provenance.

Most students found the source links useful, and what they expected in the category labelled ‘Digital source data’. They also found the AI prompt useful, but they didn’t expect to find it in ‘Transformation processing data’. Almost all felt the notification that the response was AI-generated (in the ‘About this response’ section) was unnecessary. Other data points that students felt confused and nonplussed by included the ‘CR’ logo, the identity data, the certification data (including the timestamp), and the technical terms describing the response-generation process.

Some of the data students were expecting was missing

As they reviewed the different categories of data they were presented with, students commented on data they would have expected to see. This included: environmental impact data, information about how input data would be used to train AI models, and details of how the data was processed and held by the searchbot feature. While these data points largely fell out of the realm of content credentials, it was interesting to understand the types of data students looked for when making judgements about whether to use and trust this type of AI feature.

The C2PA designs targeted some trust factors but not others

Re-examining factors that affect trust highlighted that trust is heavily influenced by implicit factors or emotional cues which are specific to individuals. Targeting subjective trust factors such as relatability, (which depends on how well something supports a person’s interests and resonates with them) was very difficult to achieve without research to understand the nuances of the target audience. Some students were reassured by the mention of AI models involved in generating the content, but for others, this merely caused confusion.

Targeting explicit trust factors like transparency and explainability in displaying content provenance was more straightforward, but it meant striking a balance. Providing too much technical detail about the production process, although this was intended to build trust by being upfront, could end up causing confusion and therefore result in the opposite effect, as I observed in the student tests.

Conveying reliability remains a challenge when AI is involved. Since AI is probabilistic, its results are variable, and in the context of a feature like an AI searchbot, it was not possible to make assurances in the provenance data. Instead, I used the progressive disclosure levels to encourage exploration and a gradual building of understanding, to help people come to their own conclusion about whether to trust the AI-generated content. I saw this working in the student tests, as they made reliability judgements about the AI responses when they saw the source data revealed.

Provenance indicators should replace the sparkles and it’s time for a unified AI vocabulary

The C2PA standard is several years old, yet content provenance is still a relatively unfamiliar concept. With new AI legislation published and the advent of more formalised requirements to disclose AI usage and application, I predict more rhetoric and discussions around disclosure, authenticity and provenance, to evolve the language and terminology around these concepts and to make them more mainstream.

Those who consume AI-generated content want to know how it was made, and the sources it came from, and those who use AI to generate or shape the content they publish have an obligation to disclose how AI was used, to ensure transparency from both perspectives. When designing AI features, consistent labelling will play an growing part in conveying meaning. As people become more familiar with AI, they will rely on these labels more and more to make judgements and decisions about AI features and AI-influenced content.

This small study has re-emphasised the importance of choosing terms and wording labels carefully, to strike a balance between technical and non-technical and to provide just enough detail to resonate with audiences but not so much that it confuses them. It highlights the need for a unified vocabulary for AI, to effectively communicate about its usage and application. As AI usage progresses, I expect provenance indicators and associated design patterns to continue progressing, and I’m keen to keep a close eye, to keep iterating on ways to display provenance data. I look forward to seeing sparkles replaced by more meaningful signifiers of AI usage and disclosure.

QuillMark, iterated. How user research helped us evolve an AI-enhanced Drupal style guide assistant

AI and UX Innovation Intern Shlok Gupta and I worked together to build and test Quillmark, an AI-powered Drupal content design tool that tackles the persistent problem of keeping content consistent with the University’s style guide. This post covers how we iterated from version 1 to 2.

Leading a UX Service, it’s critical to stay informed about the latest technologies, as these dictate user expectations and perceptions. I recognised the potential for AI to transform publisher experiences of using Drupal when Drupal AI experts Freely Give came to work with our team in January 2025. Since then, I’ve been committed to experimenting with AI application in content management, both through my involvement with the Drupal AI initiative and by supporting a succession of interns in my UX team at the University.

As a technology, AI is becoming increasingly mainstream. That said, its rapid rate of evolution and flux means adopting it consistently remains a challenge, as does designing AI features that are centred around user needs and expectations. AI features can be built and released in a matter of days, but can just as quickly become superseded and obsolete as technologies and capabilities change, and user expectations and perceptions of AI features can vary widely depending on individual experiences, trust factors and preferences. Continuous research is necessary when developing any user-centred software, but it’s particularly important when designing user-centred AI features, which I wrote about in a related blog post:

Stop chasing, keep researching: Why continuous contextual learning is the only way to build useful AI features

In the spirit of continuous research, the students who occupy the AI and UX Innovation Intern role in my UX team spend as much time testing as they do building. Shlok Gupta joined the team from May – August 2026, in which time he prototyped and tested two versions of an AI-enhanced style guide checker tool in Drupal. Shlok described his process to build and test Quillmark v1 in two blog posts, and also produced a video to show it in action:

Building QuillMark: testing a Drupal style-guide assistant with web publishers

Quillmark: the technical details behind my Drupal style-guide assistant

View a video of Quillmark v1 on MediaHopper

In this post, I continue the story, to explain how version 2 was built based on feedback from version 1, and what we learned testing version 2 with publishers.

QuillMark v1 had a non-AI stage and an AI stage and the ability to flag and fix errors

Shlok designed QuillMark v1 to have two stages as a way of making most economical use of AI time and tokens. Stage 1 used basic pattern-matching and code-based rules to catch straightforward style-guide errors without AI, and stage 2 then brought in AI to find and recommend fixes for deeper issues. This meant AI resource wasn’t spent identifying and resolving simpler problems that didn’t need it. The screenshots indicate how v1 worked.

Screenshot showing Quillmark v1 (in the right-hand panel of a Drupal site) ready to review content

Screenshot showing Quillmark v1 (in the right-hand panel of a Drupal site) ready to review content

From right-left, four screenshots showing detail of the QuillMark guide in the right-hand panel before a scan, the panel expanded to show the guide, results of a stage 1 scan and detail of selected stage 1 findings

From left to right, detail of QuillMark in the right-hand panel – before a scan, expanded to show the guide, results of a stage 1 scan and detail of selected stage 1 findings

Three screenshots of QuillMark, showing the panel ahead of running the stage 2 scan, combined results of the stage 1 and stage 2 scans and detail of selected stage 1 and stage 2 findings alongside each other

From left to right: the panel ahead of running the stage 2 scan, combined results of the stage 1 and stage 2 scans and detail of some stage 1 and stage 2 findings alongside each other

QuillMark screenshot showing how a finding from a stage 2 scan mapped to a sentence in the passive voice highlighted in the editorial interface

QuillMark screenshot showing how a finding from a stage 2 scan mapped to a sentence in the passive voice highlighted in the editorial interface

QuillMark screenshot showing the highlighted sentence changed to active voice - which resulted from pressing the ‘Apply fix’ button on the finding

QuillMark screenshot showing the highlighted sentence changed to active voice – which resulted from pressing the ‘Apply fix’ button on the finding

Tests revealed the need to simplify QuillMark v1

In the tests of QuillMark v1, participants were asked to read a guide about how the tool worked, use it to scan some pre-prepared content and then review the resulting findings and say how they would act upon them. The results indicated several areas for improvement.

All participants understood that QuillMark v1 was a tool to help them correct content errors. Despite being directed to read the guide upfront, however, some participants were unclear of the difference between stages 1 and 2.  Having run stage 1, participants were uncertain what stage 2 of the tool would do. Some identified the AI detail in the stage 2 labels and therefore realised that stage 2, unlike stage 1 used AI, but others thought stage 2 was the ‘repair’ or ‘fix’ stage to follow stage 1 as the diagnostic stage.

As they reviewed the different categories of finding in the right-hand panel, most participants said they felt the amount of information was overwhelming, and difficult to scan read. They especially found it difficult to interpret abstracted recommendations to fix errors, preferring to receive suggested fixes through the ‘Apply fix’ option. As a way to reduce the overwhelm, some participants said they would be inclined to go and fix several of the errors manually first, and then rescan, so that the tool only presented them with the errors they missed themselves.

Other observations from the test participants fell into two categories, lack of understanding of how Quillmark worked, and feedback on how they felt it could work better for them.

Participants were confused about how to use QuillMark and how it worked

Several aspects of participants’ behaviours indicated they weren’t sure how QuillMark worked. These were as follows:

  • There was some confusion over how to start QuillMark running. Participants did not immediately realise that the Stage 1 and Stage 2 buttons would start the tool. They looked for a button that said ‘Run’ or similar. When they worked out how to activate the different stages, they were unclear about what was happening in each.
  • Several participants raised questions about the cost and environmental impact of running the AI stage of the tool, saying this would affect whether they would choose to use it.
  • There was a lack of clarity from participants about whether QuillMark was using generic grammatical rules to flag errors and make recommendations or if it was using the specific University of Edinburgh style guide rules
  • Several participants did not initially recognise that QuillMark could suggest fixes as well as flagging errors.

Participants suggested ways to change QuillMark, to help it support publishing practices better

When they had got used to working with QuillMark, several participants suggested ways they felt it could work better. There were as follows:

  • A request to be able to apply bulk fixes to recurring errors, rather than addressing them one-by-one (for example, removing full stops in repeated headings or switching sets of times from the 24 hour clock to the 12 hour clock)
  • An ask from several participants to be able to preview fixes offered by the ‘Apply fix’ buttons before accepting them to change the text in the editorial interface (however, they soon realised they could switch between the ‘Apply fix’ and ‘Revert fix’ buttons to achieve this).

QuillMark v2 worked in the same way but the interface was simpler

Taking on board the feedback from the QuillMark v1 tests, Shlok made a version 2 which included several changes:

  • A single ‘Scan’ button to start both stages of the tool working, replacing the need for publishers to run stage 1 and stage 2 scans in sequence
  • Progress indicators that appeared once ‘Scan’ had been pressed, to communicate what was happening, and to affirm that the editorial style guide was being checked
  • Inclusion of an ‘Advanced settings’ panel giving users the option to decide whether to include deterministic error scanning and fixing and the use of AI to suggest replacements
  • A panel that estimated monetary cost and environmental impact of each scan resulting in the findings

View a short video demonstration of QuillMark v2 on Media Hopper

The screenshots indicate how v2 worked and show the differences between v1 and v2

QuillMark screenshot showing the tool in the right-hand panel ahead of a scan being run on a page of content titled ‘Research projects’

QuillMark screenshot showing the tool in the right-hand panel ahead of a scan being run on a page of content titled ‘Research projects’

Series of QuillMark screenshots, from left to right, showing the tool with the guide displayed, and the successive stages once ‘Scan’ has been pressed – reviewing, preparing, checking editorial style rules and then the notification that QuillMark has completed the scan.

QuillMark screenshots, from left to right, showing the tool with the guide displayed, and the successive stages once ‘Scan’ has been pressed – reviewing, preparing, checking editorial style rules and then the notification that QuillMark has completed the scan (showing that AI suggestions are included in the ‘Advanced settings’ pane

Tests showed v2 was better, but still had room for improvement

The testing protocol for QuillMark v2 remained the same as for v1, however, this time participants were asked to use the tool on three separate pages of University website content, to give them more practice of engaging with the tool to learn how it worked.

Participants found it easier to understand how QuillMark v2 worked

With the new blue ‘Scan’ button, participants were more easily able to get started with running QuillMark. Since it was running an AI and a non-AI stage QuillMark v2 took longer to produce findings than QuillMark v1, but the newly added progress indicators kept participants informed about what was happening.

The indicator with the data about environmental and monetary costs was noted by all participants, and some said this was welcome information to help them make responsible choices about AI use.

Participants were still directed to the guide, which some wanted to refer back to once they had got started with the tool, to remind them of how it worked.

There were fewer questions about the source of the rules powering Quillmark v2 than for v1, and several participants verified that the tool was using rules from the University editorial style guide by pressing the ‘Link to rule’ button on individual findings (which connected to the relevant page or section of the style guide).

Some still found the interface and display overwhelming

Although QuillMark’s interface had been simplified by the removal of stage 1 and 2 indicators, participants still commented it contained a lot of textual information and was difficult to scan. Some were unclear what the different labels in the findings meant (Errors vs Warnings vs Notes), and several found it easy to miss some of the tool’s functionality simply because there was too much to read through.

Participants were still uncertain about the ‘Apply fix’ button

As they worked through the individual findings, participants were able to recognise the need for corrections that had been highlighted, but some did not appreciate that QuillMark v2 offered a fix as well as flagging the error.

As they used QuillMark v2 to scan a page about research projects, the tool flagged an instance of poorly worded link text (‘Project Website’). The ‘Apply fix’ button contained a suggested improvement (‘Visit the Hillforts Atlas website’) but several participants did not select this, instead they manually addressed the error themselves, by checking the link destination and then writing appropriate link text. Those that did notice that ‘Apply fix’ had made a suggested improvement were happy to accept it and apply it.

Screenshot showing a QuillMark v2 finding flagging poorly written link text

Screenshot showing a QuillMark v2 finding flagging poorly written link text

Screenshot showing QuillMark v2 a fix to improve poorly written link text applied in the editorial interface

Screenshot showing QuillMark v2 a fix to improve poorly written link text applied in the editorial interface

They still wanted repeated findings to be consolidated, to be addressed together

Some categories of findings contained recurring instances of the same issue. Participants wanted these to be grouped or summarised instead of being listed individually, as they wanted to address all the findings on a particular issue together in one action (either dismissing them or applying suggested fixes), rather than going through them one-by-one. This problem was particularly prevalent on pages where the content followed a structural pattern, for example a series of headings all written in non-sentence case (with misuse of capital letters) or a labelling convention applied consistently to rows of a table (for example numerals in sequence).

Screenshot showing individual instances of an error flagged as separate findings in the QuillMark v2 interface

Screenshot showing individual instances of an error flagged as separate findings in the QuillMark v2 interface

There was concern over saving progress when working through lots of findings

In the tests, one of the pages scanned by Quillmark v2 generated 166 findings to be addressed. As participants worked through those in a particular category (for example ‘Links and calls to action’) they felt uncomfortable that they were unable to save as they went, seeking some reassurance that their work was being logged sequentially. They looked for a ‘Save’ button associated with each individual finding, and some wrongly assumed the ‘Dismiss’ button saved the change they had made (whereas in actual fact, this button removed the finding from the list to be addressed).

QuillMark was unable to reconcile multiple errors on the same content

In some instances, a piece of content contained multiple style guide errors, so it appeared in separate findings in separate categories. An example was ‘Prof Smith’s website’ used as a piece of link text. This error appeared in the ‘Spelling, wording and abbreviation’ category, flagged for the incorrect use of ‘Prof.’, but also in the ‘Links and calls to action’ category, flagged for unclear link text.

When participants encountered an error in one finding, they corrected it there, but then when the same piece of content was also flagged for another error in another category, Quillmark v2 had already noted that the content had been changed, and had no way of layering on another, different correction. In these instances, a warning flag disabled the ‘Go to field’ preventing participants from going to the incorrect content, therefore forcing them to try and find the location of the incorrect content themselves, which they found disorientating and tricky, especially in long pages of content.

Screenshot showing the blocker message that displayed when there were multiple errors on a single field of content

Screenshot showing the blocker message that displayed when there were multiple errors on a single field of content

Participants wanted QuillMark to fix heading levels and reposition inline links

Errors relating to content structure, such as incorrect heading hierarchies and inline links were flagged by Quillmark, but the tool was unable to suggest an automatic fix to apply. Several participants said they would have liked for an autocorrect functionality for these errors, since they found them difficult and fiddly to manage within the EdWeb2 editorial interface.

The next iteration of the tool will focus on one category of style rules: Links

Since the last round of tests on QuillMark v2, Shlok completed the internship, and was succeeded by Viktor Sharha in the role of AI and UX Innovation Intern, to continue the work on the Drupal style guide checker tool. Taking all the findings from the tests of QuillMark v1 and v2 together, I concluded that the most pressing problem to address was that of overwhelm.

When a page contains many errors or style guide violations, there is a dual design problem: Firstly how to detail each error, the root cause and the suggested fix, so that publishers understood what was wrong and how it can be corrected, and secondly to do this concisely, taking up the minimal space possible, to avoid publishers having a long scroll through multiple errors with the associated cognitive load.

Going forward, it made sense to focus on the first problem of presentation of individual findings first, and to do so at a site level (recognising that publishers may not do their checking on a page-by-page basis), before scaling this up to design for multiple errors. To maintain a tight remit, and to focus on the style guide violations which are highest priority to fix, we have decided to specifically look at links, to consider how the tool can spot instances of poor link text, how it can describe ways to improve these, and the options for applying automated fixes.

Reflections from the Scottish Web Folk Conference 2026 – ‘Human vs machine: who does it better?’

Scottish Webfolk has become a staple fixture in my October work calendar for the last three years. Despite the very early start to make it north for the conference starting just after nine, by the time you arrive you realise it was worth making the trip. This year there were over 100 attendees from 26 different universities, with a strong line up of presenters from Edinburgh.

Andrew Millar, Head of Digital Engagement & Marketing at University of Dundee started with his key note opening talk, which is always a highlight and this year was no different as he introduced the theme of the conference ‘Human vs machine: who does it better?’

Andrew’s talks are always grounded in realism, especially given the challenging times the higher education sector is facing, but with ample optimism and insightful narrative woven throughout so that by the end you feel energised for a day or learning, sharing ideas and discussing common challenges.

A different take on a familiar theme: AI

It’s probably not a great surprise that the conference had an AI theme. However, I felt this time it was as much about the importance and value of ‘keeping a human in the loop’ as it was about leveraging the benefits of AI-based technologies and tools.

In his talk Andrew referred to the quote: “the threat was never the machine, it’s how people use it” which really stuck with me.

AI is the latest in a whole historical timeline of technological advancements, as Andrew explained. Whilst it is revolutionary, it still needs humans to ensure that it’s used for the greatest benefit. So in that sense the day highlighted to me that machines and humans need to work in tandem, rather than in competition to get the best results.

My third presentation at the conference – ‘AI search summaries: a new lens on old content problems’

For this year’s conference I co-presented with Nick Daniels and Mel Batcharj for what was to be my last UX Service talk. We shared how we trained our web publishing community at Edinburgh on how to make their content AI-ready in the ever-evolving AI search landscape.

We highlighted how evergreen content design principles still stand strong in the new era of AI search and shared how we used the prominence of AI to teach these principles in a slightly reframed way. The aim being to enable publishers to optimise their content for AI search, whilst also tackling persistent problems with our web estate content.

The key takeaway was that website content that is well structured, well maintained and accessible is just as important for AI as it is for users.

You can read more about this training in my previous blog.

June Content Improvement Club: Is your content AI ready? 

Web governance: a problem shared is a problem halved

Another highlight of the day for me was the refreshing session which Stratos Filalithis from Edinburgh and Chris Purdy, Head of Digital Experience from the University of Glasgow hosted, which they called ‘Digital Unscripted’. There were no slides, instead an honest conversation about the challenges both universities are facing to tackle their huge, unwieldy web estates.

I’m used to starting off presentations at conferences explaining that Edinburgh has a huge web estate with over 1,000 web publishers (today was no different). This is often met by surprised faces, especially for those universities with a centralised digital marketing team who oversee the production of all web content. So, it was somewhat comforting to hear that Glasgow faced similar challenges to Edinburgh in terms of managing vast amounts of diverse web content and that both institutions were sharing knowledge and expertise to try to bring order to the chaos.

I found this interesting as a lot of my work recently has been focussed on developing guidance (alongside subject matter experts) for website owners on their key responsibilities when creating and managing a website.

The difficulties and governance risks associated with large web estates felt like a very relevant topic to be discussing given AI’s ability to unearth long forgotten about content in its answers, but also in the wake of the PHP vulnerability incident which highlighted the importance of ‘knowing what you own’ from a web registry perspective.

Scottish Web Folk – much more a community event than a traditional conference

I think the ‘Digital Unscripted’ session, along with the panel discussions where representatives from different universities shared their perspectives on challenges they were facing for me sums up that really Scottish Web Folk is more a community event than a traditional conference. There’s a real sense of sharing solutions to mutual problems and discussing the challenges that nobody has yet managed to resolve, rather than just highlighting successes. I think it’s the community emphasis and collegiality that keeps bringing people back year after year.

Using a survey to understand digital design tool usage across the University (as part of the Figma design component library project)

This year, as part of the Figma component library project, we researched how colleagues across the University use digital design tools in their work. This post explores how we approached the research, from designing and distributing the survey through to analysing the responses and identifying emerging themes. 

In a previous blog post, intern Kaile Liu introduced the wider Figma project, and its aims and objectives: 

What I am working on as a Digital Developer Intern (Figma) 

Request access to the project brief for more details:

Figma project brief

We needed to understand how digital design work happens across the University 

To begin with, we needed to understand how staff across the University design, collaborate and deliver digital work. 

There are a wide range of tools, workflows and practices currently in use. Understanding these existing ways of working would help us explore where Figma fits within the wider digital landscape, rather than looking at Figma in isolation. It also supports the work we are doing to improve our Figma component design library.

To hear from colleagues across a wide range of roles and disciplines, we chose a survey as the main method for this research.

Five questions framed our research 

As well as understanding what tools people use, we wanted to find out how those tools support their work and where there may be opportunities to improve existing ways of working.  

Our research was underpinned by five key questions: 

  • What tools are people using day-to-day? 
  • How do those tools support their work? 
  • Where do people experience challenges or inefficiencies? 
  • How do teams collaborate when creating digital content and services? 
  • Where are there opportunities to provide better support, guidance and training? 

These questions helped us build a broader picture of digital working practices across the University. Additionally, working this way allowed us to look beyond any individual or singular tool.

Hearing from more than just traditional design roles was important 

Design work happens across many different roles at the University, including roles not traditionally associated with design or that do not have ‘design’ in the job title. For example, a communications officer creating digital content, a developer sketching a wireframe, a UX practitioner co-creating a prototype or a learning technologist building online resources may all be making design decisions as part of their work. 

This meant looking beyond traditional design roles and hearing from colleagues who use digital design tools in different ways.  

Initial conversations helped us shape the research 

Before designing the survey, we carried out an initial set of user interviews with staff from across the University. We also drew on existing University communities and communities of practice to understand the different roles involved in digital design work. These conversations gave us an early view of how colleagues use and collaborate through digital tools. They helped us identify the groups we wanted to hear from and shape the questions we wanted to ask in the survey.

Adapting our survey to broaden our understanding of ways of working

Based on these interviews, we recognised that a single set of questions would not capture the differences between roles particularly well. Instead, we adapted the survey so that respondents would answer shared questions and then be filtered to questions relevant to their role and ways of working.

This, in essence, would give us a more consistent set of responses while still allowing us to explore some of the differences between roles.

Role categories allowed us to tailor the survey 

We created a filter with seven options: 

  • Marketing and Communications 
  • UX and Human-centred Design 
  • Graphic and UI Design 
  • Developer 
  • Project or product manager 
  • Research 
  • Other (for participants who did not feel represented by the other groups) 

Participants selected the options that best reflected their work, determining which role-specific questions they were shown. They also provided their job title.

The survey combined common and role-specific questions 

The survey consisted of three main sections. The questions included a mixture of multiple-choice and open-ended responses. This allowed us to identify broader patterns while also giving colleagues opportunities to explain their experiences in more detail. 

The opening section included five common questions about participants’ roles, the types of digital design work they do and the tools they use, including tools for collaboration, diagramming and presentations. 

The middle section included around four to six role-specific questions, depending on the participant’s selected group. These explored aspects of digital working relevant to each group. For example, marketing and communications colleagues were asked about the tools they use when designing marketing materials, while developers were asked about the tools and approaches they use during the design phase of projects. 

The closing section included 15 common questions focused on colleagues’ experiences and expectations of digital design tools, including Figma. We asked which tools colleagues would be likely to use, whether they experienced challenges with their current tools and how their work might be affected if they lost access to Figma. 

We wanted the survey to reach as many colleagues as possible

We distributed the survey primarily through existing Teams channels and University communities, which helped us reach a wide range of roles. These included groups focused on marketing and communications, human-centred design and digital development, as well as distribution via heads of IT. 

The survey received 55 responses  

We heard from 55 staff across a range of roles. Based on the categories participants selected when completing the survey, these were as follows: 

  • Marketing and Communications: 20 
  • Developers: 10 
  • UX and Human-Centred Design: 6 
  • Graphic and UI Design: 3 
  • Project or product manager: 2 
  • Other: 14 

Although seven categories were available in the survey, no participants selected ‘Research’ as their primary category, so it is not shown in the breakdown. 

The ‘Other’ category included colleagues working in areas such as product and project management, learning technology and academic support. 

We refined the groups when analysing the responses 

The survey categories gave us a useful starting point, but participants did not always recognise themselves within the roles specified (even when their job title directly reflected one of the options). For example, a participant selected themselves as ‘Other’, but their job title indicated that their work was within graphic design.

In order to evaluate the responses effectively, we revisited participants’ selected categories alongside their job titles and descriptions of their work.

This helped us six create groups that gave us a more useful basis for comparing responses:

  • Developers and technical specialists 
  • UX practitioners 
  • UI and graphic designers 
  • Marketing and communications staff 
  • Product and project managers 
  • Learning technology and academic support staff 

Some of these groups needed to be created on the back of this, as we had not identified those specific roles. For example, Learning Technologist was not an option for participants to select, but emerged as a useful category based on the number of respondents who had this as their job title. 

After this assessment, we were in a better place to begin analysis, as these groups gave us a more useful basis for comparing responses while still reflecting the range of roles represented.

Analysing the survey results required two approaches 

The survey produced both quantitative and qualitative responses, and we used different approaches depending on the type of question and response. 

For the role-specific questions, we used FigJam (Figma’s whiteboard for online collaboration) to review and compare responses across the different groups. This included both quantitative and qualitative responses, allowing us to look at the detail within each role as well as compare responses across the wider dataset. We used Power BI, a data visualisation platform, to explore the common quantitative questions, which gave us a consistent set of responses across all participants. 

We used FigJam to explore the role-specific responses 

We created a large collaborative FigJam board to organise the responses. FigJam supported discussion within the project team, allowing us to work in tandem and analyse the data as a group (in person and asynchronously).

We read through responses from each group, compared them across different roles, and identified recurring patterns and themes. FigJam also allowed us to consider the quantitative responses alongside the more detailed qualitative answers, giving us a fuller picture of ways of working. 

Screenshot of three frames on FigJam full of stickies. There is a blue section, titled Developers, an orange section titled Marketing and Communications, and a green section titled Graphic visual and UI design.

Section of our analysis on the FigJam board

Power BI helped us explore the common quantitative responses 

Power BI allowed us to visualise responses across the opening and closing questions. We wanted to do this as they were common questions asked of all participants. Power BI allowed us to bring our data to life by presenting the data in a more visual way, highlighting the common patterns and themes more easily.

Screenshot of Power BI visualisations. There are seven individual bar graphs of different colours, alongside a larger bar graph with these seven combined. The large bar graph has a right skewed narrow tail. There are also two sets of pie charts, each consistening of seven smaller charts and one larger combined. The data itself is not visible.

Section of our Power BI visualisations for the project.

Grouping responses helped us identify recurring themes 

The open-ended questions in the role specific sections had more detailed responses. This required us to compare responses across the different role groups and group similar comments together to identify recurring themes. 

As we worked through the responses, we also noted areas where a follow-up conversation with a participant could provide more context.

We also captured key quotations that illustrated what participants were telling us in their own words. These will be useful for when we playback findings to colleagues and complete a written report. 

The research will inform the next stage of the project 

The research highlighted both the diversity of digital working practices across the University and common themes across different roles. 

The next stage will bring these findings together into a report. It will highlight where Figma currently supports existing workflows and where its integration will provide more consistent, unified delivery of digital practices across the University.

Testing with users informed the Institute for Academic Development’s content migration

The User Experience Service worked with colleagues from the Institute for Academic Development to test their content with users. Results from the testing supported content decisions in migration work that these colleagues carried out later in the summer.

This is part two of a two-part series. Part one is available here:

How we helped the Institute for Academic Development to establish user needs for their website

We had established some user needs but wanted to learn more

In two online sessions, we had worked with Institute for Academic Development (IAD) staff to establish a shared understanding of what users needed from their service.

We now wanted to find out about the experience of students who use the IAD website. Specifically, we were interested in two audience groups:

  • Postgraduate research (PGR) students
  • Students from all stages who might need academic skills support

Along with tests on the existing site, we also wanted to try out some new layout ideas and learn about what types of content these audiences would find most useful.

An in-person event gave us an opportunity to hear from PGR students

We knew that presenting information about workshops would be a priority for the new sites and we wanted to know the best way to present this. To inform the design, we needed to speak to people who might use the website to look for these workshops.

Fortunately, the perfect opportunity was just coming up. On 17 June, a large group of research students would be attending PhD Horizons, an all-day careers event at McEwan Hall. We planned two simple research tasks that we could carry out on the day. Emma Horrell and I attended the event with UX team interns Hannah Watson and Shlok Gupta. This was a great opportunity for Hannah and Shlok to gain hands-on experience of in-person UX research.

Usability testing helped us to learn about the existing site

The first task was a quickfire usability test asking PhD students to use the IAD website. For the testing, we handed PhD students an iPad and presented them with two scenarios:

  • Imagine your viva is coming up. How would you find a workshop to help you prepare for this?
  • You’re going to a conference and want to get better at presenting. What can you find on this site that could help you to do this?

Paper prototyping gave us insight into alternative designs

For the second task, we presented four paper prototypes of potential homepages for postgraduate researchers. PhD students compared the prototypes and talked about which areas of the site they would find most useful. This gave us an indication of how closely these homepages aligned with their needs, and it also provided insight into how the language used in the prototypes might be understood.

Clipboards and sticky dots on a table.

We used paper prototypes of homepages to test ideas with students.

A piece of A4 paper with a rough mockup of a webpage. Page is titled Support for postgraduate researchers. There are 4 boxes with titles Explore our workshops and courses, Plan your PhD journey, Browse our writing support courses, and Not sure what training you need? Launch the Development Needs Analysis tool. There are sticky dots on the page.

Students used sticky dots to indicate which areas of the paper prototype would be most useful to them.

What we learned from the usability testing and paper prototyping

In the end, we heard from about 20 PhD students who told us how they might search for workshops on the IAD website. From these conversations, we learned how these students responded to items presented on the live site and what they thought of some potential alternative ideas.

In the usability testing activity, we learned which ways of presenting workshops were perceived as being more useful than others. For example, most participants did not use the embedded PDF brochure, despite it being prominently located on the PGR landing page. Instead, we saw users gravitate to webpages showing workshops grouped by theme, or as an A to Z list.

Screenshot of the Postgraduate Researchers page on the Institute for Academic Development website. An embedded brochure is at the top of the page.

Users did not interact with the embedded PDF brochure.

Observations like this gave us a steer on which categories and methods of presenting workshops would be most useful to PGR students in future designs of this content.

In the homepage paper prototype activity, one of our mock-ups included an option to complete a Development Needs Analysis. This is a form that new PGR students can complete to assess their learning needs. But when we presented the text ‘Development Needs Analysis’ in our prototype homepages, most students could not guess what this was. That seemed like something that could be addressed in the new website.

We shared the findings with colleagues tasked with developing the new site for postgraduate researchers.

We ran usability tests on the academic skills content

Alongside this work, we were helping IAD staff to develop content for a new academic skills website. We agreed that a short round of usability testing on the existing IAD site would help us to get a better understanding of what it’s like to use it. We wrote a script for a 15-minute test and recruited four students working on summer internships at the University. A week later, we were on a Teams call with those students, watching them use the site.

The tasks we set in the usability tests were as follows:

  • You’re an undergraduate student at the University of Edinburgh. You’ve just started a new course. You’ve got your reading list, and you’re feeling overwhelmed by it. Your student advisor has said to book a one-to-one meeting with the Institute for Academic Development (IAD). You land on the IAD website homepage. Can you find out how to book a one-to-one?
  • You’re a postgraduate taught student. You’ve seen a poster advertising dissertation workshops. You arrive at the IAD homepage. Can you find a workshop that will help you with writing your dissertation?
  • You’ve just completed your first year of your History degree. Your exams didn’t go so well and you have a resit coming up in August. Your friend said there’s some advice on preparing for exams on the IAD site. Starting from the homepage, can you show me what advice you can find?

What we learned from testing the academic skills content

None of the participants clicked on the Study Hub resources. The Study Hub contains valuable content that could help students with the problems we presented in the test. But none of the participants followed links to the Study Hub, despite these links being present on several of the pages that participants used.

The Exam Bootcamp page did not lead participants to other exam content. When looking for support for exams, participants scanned the homepage for any mention of exams. This led them a page about Exam Bootcamp, a resource available on the Learn VLE. Participants seemed pleased to find this but didn’t find any of the other available resources or workshops about exams.

Participants did not seem phased by the amount of dissertation workshops on offer. When tasked with finding a workshop to help with dissertations, participants reached a page that listed 17 different workshops on the topic of dissertations. We had anticipated that that might be a problem. But none of the participants seemed worried about how they would choose a workshop to attend. They scanned through the list and picked those that looked like they might be relevant.

The new sites are now live

The teams working on the new sites took the findings from the strategy sessions and the user research and fed this into the development of four new sites. The new sites went live in July, using layout, wording and information architecture choices that reflect some of the insights gained from the content review work.

For example, the Postgraduate Researcher and Career Development site puts PGR workshops in a prominent location on the homepage. Users can browse all workshops, or view them grouped by theme. Next to that, a card shows development support options for PGR students at different stages of their studies (‘Starting out’, ‘Keeping on track’ and ‘Final stage’), echoing the variation between needs at different stages that we identified in the content review.

Postgraduate Researcher and Career Development

Screenshot of the Postgraduate Researcher and Career Development homepage. There are cards on the homepage reading Explore our PGR workshops, Find development for every stage, and Not sure what training you need?

In a similar way, the new Academic Skills homepage leads with the priority content we identified in the review. Study Hub is given the most prominent location at the top left of the screen on desktop, and at the top of the screen on mobile. Cards for workshops and one-to-one meetings show the support services that are available to students. Elsewhere on the page, students can find support for topics such as writing, making notes, literature reviews and exam preparation.

Academic Skills

Screenshot of the Academic Skills homepage. Cards on the homepage are titled Visit Study Hub, Browse our workshops, and Book a one-to-one.

Reflections on this work

This work was a great example of how you can take an evidence-informed approach to content strategy when you’re working to a tight deadline. It also showed how you can use a website migration as an opportunity to review the content you have and think about how it supports your service. By drawing on existing knowledge about users, and testing ideas and prototypes with them, the teams were able to make decisions about their new websites with user needs in mind.

If you’re reviewing your content

If you’re thinking about reworking what you have on your website, get in touch. The User Experience Service can help you to evaluate what you have, develop your content strategy and bring user research into the process.

Contact the User Experience Service

How we helped the Institute for Academic Development to establish user needs for their website

This summer, the User Experience Service worked with colleagues from the Institute for Academic Development to review their content ahead of a website restructure. In this post, we look at how we developed a shared understanding of user needs to help guide this work.

Supporting the migration of IAD content

At the end of the academic year 2025 to 2026, the services performed by the Institute for Academic Development (IAD) moved to different areas of the University. As part of this change, the IAD website was being shut down.

Content from the IAD website was moving to four new sites:

In June 2026, there were 900 published pages on the IAD website. Time was tight, with the content needing to have moved to its new location by August.

Members of the User Experience Service (Emma Horrell, myself, and interns Hannah Watson and Shlok Gupta) worked with colleagues from the IAD to prepare for this move. In this post, I’ll outline how we worked together to establish who uses the content and what user needs it supports.

We ran two online sessions to identify service users and their tasks

In a content review, we typically start by working out who’s using the site and what they’re trying to do when they arrive at it. It’s also useful to know what kind of events trigger people to visit the site and what information is likely to be most valuable to them at that point.

To answer these questions, Emma Horrell and I ran two online sessions with colleagues who were familiar with the existing content and who would be involved in creating the new sites. We wanted to draw together existing knowledge about users of the IAD’s services so we could establish what these users need from the IAD website. Each online session brought together a small group of specialist staff. We met over Teams and used Miro to gather the information.

Session 1: Postgraduate research students

In the first session, we focused on content for postgraduate research (PGR) students. Through discussion, it became clear that this audience group has shifting needs as they complete their programme of study.

We grouped the student needs on Miro and three stages became evident:

  1. They become a PGR student
  2. They get started with their PGR programme
  3. Later in the programme
Screenshot of Miro showing three circles with sticky notes in each one. Text on the sticky notes is too small to read.

We collected user needs and grouped them into three stages.

New PGR students have distinct needs

We learned that new PGR students might hear about the University’s academic development services in Welcome Week or in a conversation with their supervisor.

At this early stage in their programme, their tasks might include:

  • find workshops that are relevant to me
  • sign up for newsletters that tell me about future workshops
  • learn what I can expect from my upcoming programme of study
  • connect with other PGR students at the University of Edinburgh

PGR student needs change as they progress in their studies

Later in their PhD, these same students will likely have different needs. For example, they might be looking for a particular workshop that will help them to:

  • write a dissertation
  • put together a literature review
  • prepare for a viva

They might also be thinking about what comes after their PhD and what they need to do when they take the next steps in their career.

Some tasks apply at all stages of a PGR student journey

At all stages, PGR students could have these needs:

  • find out which services support academic skills development
  • find out what these services do
  • engage with these services in the event of questions
  • find out what’s on (workshops, support sessions)
  • get updates about upcoming workshops (for example through a newsletter)
  • access policy and standards information

At times, PGR students may also need to know about other University services that can support them. For example, Edinburgh Research Office and Edinburgh Innovations can provide support with funding applications. The Student Counselling Service can support with health and wellbeing issues.

Presenting available workshops was the priority

Our discussions made it clear that presenting information about workshops was going to be the top priority on the new site for PGR students. On the IAD website, workshops were presented in a variety of ways:

  • in an A to Z list
  • listed by theme
  • on a page called “At a glance – workshops this month”
  • in an embedded PDF brochure

Session 2: Users of academic skills content

Alongside this work, other IAD staff were preparing content for the new academic skills site. This would have a different purpose to the site for PGR students, and it would be pitched at a wider audience of students at all stages of study.

An online session helped us to understand user needs

We ran a second session with a small group of IAD colleagues to learn about the existing academic skills content. Specifically, we wanted to know:

  • who uses academic skills content
  • how they arrive at it
  • what they need from it

As in the first session on postgraduate research students, we gathered together known user needs and looked for common themes.

Screenshot of sticky notes on a Miro board. Groups of sticky notes are titled Writing, Workshops, AI, How to get in touch, Exams, Wellbeing, Finding literature and Planning and organising time.

We used Miro to collect the needs of students looking for academic skills support.

Through these discussions, we developed some scenarios to illustrate how a student might arrive at the site. For example:

An undergraduate student gets a poor grade for an assignment. They speak to their student adviser about how to improve their essay-writing skills. The adviser directs them to the IAD. The student goes on the IAD website to look for resources or workshops about essay-writing.

By thinking about user tasks in this way, we developed our understanding of what it might be like to be a student scanning the homepage, looking for content that will help them to complete the task they have in mind.

We also established that the priority for the new academic skills website would be content that supports the following:

  • Booking on a one-to-one study skills consultation session.
  • Finding a study skills workshop.
  • Finding online resources to support study skills such as writing and presenting.

Understanding student users and their priority content like this gave us a foundation for what we needed to do to assess the site and consider alternative designs.

The sessions uncovered a need for testing

Through these two sessions, we collected existing understanding of who was using the IAD website and what they needed from it.

This led us to the next stage of this work. We wanted to know what it was like for these users when they interacted with the site. Could they find what they were looking for? What might be the best way of organising this content on the new sites?

To answer these questions, we carried out two sets of user research:

  1. To support the work developing content for PGR students, we joined a Careers Service event at McEwan Hall to hear directly from those students, seeing how they responded to existing pages and some prototyped alternatives.
  2. To support the work developing academic skills content, we ran a short series of usability tests to observe students using the existing IAD website.

Read more about this work

This is part one of a two-part series. In the next post, I’ll explain how we used paper prototypes and usability testing to learn how users interacted with various site layouts.

Testing with users informed the Institute for Academic Development’s content migration

Architecting a new website to support dementia research at the University

Alzheimer’s and dementia research is a flagship strength of the University.  Websites of different research units play a key role in supporting this work. The UX Service worked with the Alzheimer Scotland Dementia Research Centre to design and launch Dementia.ed, a new home for dementia research at the University.

The University of Edinburgh is internationally recognised for research on dementia and Alzheimer’s, with different research activities occurring across groups, labs, centres in and across the various Colleges. The team at the Alzheimer Scotland Dementia Research Centre had identified a need to join up different research activities, to ensure connections could be made across different research initiatives by facilitating an interdisciplinary University network.

A website was the proposed way to meet this need, with the intention that this site would provide visibility of the different research areas, to enable academic researchers to exchange knowledge and expertise within the wider dementia and ageing research community. The site was also anticipated to have a secondary, non-academic audience: people with a general interest in learning about and supporting dementia research.

With the go-ahead to build an EdWeb2 site confirmed, Angel Garmpi from the ASDRC contacted the UX Service for help to shape the new site.

Before creating a site, we thought through a content strategy

In the seminal textbook on content strategy, ‘Content strategy for the web’ by Kristina Halvorsen and Melissa Rach, the authors introduce content strategy as a way to avoid being tormented by a burden of excess content to be managed. Starting a site from scratch is a golden opportunity to make decisions upfront to prevent content spiralling out of control, leading to website bloat and content debt further down the line.

The framework in the book outlines four elements of content strategy:

  1. Substance – what content does the site need and why?
  2. Workflow – how does the content get created and maintained?
  3. Governance – who owns the site and is responsible for its content?
  4. Structure – how should the content be organised so it can be found?

These elements served as a useful way to tease out key considerations from the start, setting a path for good decision-making and planning for the longer-term, past the website launch.

Substance: Content for dementia researchers and for people interested in dementia research

By considering questions about what the site was for, and who would visit the site and why, we were able to establish the site substance.

The ASDRC team had identified five main purposes for the site:

  1. To facilitate a network where researchers could come to collaborate with others on research work
  2. To showcase and bring together the different groups, centres and parties involved with and supporting dementia research
  3. To be a source of information where people interested in dementia research could come to read about latest insights
  4. To serve as a volunteer hub to capture details of people wishing to participate in dementia research (or make tissue donations)
  5. To act as a payment portal where supporters of dementia research could make financial donations.

Using these purposes as a starting point, we defined primary and secondary site audiences and their associated tasks to flesh out detail of the site substance.

Site audience 1 (primary): Academic researchers working on dementia and ageing

Needs and tasks:

  • Meeting other researchers and groups working in related fields
  • Establishing peers to collaborate with and exchange knowledge
  • Getting support and help with grant and funding applications

Content to support:

  • Contact details of the different groups
  • An entry point to join the network
  • Ways/touchpoints to stay informed such as events

Site audience 2 (secondary): People interested in tracking dementia research developments

Needs and tasks:

  • Finding out about dementia research in relation to a specific condition of interest
  • Staying up-to-date with advances in specific areas of dementia and/or ageing research
  • Identifying relevant academic papers and publications to form the basis of journalistic content

Content to support:

  • Publications and updates about dementia research
  • News articles about topical research areas

Site audience 3 (secondary): People wishing to take part in dementia research

Needs and tasks:

  • Registering to donate brain tissue
  • Volunteering to participate in academic research studies

Content to support:

  • A place to register interest in becoming a research participant

Site audience 4 (secondary): People wishing to make monetary donations

Needs and tasks:

  • Donating money to contribute to the progress of dementia or ageing research

Content to support:

  • A portal where financial donations can be made

Workflow: content is curated and published by the ASDRC

Considering the site substance led us to think about where this content would come from – the sources and the subject matter experts behind it. All too often, websites are built and filled with content for a launch, but without mapping out the content sources and understanding the underlying creation processes, launch-day content is at risk of quickly going stale because there is no plan in place to maintain it.

To fulfil its primary purpose of facilitating dementia research networks, the bulk of the content for the Dementia.ed site was planned to come from the different University-wide dementia research groups and initiatives. This content would be captured as links to the different sources, to ensure it was current. Events from the different groups needed to be highlighted as network-forming opportunities, so these were planned to be pulled together in a single location on the Dementia.ed site by the ASDRC team. It was also felt that the site should provide a round-up of news from different areas, which again would be curated by the ASDRC team.  Finally, the site would include static content such as: practical information about the ASDRC and the dementia research network, contact details, information about how to join the network and details for how to contribute, either donations or by participating in a study.

Taken together, this meant the initial content-related work requirement from the ASDRC team would be: to create the static content for the site in the different sections and to establish the correct links for the different research groups. Going forward, there would be a maintenance required to check the static content and the links remained correct, and to produce events and news content, which all  seemed doable given resource in the ASDRC team.

Governance: ASDRC team owns the site and manages the content

Considerations of how the new site would be governed were related to the workflow processes. A prerequisite for obtaining an EdWeb2 site is the need for a named site owner to take overall responsibility for the site, and the same ASDRC team member responsible for creating, curating and maintaining the site content had been assigned the role of site owner. This kept things straightforward in terms of decision-making about the site’s content and overall direction, although there was a need for the site owner to forward-plan a content calendar for maintenance, curation and creation tasks, and to assign responsibility to other ASDRC team members as a back-up to avoid a single point of dependency.

Structure: content arranged under common user tasks

Starting the design of a new website from a blank canvas can be difficult as there are lots of questions to consider at once. What should the menu contain? What should be on the top of the homepage? How should the content link together?

Another core textbook useful in site building is ‘Information Architecture’ by Peter Morville and Jose Arango. In this book, the authors encourage thinking about how to group or categorise content (in an organisation scheme) as well as how to establish relationships between those groups or categories (in the organisation structure).

Typically, websites adopt ambiguous organisation schemes – with content arranged by topic, subject, audiences and/or tasks to encourage browsing and discovery. This approach was appropriate for the Dementia.ed site to enable it to support the different audience groups and their associated needs and areas of interest. We considered a task

For the organisation structure – a non-hierarchical flat arrangement made most sense. Given the site was anticipated to have a relatively small number of pages, and would have clear calls-to-action for the different audiences, it didn’t seem necessary to plan a heavily hierarchical structure. For the content in the different research group sites, a hub-and-spoke model could be adopted, with Dementia.ed acting as the central hub, linking off to the different sources.

Before building, we needed to sense check the proposed site map with users

With ideas for the site mapped out, it was tempting to just get building in EdWeb2. Before creating the new site, it was wise to test assumptions with some testing.

Usability testing – where audiences are given a task to perform online and are observed as they do it – always provides an excellent source of insight into how audiences naturally interact with a digital artefact like a website, how they make sense of the menu labels and use on-page content elements to find their way to complete their tasks.

When it comes to new sites, however, it can be difficult to know what to test with users. If you mock-up a prototype, which parts should you include and which can you omit? Asking participants to make sense of menu labels that lead to empty pages can feel unnatural, and setting participants tasks can fall flat if there’s no real-life site content where they can find the information they need to complete tasks authentically. All that said, it’s impractical to build out an entire site just for testing purposes, as it’s likely the testing will reveal ways it needs to change which can end up adding more work.

We used another research network site as a proxy to test with

Thinking through the options with the ASDRC team, we came up with a plan. The ASDRC team had started looking at other research network sites for inspiration and ideas about how to build Dementia.ed, so we decided to adopt one of these sites for the basis of a test.

We chose the Edinburgh Mental Health Network (EMHN) as a testing site. This site included top level menu items: People, Research, News, Events, Blog and Contact. It had a total of 13 main pages with additional pages within the News, Events and Blog sections.

Access the Edinburgh Mental Health Network website

Since the site supported researchers in mental health as well as audiences with a more general interest in the topic, we were able to shape a usability testing protocol to use the site to assess how two audience types sought to complete representative tasks:

Audience 1: Academic researcher wanting to join the research network

Representative tasks:

  1. Find out what the network is, and its goals and values
  2. Find out how to become part of the network and what this would entail
  3. Find out about upcoming activities
  4. Find research opportunities in the network
  5. Find out how to develop your research project and involve people with lived experience
  6. Find content you can use in support of a funding application

Audience 2: Non-academic with a personal interest in the research area

Representative tasks:

  1. Find out what research is being done across the network
  2. Find out how the network is managed
  3. Find out how to keep up with what’s happening in the network
  4. Find out how you can contribute as a research participant for studies
  5. Find out how to donate money to support the research

Some of the tasks mapped clearly to specific sections of the EMHN site, whereas others were more ambiguous – they could have been completed in multiple sections. In all cases, it was useful to see how participants approached the tasks, so we could use their expectations to design Dementia.ed in the most audience-focused way possible.

We learned which menu labels worked, and which didn’t

The results of the testing gave us a clear steer on which elements of the Edinburgh Mental Health Network site supported the different user groups completing their tasks and therefore would be a good fit to adopt in the new Dementia.ed site. The tests also showed areas where the approach of the EMHN site was lacking – and therefore where different structuring and/or labelling was more appropriate.

The Events and News sections worked well

To complete tasks about finding out about activities and keeping up with what was happening, participants naturally gravitated towards the Events and News sections of the EMHN site, where their expectations of finding events to sign up to and articles to read were met.

The EMHN site was lacking a specific place for information about the network

When working on tasks geared at finding out what the network was, what its goals and values were, and detail about how the network was run, participants looked for an ‘About’ section when they used the EHMN site. When they realised this didn’t exist, they tried multiple other locations including the homepage and the Research section. This suggested an ‘About us’ section would be useful to include in the Dementia.ed site.

The People section didn’t resonate as the way to become part of the network

The EMHN site presented information for academic researchers to join the network in a ‘People’ section, which also contained information about the management of the network as well as researcher support resources. In the tests, participants struggled to find this content when tasked with finding out how to join the network, suggesting that a more explicit ‘How to join’ section would be more helpful to support this key task.

The Research section was too broad to usefully support specific tasks

Test participants gravitated towards the ‘Research’ section of the EHMN site for various tasks: to find research opportunities, to find help with research project development, to find content to support funding applications, to find out how to be a research participant and to find out how to financially support research activities.

On arrival in this section, when they realised it only contained content detailing different research activities, they went elsewhere. This suggested the need to separate out content about research activities from content supporting specific research-related tasks and to label it accordingly. In other words, instead of having a large, general ‘Research’ section, opt for smaller, more specific sections such as: ‘Research areas’, ‘Participate in research’ and ‘Donate’.

The new site has a flat structure, with content arranged by topic and tasks

Keeping site of the content strategy work, and using the results of testing the proxy site to make key decisions about structure, the new Dementia.ed site was built and launched.

The site contains six main sections, adopting a mix of topics and tasks to signpost the different sorts of content. The ‘About’ section contains background information about the dementia research network, with a dedicated page bringing together links to the different research centres and groups in a hub-and-spoke approach. A ‘What’s On’ section combines pages to Events and News content. The other four sections are labelled in task-focused ways, to ensure relevant audiences can quickly get to the content they need to complete their tasks: Join the network, Get involved, Contact us and Donate.

To ensure audiences reach content in the most direct way possible, the content in each section is either available as a top-level page, or a sub-page, avoiding long pages and deep hierarchies which may confuse those visiting the site.

Access the Dementia.ed website

If you need help architecting your website, get in touch

From start to finish, the process to design and launch the Dementia.ed website took a few months, including a few meetings between the UX Service and the ASDRC team, some short usability tests and an overall site review.

The ASDRC team provided the vision of what they wanted to achieve, and subject matter knowledge of their audience base and their associated tasks. With this information, the UX team could apply expertise to translate goals into a user-centred website, which is manageable for the ASDRC team to maintain.

We’re happy to offer help, advice and support to other University teams. If we can help, please contact us through the UX Service website.

Contact the UX Service

UX for real-world higher education – UCISA UX Community Day recordings now available

If you design, build or manage digital experiences in higher education, the content from this year’s UCISA UX Community Day has plenty for you. The day included a panel discussion, case studies and practical advice from UX professionals, all focused on user-centred approaches in the university context.

As co-chair of the UCISA UX Group, a highlight of my year is planning and organising our annual UX Community Day, which sits alongside our regular webinars. Now in it’s third year, this all-day event is our flagship. It’s aimed at anyone striving to make digital products, services and platforms better for those who use them.

Read more about the UCISA UX Group on our UCISA webpage

Read more about previous UCISA UX Community Days in previous blog posts:

Growing our UX maturity by leading a Community Day for UK higher education institutions

Sharing and gaining knowledge at UCISA’s UX Community Day 2025

Eighty UX HE professionals sharing, collaborating and ideating: The magic of UCISA UX24

Clearly, I’m biased, but here are my top five reasons why this event is worth your time:

  1. It provides a safe-space, non-performative opportunity for higher education professionals to speak and share – to present ideas and work and gain feedback and input – no conference pressure, no vendors or suppliers, just a supportive community ready to listen.
  2. Depending on the UX maturity of your organisation, doing UX in higher education can be lonely at times, as UX practitioners can be outnumbered by other roles. Bringing user-centred folks together to talk openly about their UX work not only fosters welcome discussions, critique and questions, it also provides reassurance, boosts confidence and sparks new ideas to take forward for everyone taking part.
  3. It offers the chance for attendees to build and expand networks and in the process, get real-life practical help from those in-the-know. Not sure how to structure a workshop? Struggling to analyse your research data? Looking for some metrics to show impact? Show up and ask your questions – chances are, someone will be able to assist.
  4. Often said, but people make our universities what they are. Taking time out for this event, to specifically think about the digital experiences of those groups of people, and how we can work to make them better, is the right thing to do and is therefore time well spent.
  5. It’s completely free to UCISA members (which includes most UK universities). Need I say more?

So much for the plug. If you missed attending the event live on 26 September 2026, don’t worry, the recordings are now available. More details below.

This year’s theme spoke to UX trends as well as the broader state of the sector

Our chosen theme for the 2026 community day was UX for the real world: having lasting impact in higher education. This theme served a dual purpose. Firstly, it acknowledged how UX practice in higher education is changing: less by-the-book, more adaptable to the realities of the way things work in the ecosystem. Secondly, it captured the potential for UX to be a ‘north star’ – a prioritisation mechanism to bring clarity and direction to the sector at a time when change often has to happen fast, decisions need to be made quickly and therefore unpredictability is rife.

The theme therefore worked at two levels:

  • Top down (strategic): it reflected a growing move to ‘do UX without telling me you’re doing UX’. In other words, treating UX as part of how the organisation already works, rather than as a separate function.
  • Bottom up (practical): it spoke to the need for UX practitioners to be opportunistic and adaptable in the midst of shifting conditions – for example, moving between short- and long-term priorities, bringing users and designers closer together, and keeping track of successes.

With this in mind, the overarching theme cascaded into five sub-themes:

  1. UX within the machine: Making UX happen relies on aligning with ‘how things work’ in your organisation – understanding how things actually get done in an organisation, and planning UX work to fit.
  2. Playing the long game: You’ve reached the end of the project or the initiative, but now what? Projects are often short term, so how do we keep UX going once they end?
  3. Putting numbers on it: How have you demonstrated the impact of UX in ways decision-makers understand? Making the case for sustained UX means showing its value, including in numbers.
  4. Designing with, not for: Co-designing with students, staff and other audiences – creating a more direct link between user needs and user-centred design.
  5. AI in UX: Beyond the chatbot: AI is everywhere – but how is actually being used in UX practice? Using AI as a tool to support UX work, and recognising its growing role in shaping user experiences.

With the theme set, we went public, we started accepting speaker proposals and opened the event for sign-ups in July, ahead of the event date on 16 September 2026.

The event was met with a positive community response – from the proposals, to the sign-ups to engagement on the day

A real test of a theme is the speaker submissions and the response was strong, particularly for the ‘UX in the machine’ subtheme. It’s always a great privilege to work with the UCISA events team and fellow committee members to select and sequence the sessions and to curate the schedule for the day. We aimed for a balance of themes, talks that would appeal to seasoned UX practitioners and newcomers alike, with a mix of formats. Taking on board feedback from previous events, we also built in time to pause, reflect and interact.

The event attracted nearly 200 registrations from UCISA members, with over 100 people from 49 institutions attending on the day. It was interesting to see a split across job roles of attendees. As expected, UX/UI design and content roles were the best represented. But the audience stretched well beyond them, taking in leadership and strategy, digital learning and education, as well as technical and systems roles. We were also joined by people working in digital experience, engagement and communications and people in library services and data and accessibility positions.

I’ve often found activity in the shared chat to be a good gauge of the success of an online event. On the community day, the Teams chat was lively from the start, with people introducing themselves at the start, then going on to share comments, questions and resources, right up until the event closed at 4pm. Even once the event was over, messages were being exchanged to set up follow-ups, suggesting the day had succeeded in bringing the community together over shared goals and challenges.

Watch the recorded sessions in order, or just pick those of interest

There’s nothing quite like attending a community event live to experience it as it happens and to be part of the conversations in your own way. That said, it’s not always possible to take a day out of the working week, so thankfully UCISA records the event and makes the recordings available through the UCISA Connect portal (free to access for all UCISA members).

Access the recordings through UCISA Connect portal (log in required)

You’ll find the recordings available in two slots – one morning and one afternoon. Underneath the recordings, you’ll also find slide decks provided by the speakers to help you follow along in your own time and revisit specific topics of interest.

Here’s a synopsis of each of the sessions to guide you through:

UX Within the Machine: Making user experience work inside complex institutions (Sam Fanning, Director of Digital Experience at University College London)

Long-standing UX community contributor Sam Fanning shared how UX works at UCL. She covered UCL’s product delivery model, how the digital experience team is set up, and how they measure the success of their UX service. Sam also explored how UX can get involved earlier, and the challenge of managing dependencies between the SaaS products and bespoke builds that underpin many UCL services.

How usable is it, really? When UX evidence starts to add up (Juliana Matos, UX Manager, King’s College London)

Juliana shared her experience of tackling a long-term UX problem: demonstrating impact through metrics. She discussed sample sizes, the familiar ‘five users is enough’ rule, and the difficulties of producing statistically significant numbers. Using the Moodle VLE as a case study from King’s, she showed how the team applied the well-established System Usability Scale (SUS), and why hearing directly from users was necessary to add qualitative depth to the scores.

‘Clearing’ a path through the chaos: transforming operational processes one-on-one (Jen Doyle, Senior Content Designer (Content Operations), University of Edinburgh)

Improving a key stage of the student journey is something anyone doing user-centred work in higher education can relate to. Jen Doyle described how she and her team analysed each stage of the Clearing process at Edinburgh, looking for ways to improve the content and streamline the processes behind it. They were able to make changes and improvement resulting in easier, more efficient updates for staff and a more consistent experience for everyone.

UX within the machine – panel discussion (Chaired by Emma Horrell, UX Manager, University of Edinburgh, Panellists: Sam Fanning, Juliana Matos and Neil Allison)

 Clearly I’m biased (again), but this 30-minute panel flew by and is well worth revisiting, even if you caught it first time round. Sam, Juliana and Neil described what ‘the machine’ looks like in their organisations, sparking ideas for making UX fit. They shared how they prioritise UX resource, show impact, and win buy-in and championing from the top, along with honest reflections on what has and hasn’t worked and advice for people new to the field. If you’ve ever wondered how to get UX applied where you work, there was plenty to take away.

Designing for students: Using quantitative UX research methods to design a new VLE student experience (Adam Ellison, Phil Tubman and Stephen Owens, all from Lancaster University)

The team from Lancaster shared a case study on improving the UX of their VLE, tackling inconsistent structure, content overload and unclear navigation. For me, this talk was a welcome chance to revisit how my own UX Service approached work on Learn, the primary VLE at Edinburgh. It was a chance to learn from Lancaster’s experience, and join a wider conversation about why VLEs deserve more UX attention.

Designing the Digital Journey to Wellbeing Support: Co-creation in Practice (Holly Thomson and Jade Alexander, User Researchers at the University of Glasgow)

Holly and Jade shared how Glasgow co-designed Wellbeing Home, a one-stop shop for wellbeing services, and Support Finder, a self-triage tool, working together with students with lived experience. They described how they used card sorting and a build-your-own homepage activity to work out how to organise services in ways that made sense to students and explained how they tested designs based on students’ real-life needs and scenarios, to ensure every user journey had clear steps for students to follow. They concluded with tips on making co-design work in practice.

AI in UX. Beyond the chatbot (Paul Boag, Boagworks)

UX expert Paul Boag shared practical ways to get more value from user research data using AI. Drawing on examples, he shared how feeding AI forgotten reports, surveys and slide decks on can surface fresh patterns and insights, and can shape virtual personas. He also revealed how AI can highlight gaps, to signal where more research is needed. His talk sparked discussion about the importance of being transparent when using AI to analyse and apply research data.

Exploring AI and the UX process (Madeline Linnell, Website and Application Developer, Oxford Lifelong Learning, University of Oxford)

Madeline shared the process her team at Oxford followed to develop an AI-assisted way for users to discover courses on the Lifelong Learning website. She covered the research process (including use of AI tools such as Google AI and Cursor) the ideation phase (using AI to brainstorm creative ways to discover academic learning opportunities), prototyping (using Google AI to generate interactive screens from static files) and user testing (using AI to analyse the findings), concluding with what the team learned and their plans for implementation.

Workshopping essentials: the fundamental principles for inclusive, democratic engagement and design (Neil Allison, Head of Prospective Student Web, University of Edinburgh)

Neil gave a practical talk which served as a primer on designing workshops. He unpacked when something should be a workshop rather than a meeting or an email, with workshops best suited to collaboration, democratising decisions, generating ideas and prioritising. He explored how to strike the right balance of facilitation to guide people through the divergent, emergent and convergent stages, and shared examples of successful workshops he has run at Edinburgh.

Thoughts, suggestions, reflections? – reach out to me on email

A UXer through-and-through, I take my responsibility of organising a user-centred UX Community Day seriously – therefore I’m always looking for feedback to guide how we do things better. If you signed up for the UX Community Day, you will have received a feedback form. If you didn’t, or the window to complete the form passed you by, you can still share your thoughts by emailing me directly. Drop me a line at ehorrell@ed.ac.uk I would love to hear from you.

 

 

What I learned in my summer internship researching digital content accessibility

In this post, User Experience team intern Hannah Watson shares her work over the summer researching digital content accessibility with the EdWeb 2 publishing community.

Introduction

As part of my internship with the User Experience Service, I have been investigating digital accessibility at the content level across University web pages, particularly on EdWeb 2 sites. Digital accessibility at the content level is about applying principles of accessibility to how web content is written and formatted. This includes, but is not limited to, content features such as heading levels, links, and alt text. It does not include anything that is not involved with content design or that is controlled at a higher level by the Content Management System (CMS), such as font, font size, or colour contrast.

Research aims and scope

The specific research questions for this project were:

  • How do web publishers learn about digital content accessibility?
  • What do web publishers know about digital content accessibility?
  • How do web publishers implement digital accessibility requirements and principles in their content?
  • What challenges do web publishers face in creating digitally accessible content?

More comprehensively, I was investigating the accessibility of:

  • Heading levels
    • Ensuring that heading levels are used correctly
    • Not using heading levels for emphasis
  • Links
    • Clear and concise link text which describes the linked destination
    • Link text which makes sense on its own
    • Avoiding URLs on web pages
  • Lists
  • Alt text
    • Clear link text that describes the image
  • Images
    • Ensuring that all images have appropriate alt text
    • Avoiding images of text
  • Videos
    • Including human-corrected captions with any videos uploaded to web pages
    • Making sure that transcripts are available for all videos uploaded to web pages
  • Italic, bold, and underlined text

For more detailed guidance on these topics, please refer to the University of Edinburgh editorial style guide.

Editorial style guide | Information Services

This research has been necessary for the User Experience team as it allows us to identify which areas of content accessibility are challenging for web publishers. From this, the team can adapt guidance and training to provide extra support on these more challenging areas where possible.

Research methods

Interviews with web publishers

I started my research by setting up short, informal interviews with six University web publishers. In these interviews, I asked the publishers about their experiences of creating digitally accessible content. The aim of these conversations was to understand how publishers learned about digital accessibility and what challenges they face when making content as accessible as it can be. During the interviews, we referred to web pages that these publishers work on to get concrete examples that illustrate the topics we discussed. Through these interviews, I was able to identify a number of trends, particularly in the challenges that the interviewees and their colleagues face.

Survey of EdWeb 2 publishers and the Web Accessibility Special Interest Group

Following these interviews, I created a survey to further investigate the findings and collate more supporting evidence for these findings. The survey consisted of 10 questions, three of which were demographic based as a filter, with a final question asking permission to follow up with those who responded.

The other six questions asked about:

  • which actions the participants took to make their content digitally accessible
  • what resources they used to do so
  • what challenges they face in doing so

The findings from the survey were effective in making the information gathered in interviews more robust and provided further evidence for some trends that were identified previously.

To publicise this survey, I sent a brief statement explaining my research into two Teams channels, the Web Accessibility Special Interest Group and the EdWeb 2 Community. Overall, the survey received five responses, and while this is a limited number, I found that it helped to support the findings from the interviews.

Analysis of Effective Digital Content workbooks

The final method of research that I used to learn about the digital accessibility of content on University web pages was by looking at pages that were submitted within Effective Digital Content workbooks as part of the course. The course requires learners to choose pages from a University website and assess the effectiveness of the content. By looking at the pages that learners selected, I was able to use active examples and make note of content accessibility issues that were present on live pages. This process also highlighted a number of trends.

This method of research was also useful in that it provided two separate sources of data. Firstly, the answers in the workbook helped me to gauge learners’ understanding of the principles covered in the course. Secondly, the web pages linked by those taking the course allowed me to have live examples of content to assess against content accessibility principles. While there was not necessarily overlap between the answers in the workbook and the pages submitted as part of the workbook (as some people may not have been fully or at all responsible for the content on those pages, or the content could have been updated since the workbook was submitted), it was useful to see these things separately.

Findings about accessibility

Combining findings from all areas of research for this project, I have identified a number of trends in the accessibility of digital content.

Headings, alt text, and links are the principles most often put into practice by participants

In the interviews and the survey, participants were asked what principles of accessible digital content they actively used when designing content. More than half of participants mentioned three principles in particular, with all participants mentioning at least two, which were:

  • using the correct heading levels
  • adding meaningful alt text to an image
  • writing clear and descriptive link text

Interestingly, while these principles were mentioned frequently by participants, and evidenced by the websites that we discussed in interviews, they are also principles that are often missed on University web pages. This is discussed in more detail in the corresponding sections further on in this post.

PDFs are hard to avoid

One standout finding from the interviews was that web publishers sometimes struggle to find an effective alternative to PDFs. While PDFs are not necessarily accessible, they do have benefits which make them useful for publishers. For example, they are downloadable, searchable, and cannot be easily edited without permission from the owner. There are ways to make PDFs more accessible, such as avoiding decorative images, adhering to accessible content design principles within the PDF, and checking colour contrast. However, the interviewees were more in favour of finding a way to turn their content into a webpage as this is more likely to result in accessible content.

Out of the survey responses, three also mentioned that they had recently chosen to publish content as a web page rather than as a PDF, highlighting their knowledge that a web page is preferable in terms of accessibility. However, their personal preferences or how difficult they found this is unknown as I was unable to follow up with these participants.

Heading levels are often skipped and headings vague

Across web pages that I assessed for correct heading levels, there were several which skipped heading levels throughout the page content, such as going straight to a heading 3 without that heading being nested within a heading 2 section. Additionally, headings on the University web pages that I investigated were often generic, instead of being specific about what a page or section will contain, which is the recommended approach.

The fact that the web publishers who were interviewed and surveyed were aware of the importance of correct headings levels and specific headings, and that the majority of these publishers have attended a staff training or used the editorial style guide, suggests that the guidance provided is accurate and useful for publishers.

To increase the use of correct heading levels, as well as clear and descriptive headings on web pages, participant responses suggest that increasing the reach and engagement of existing training and resources involving headings would be effective. This includes training provided by the User Experience Service, such as Effective Digital Content or Content Improvement Club, and the University of Edinburgh editorial style guide.

Alt text is well written but sometimes missed

In three of the interviews, and in two survey responses, participants mentioned that they struggled to find the time to add alt text to images on their web pages. However, participants in the interviews also stated that they understood the importance of alt text, and what writing meaningful and clear alt text involves. This is reflected in the answers in Effective Digital Content workbooks. The course contains a question asking learners to write meaningful alt text for two images, and this question is often answered well. This suggests that web publishers understand how to write alt text, and that the obstacle in doing so is more likely to be related to time and resource.

One way that time constraints on adding alt text could be improved is to reduce the number of images on a web page, which will not only make this task more manageable but also is also more sustainable.

Explaining acronyms and abbreviations is common

All five responses to the survey stated that they had explained an acronym or abbreviation recently. Although this did not come up in the interviews often – only once – the survey responses suggest that this is common practice for web publishers who are familiar with digital accessibility principles.

Link text is frequently inline or not descriptive

Participants stated that they understand the importance of clear link text as a principle and make effort to implement this into their content. However, similar to headings, this sentiment is not reflected across numerous University web pages. In some of the pages submitted as part of the Effective Digital Content workbooks that I investigated, there were frequent occurrences of inline link text, or link text that does not clearly describe the linked destination.

To increase the writing of link text on a separate line, as well as clear and descriptive link text, participant responses suggest that increasing the reach and engagement of existing training and resources involving links would be effective. This includes training provided by the User Experience Service, such as Effective Digital Content or Content Improvement Club, and the University of Edinburgh editorial style guide.

Findings about staff experience and engagement

A trend from both the interviews and survey responses is that for many of the people involved with this research, their knowledge of digital accessibility started with a personal interest. Specific examples of this that participants mentioned include learning about accessibility as a student (and then going on to be an accessibility advocate for a student society) and working with disabled students and learning through experience.

During the interviews, multiple staff members mentioned that they believe, through their experiences, that a large part of issues with creating accessible digital content at the University surrounds communication. A combination of factors was discussed that had communication at the centre, including:

  • the importance of digital accessibility not being widespread enough.
  • consistency about expectations between schools or areas of the University, such as one page being edited by multiple schools or areas and having different standards.

Job-based constraints were also mentioned frequently, such as limited time to add alt text to all images on a web page and working on a page where the lead publisher takes a design forward approach which can sometimes clash with accessible content principles. These answers were also reflected in the survey responses, with all of the responses mentioning either one or both of these problems.

What resources do web publishers use for learning about and developing their digital accessibility skills?

The primary resource that participants mentioned using to learn about digital accessibility and develop their skills was University-provided staff training, with 10 of 11 participants mentioning this. Effective Digital Content was specifically mentioned twice, and Content Improvement Club three times. In the interviews, two participants mentioned more general staff training, with one survey response saying the same. In the survey responses, three participants also said that they used training provided by the Disability Information Team.

The Web Content Accessibility Guidelines 2.2 (WCAG) is another resource that was frequently mentioned by participants, with six in total saying that this is a resource that they use as guidance on accessibility.

The University of Edinburgh editorial style guide was mentioned by five participants as a resource that they used to provide guidance on digital accessibility, in which the guidance reflects what publishers learn in training courses, meaning that information they take from the style guide is in line with accessibility and content design training.

What I learned from researching content accessibility approaches in EdWeb

I learned a lot during my time researching how web publishers at the University of Edinburgh approach creating accessible digital content. I thoroughly enjoyed the opportunity to work with staff from a variety of different areas of the University. This helped me to learn how to identify commonalities in interview and survey responses despite the areas of work being distinct. I also enjoyed this because I was able to learn much more about the work that goes on across the University and what the work of other teams involved.

I also particularly enjoyed the format of my internship being a combination of individual work and working with other members of my team and the Disability Information team. This balance allowed me to set my own goals while still getting the opportunity to work and learn collaboratively as part of a team, prioritising my tasks between my own and those that I was working on with others.

A limitation of my work was that it was much easier to contact and work with staff who have a genuine interest in the subject area of accessibility, which does not lead to research that is representative of how University staff approach digital accessibility as a whole. Having done the research that I have so far, a continuation of the research would be most beneficial if it focused on the experiences and approaches of staff who are less familiar with digital accessibility requirements and principles, as this would create a more well-rounded understanding of web publisher accessibility approaches at the content level.

Furthermore, another limitation of the work was the time constraints which have restricted my ability to develop solutions to issues identified during the research period. This was expected to a degree, as the solutions depended on the outcomes of the research, which has taken the majority of the 12-week period. The positive outcome of this is that the research and findings will provide the User Experience service with information that allows for both further research and for adaptations to guidance and training if it is deemed necessary.

Conclusion

This research has helped the User Experience Service to assess their training and the resources that are available for publishers to learn more about digital content accessibility. Keeping in touch with the publishing community through projects like this helps the team to direct their effort to real challenges that publishers face on a day to day basis.

The research completed throughout the duration of this project will be supported and advanced by further investigation, particularly by communicating with web publishers who are less involved with digital accessibility as a whole. This will likely help in providing more concrete solutions to digital content accessibility issues across EdWeb 2 pages.

Hearing about approaches to student voice at the Learning and Teaching Conference 2026

In June, I headed over to the Nucleus Building to attend the University’s Learning and Teaching Conference. The theme of the conference was student voice, and I was there to present about how the Careers Service and the UX Service had involved students in the design process for a recent website development project.

Emily McIntosh’s keynote “Nothing about us without us”

The day kicked off with a keynote from Emily McIntosh, Chief Student Officer at Harper Adams University. Emily set up the theme of student voice and covered some important distinctions for us to keep in mind throughout the day. Emily compared two ways that students are involved in decision-making processes: representation and partnership.

Representation is where staff within a university create something – a new policy, for example – then show student reps and ask for their opinion. While this is better than not involving students at all, it can lead to student reps feeling that they are responding to decisions that have already been made. The input from student reps sometimes comes at late stage, and can lead to tweaks but not necessarily substantive change.

Emily suggested that “you said, we did” posters can be indicative of this sort of approach. She contrasted this with examples of universities working in partnership with students from the beginning. In this way, students are actively involved in helping to shape services and policy. Emily described this as a relational, dialogic culture of ongoing collaboration, where students and staff work on problems together. This approach is sometimes described as co-design.

For me, there were some clear analogies in what Emily was describing to processes we advocate for in user-centred design. For example, bringing user input into the early stages of a design process is generally more effective than getting the same input later in the day. In addition to this, user-centred design often emphasises the benefit of ongoing collaboration and continuous iterative improvement, keeping users involved before and after the launch of a product.

Working in partnership like this is not necessarily a simple thing to achieve. But I appreciated the message: incorporating students into how we decide to do things as a university brings some significant advantages. Whatever the practical challenges of implementing this approach, moving from designing-for to designing-with offers us opportunities to better meet service users’ needs.

Highlights from the rest of the day

The rest of the day made me think about issues around student voice.

Neil Speirs (Widening Participation Manager) raised the issue of working class students often being absent from student voice projects. Neil’s talk made me think about how good quality user research often depends on recruiting a diverse group of participants. Recruiting the right research participants can be difficult. We see that all the time in our work. But without a diverse range of voices, user research can miss important perspectives and draw an incomplete picture of what matters to users of a product or service.

Neil Speirs speaking in a lecture theatre in front of a slide titled A neutral proposition

Neil Speirs presenting at the Learning and Teaching Conference. Slide reads: “A neutral proposition? We must be aware that student voice does not stand as a neutral proposition. Without democratic and participatory structures and cultures, student voice projects in learning and teaching will not provide democratic and participatory outcomes. We might reflect on the class based inequalities that are ‘baked into’ (Major et al, 2023) our structures, cultures, policy, curriculum and pedagogy. Which students are readily able to participate and benefit from student voice projects? Which students are invited to take part? Do we even notice? Is it the usual suspects? Do all students know the rules of the game and therefore why and how it might be advantageous to partake in student voice projects? Is the working class student voice heard in these spaces?”

Clare McKay (Student Recruitment and Admissions) and Christine Emmerson (Communications and Marketing) talked about how student recruitment markets can help us to anticipate student needs. One important point they made was that marketing is not just about promoting courses and programmes. It’s also about understanding our audience and whether there is a need. This resonated with aspects of UX work, where an understanding of user needs can tell us not just how to communicate information, but also whether that information is needed in the first place.

Jenna Fyfe and Carine Abraham (Edinburgh Medical School) talked about research they had done with alumni networks. Their project focused on alumni from three online postgraduate programmes at the Medical School. Alumni described how they create networks with each other following graduation, and how they often retain an interest in maintaining contact with programme teams from the University. This contrasts with a picture where alumni wish to maintain contact with university-wide structures. I found this interesting, because we sometimes think of people outside the University seeing us as a single entity unified by an overarching brand. In contrast, the alumni in this research weren’t thinking about a monolithic institution. Instead, they retained an interest in personal connections with a small, specific part of the University.

My talk was about our work with the Careers Service

In my talk, I gave an overview of how the Careers Service developed their website in 2025. In particular, I highlighted how the User Experience Service assisted with the design of collaborative activities that brought student voice into the work. I wanted to show how bringing student input into this sort of design activity makes the work better, and I talked about the benefits that the Careers Service achieved by approaching the project in this way.

You can read more about our collaboration with the Careers Service elsewhere on this blog:

Blog posts about work by the UX Service and the Careers Service

Reflections on the day

I thought there was a lot of value in holding a conference like this, with most of the speakers being staff and students from the University of Edinburgh. Presenters were speaking from a broadly similar context, which meant the talks had a high level of relevance to our work. There were a lot of familiar faces and it was good to catch up in the breaks with colleagues I hadn’t seen for a while.

My main takeaway was that student voice projects often have a lot of crossover with user-centred design approaches. In both cases, we’re interested in making things better by involving end users. I felt it was a great opportunity to hear about how student voice projects have been informing colleagues’ work at the University.

About the conference

University of Edinburgh Learning & Teaching Conference

Testing alternatives to video in student communications

Our previous research found that students often skipped videos on University webpages. This follow-up explores whether audio, timestamped videos or structured transcripts better support students in finding the information they need. 

In June, I wrote a blog post about whether students watched videos on University webpages, describing my research as part of the UX Service. 

Do students actually watch videos on websites? 

As part of that research, we ran a small qualitative study with three student participants, observing their natural behaviour as they encountered video content embedded in a student-facing newsletter from the University. We asked a simple question: when a University-wide communication points students to video content, do they actually watch it? 

Our first case study suggested the answer was often no. Two of the three participants were not engaged by video content, with one actively avoiding it and another being highly selective. In contrast, one participant was receptive to video, likely due to its relevance to his current situation, highlighting the importance of timing and context in determining video engagement. 

This blog post continues where that left off, testing why and what students would rather have instead. 

We investigated student reactions to alternative forms of video content

The obvious next question was what students would choose instead of a video if given the option. We decided to run A/B tests, which are controlled experiments that typically compare multiple versions of a digital interface, for example, a webpage to determine which performs better based on user behaviour. We recruited six participants and showed them the same newsletter in three different formats. 

  • Two participants were shown an audio version. 
  • Two were shown a video with timestamps. 
  • Two were shown a transcript, restructured into digestible text rather than auto-generated wall of text. 

The findings from this short experiment were interesting. We learnt that, of the three alternatives, the audio format was the least well perceived by the participants. The main reason was because an audio version offered no visual elements, no way to scan ahead, no quick way to extract the one piece of information someone actually needed. Even when the participants were explicitly told that audio was lighter in weight and reduced the carbon footprint of the web estate, it was still the least preferred option. 

The video with timestamps on the other hand, was much more positively received. Timestamps allow people jump straight to the relevant section rather than watching linearly, therefore allowing a more efficient way to consume information. 

Of all three alternative formats, the transcript was the clear preference of the participants, since this permitted them to quickly skim through the content and find the key information they were looking for. The transcript participants favoured wasn’t just “text instead of video”,which can often appear as a ‘wall of text’ it was text that had been deliberately restructured to pull out the key takeaway points. Taking these findings together, we learned that, for video transcripts to be easily read, and to ensure people can extract most value from them, it is worth putting effort into structuring and presenting text from transcripts, and not simply relying on the auto-generated ones.  

Our research revealed ways to make video content more impactful and effective

From a relatively small experiment, we learned a lot about the effectiveness of different formats of student communications and student preferences. The main findings were as follows: 

  1. When it came to assessing content in student communications, students did not seem to be primarily concerned with the digital sustainability of the format. While digital sustainability considerations did not drive student format preferences, it’s worth noting that the most preferred format (structured text) had the lowest environmental impact of those tested. 
  2. Audio did not appear effective for this type of student communication. Audio formats lack the scannability that students rely on, making them a less suitable choice for this type of communication. 
  3. Complying with accessibility requirements does not guarantee usability or a good user experience. A transcript can meet accessibility requirements, yet still fail to provide adequate scannability, highlighting the importance of considering both accessibility and usability. 
  4. Students prioritise their time and seek relevant information, emphasising the need for formats that can be easily updated and kept current. 
  5. More research is needed on video and text complementarity. Further study is required to determine the most effective ways to combine video and text to support communicating with students in the most effective ways possible.  
  6. When researching the effectiveness of student communications, hearing direct feedback from students is essential. A useful next step would be to ask students directly about their experiences with structured text content versus timestamped video and compare the accuracy as well as the completeness of their recall. 

We’re planning to continue our research into effectiveness of video content

Our findings have implications for the development of educational content. We shared our results with the Careers Service, who were interested in using our research to inform their content development. This collaboration has the potential to impact how they work with media in the future. Looking ahead, we would like to investigate how students interact with videos on social media platforms like LinkedIn, as their attitudes towards video seem to differ depending on whether it’s for entertainment or educational purposes. This could provide valuable insights into the role of video in student learning and engagement.

Can AI do a content audit of my website? A review of different ELM models and AI products

Content audits are a long-time staple of the user-centred website toolkit, but they take time and effort to complete, which many website owners struggle to find. In my ongoing AI experimentation, I tried using AI tools to help me with auditing web content.

Every day brings a raft of new AI developments. When choosing how to use AI, I’m less sold on getting it to do creative tasks for me because I like doing those myself. Instead, in pursuit of freeing up my time, I prefer to use AI to help me with the tedious tasks I never get around to, the ones I know will take me longer than I think, the ones I put off again and again. Here’s looking at you, content auditing.

Websites are easy to grow but difficult to keep in check

When people come to the UX Service for help with their websites, they tend to use several phrases to describe their site. These have included:  ‘It’s a mess!’ ‘It’s a bit out of control’. ‘It’s grown arms and legs’. None of this is unusual when you consider the lifecycle of websites. Over time, websites change hands. Pages once carefully crafted are easily forgotten as new editors take over. New content is created without knowing what’s already there. Very often, content needs to be published to a deadline and it’s quicker to publish a new page than seek out and amend an existing one. Seldomly do publishers want to get rid content that’s taken time and effort to produce.

Read some thoughts on content housekeeping by UX Service team members (current and past):

Digital housekeeping – applying content management practices to improve digital sustainability by me

Be a gardener by Ari Cass-Maran

Content Audit Findings and the 100k Challenge by Milo McLaughlin

How to get a grip of your website (and then keep hold) by Neil Allison

A content audit is the best place to start improving a website (not the homepage)

When people seek to bring order to disarray on a website it’s difficult know where to start, so typically, they start on the homepage with ideas to change images or add new content. While this gives sites a superficial makeover, it doesn’t really help site audiences in search of content. Fewer and fewer website audiences are starting out on homepages, instead they’re parachuting directly into site content from Google search and increasingly, from AI summaries. What does this mean? It’s more important than ever to make sure your web content is up-to-date, accurate, useful and relevant for your website users. Here’s looking at you, content auditing.

Read more about getting your content ready for AI from a recent Content Improvement Club session:

Auditing a website needs a methodical approach (which is where AI can help)

If audits are so important, why are they so easy to put off? Short answer, they can take prolonged amounts of time and concentrated effort. The larger and more complex your site is, it’s likely that there will be more content to review, and the more difficult it may be to maintain a consistent auditing approach. Going through multiple, long pages of web content, it can be easy to run out of the time you’ve allocated for your audit. Breaking the audit into chunks may lead to different results from different sessions as biases creep in. Bringing colleagues in to help can share the load, but can also introduce new perspectives and contradictory auditing decisions. There’s a need to maintain ruthless objectivity, to counter human subjectivity and avoid errors infiltrating. Here’s looking at you, AI.

If you can plan and describe a website content audit, you can ask AI to help you do it

Website content audits can be done for many purposes, but commonly, doing a content audit involves making decisions about what content to keep, what needs to go and what needs to be changed – all in pursuit of reaching a future state of your website. A good way to decide what content to keep is to consider who it’s for and the purposes it serves. If a piece of content doesn’t meet audience needs and doesn’t align with the site purpose, then it’s probably a candidate for deletion. A good content audit of a website therefore typically requires several things:

  1. An inventory of every piece of content in the site. Typically arranged in a spreadsheet, with one row per webpage. Can be organised into sections or content types, or grouped by root URL
  2. A note of the site audiences, ideally in order of priority (answering the question ‘Who is this site primarily for?)
  3. A list of the main reasons those audiences visit the site (answering the question ‘What tasks do people complete on this site?’) together with a list of the main goals of your site (answering the question ‘What do we want people to do on this site?’)

Considering each of these things as different pieces of data, where item 1 is data to be audited, items 2 and 3 are factors or auditing criteria used to decide what happens to those data (typically one of three outcomes: ‘Keep’, ‘Delete’ or ‘Modify’), I reasoned that I could give AI these data, tell it what a content audit was and and ask it to complete this process for a given website.

I experimented with ELM to help me do a content audit of a website

The University’s AI platform, ELM was the natural place to start experimenting with applying AI to aid content auditing. The ELM interface allows you to input a prompt, as well as add additional files and data sources. It also allows you to select different models to handle queries. I was curious to see the differences between the different models – both non-reasoning ones and reasoning ones so I needed to design a prompt that would work for both.

Read more about ELM and its models

ELM website

ELM new model guides (University log in required)

Being mindful of the environmental cost, I wanted to start with a small AI model (Llama 3.3)

AI comes with a significant environmental cost, which is highly dependent on the LLMs you use. Powerful models with more reasoning power are especially resource-intensive to run, so to avoid unnecessary wastage of tokens and other resources I’ve found it’s sometimes best to start with a smaller model and size up accordingly, as the need arises. The trade-off is that smaller models have limited reasoning power, however, so the responses they provide and the tasks they are able to complete can be limited.

Based on my previous experience of using AI and thinking about its potential application to a typical content auditing process, I had a hunch that providing a small model with something like a site inventory (such as a XML sitemap) and detailed data about website audiences and purposes could overload its context window and produce inferior results.

I therefore decided to begin with a lightweight prompt ‘Can you help me do a content audit of this website? (with a link to the website)’ to give a small model something manageable and learn how it approached handling the query. I chose a website that the UX team had recently worked on, so I was familiar with its content, and I entered the prompt, starting with the Llama 3.3 model – an open weights model (costed on computing power rather than per-token usage) which is hosted in University data centres (instead of being cloud-based). I toggled on the option to include web search so ELM was able to access the website.

The same prompt to different ELM models produced varied results

Having starting with Llama 3.3, I moved on to other models: Open AI’s GPT 5 and then Anthropic’s Sonnet 4.6. To assess and critique responses from each, I adopted the mindset of a person reluctant to begin a content audit and looking for AI help to kick-start action on this, and used this as a roleplay to review and compare the responses.

The response from Llama 3.3. was too generic to convince me to start auditing

Despite a glitch in formatting the response, Llama 3.3. provided some general information about content auditing. Its response included an initial assessment of the site, which made objective judgements of the homepage (assessing it as ‘well-structured, with a clear introduction’), the navigation menu (labelling it ‘easy to use with links to key sections’), content quality (noting it as ‘high-quality, informative and engaging’) and accessibility (pulling out that the site had an accessibility statement).

The response then acknowledged that in order to do a thorough content audit on the site in question there was a need to provide guidance on the aspects to focus on, and rounding off, the response noted some of the aspects a typical content audit would examine: content quantity, content quality, accessibility and consistency across different parts of a site.

Reviewing this response, I felt that if I was a site owner looking to improve my site, this response could very easily persuade me that I didn’t need to bother auditing it at all – and I wasn’t left any wiser about how to meaningfully provide guidance on the aspects to focus on if I did decide to proceed with an audit.

Likelihood to convince me to audit: 3/10

Open AI’s GPT 5’s response was detailed and thorough but a bit overwhelming

As would be expected from a reasoning model, the response when I used GPT 5 was more detailed than Llama 3.3’s. Arranged in two sections, the first part included preliminary observations using the site URL provided – which were split into the following categories:

  • Information architecture and navigation – commending the site’s positioning on user intents, yet suggesting improvements with more explicit audience pathways and call-to-action signposting
  • Content coverage and freshness – using the site’s range of content types to assess content structure and assessing freshness based on recency of news items
  • Call to actions – acknowledging that the site had were multiple calls-to-action which could be combined into primary CTAs for each audience
  • Accessibility and usability – calling out the need for more meaningful and less duplicated alt text on the site images
  • Trust and governance signals – recognising the role of the site’s strong branding to demonstrate authenticity
  • SEO – suggesting the site’s page titles and meta descriptions were reviewed for relevance

The second part of the response outlined a plan for a full audit. The plan began by posing three questions to be answered:

  1. What are the primary goals? e.g. increase visitors, boost traffic, showcase outputs
  2. Who are the priority audiences? (with suggestions of the different groups)
  3. What’s in scope? (suggesting either the entire site or specific sections)

The response then provided a suggested series of 7 steps to follow to complete the audit:

  1. Inventory and crawl (with detail of the inventory to start with and the tools to complete the crawl)
  2. Editorial quality review (with detail of criteria to score each page – such as inclusivity, accuracy, audience-fit etc)
  3. UX and IA review (with suggestions to evaluate navigation labels and connections between content and pathways to achieve top tasks)
  4. Accessibility (with suggestions of accessibility checking tools to use as well as manual checks to complete)
  5. SEO and technical (with suggestions to check data points like metatags, internal links, robots files and performance)
  6. Analytics (with suggestions to use analytic data such as bounce rates, page views, site searches etc)
  7. Recommendations and roadmap (detailing prioritisation measures to make the identified site changes)

It rounded off suggesting a template structure for the content inventory (and offering to create this template as a Google Sheet), an idea for a scoring rubric, and reiterating different tools to use (naming products like Screaming Frog, Google Analytics and SEMrush). To conclude, it offered to do the audit, requesting answers to the three initial questions and asking for access to a sitemap, external tools and governance documents (such as style guide and brand voice guidelines).

Reviewing the response, which was presented as text content, I found it a bit too much to scan and digest, and putting myself in the place of someone facing content auditing with some reluctance, it felt that this large amount of detail could be off-putting to even start the process. I could appreciate the relevance of suggesting of external tools, but knowing that each of those would require setting up accounts and logins, as well as going through a process to integrate with Open AI, it had the effect of making me mentally push content auditing to the bottom of a to-do list.

Likelihood to convince me to audit: 6/10

Anthropic’s Sonnet 4.6’s response gave structured guidance which was easier to follow

The output from the Anthropic model stood out compared to the others from ELM since it was not just plain text, instead, it had been structured with headings, tables and icons. This made scanning and digesting this response much easier than the others.

The response was structured into 10 sections, as follows:

  1. Site overview – table summarising the site name, owner, primary purposes and date of confirmed content
  2. Information architecture – list of the top-level navigation elements with an assessment of what worked well and issues identified in a bulleted list
  3. Content inventory – table of pages, with noted audiences, details of last update (if known) and inferred status (on a red, green and amber scale)
  4. Calls to action – list of CTAs on the homepage, with assessments and recommendations
  5. Content freshness – table containing list of most recent items in each section and related assessment
  6. Accessibility – list of accessible features that were present, table of items that needed checking and recommended accessibility tools to make the assessment
  7. SEO and technical – table containing an SEO assessment, considering elements like meta descriptions, structured data, mobile responsiveness – each with an allocated status, associated notes and combined recommendations
  8. Content quality and tone – list of observations and recommendations
  9. Priority recommendations summary – table of priority actions with an effort/impact score
  10. To complete the full audit I need – list of requirements to complete the full audit (including sitemap, CMS page list, Google Analytics data, etc.)

Content within each of these sections was comparable to that included in the GPT 5 response, but in the Sonnet 4.6 response, these sections referred directly to information from the site in question – for example, instead of making general assessments of site aspects like navigation and content coverage, it contained direct detail about the site itself with mention of specific pieces of content, noting what was already known (for example, referring to issues noted in the accessibility statement). Arranging this in tables with use of icons and red, green and amber priority indicators made it easy for me to pinpoint what to address and with what urgency, and having this consolidated in the final table was easier to parse than reading it in text format (as in the GPT 5’s response).

Reviewing this response from Sonnet 4.6, it was the best of the bunch from ELM, it contained enough detail to help me get started but not so much that it was overwhelming. Putting myself in the position of a time-poor site owner unsure where to start, I had clear pointers of the priority areas and a mix of quick wins and bigger tasks to choose from.

Likelihood to convince me to audit: 7/10

Three sets of two side-by-side screenshots, showing the output from Open AI GPT 5.5 (on the left) compared to the output from Anthropic's Sonnet 4.6 (on the right) showing Anthropic's more structured and visually appealing approach to presenting the responses

Three sets of two side-by-side screenshots, showing the output from Open AI GPT 5.5 (on the left) compared to the output from Anthropic’s Sonnet 4.6 (on the right) showing Anthropic’s more structured and visually appealing approach to presenting the responses

As well as ELM, I experimented with Chat GPT and Claude to provide content audit help

Models within ELM can also be accessed through interfaces provided by AI products. Two of the best-known AI products are Chat GPT (by Open AI) and Claude (by Anthropic). I was curious to see the responses each of these products would provide in response to my query asking for content auditing help so I prompted each of them in the same way. In both cases I used the free versions.

Chat GPT gave a suggested approach and produced initial findings

The response from Chat GPT was comparable to the output from ELM choosing an Open AI model, but the response was structured and annotated with tables and icons, in a style similar to the Anthropic response. It began with a suggested audit approach which set out different areas to consider and related questions. In the next section (headed ‘Initial observations’) it contained an analysis of the site itself, which included site strengths and highlighted the following six areas for improvement on the site in question:

  1. Homepage is quite academic – pulling out some audience questions that would be relevant to be answered on the homepage instead of the content that was there
  2. Navigation could be more task-focused – noting that the current navigation was driven by organisational structure rather than tasks users would want to complete
  3. News archive – picking out the need for recent articles
  4. Calls to action – acknowledging that many of these are worded to provide information not prompt an active response
  5. Content consistency – advising a uniform page structure for easier reading
  6. Audience segmentation – recommending clearer delineation between groups of site users

The response concluded with an example audit table and some content recommendations, organised under ‘Keep’, ‘Improve’ and ‘Review’.

Reviewing this response as a whole, I liked the up-front suggestion of an auditing approach and the example audit table which was then supplemented with detail about the site itself and recommendations relevant to different aspects of the site in question. This helped me visualise what the audit outputs could look like, and I could see ways to get there.

Likelihood to convince me to audit: 7/10

A screen showing the output from Open AI after being prompted for help content auditing. The suggested audit approach laid out in a table with columns for the different areas of auditing and associated questions to ask in the right-hand column

A screen showing the output from Open AI after being prompted for help content auditing. The suggested audit approach laid out in a table with columns for the different areas of auditing and associated questions to ask in the right-hand column

Another screen showing the output from Open AI after being prompted for help content auditing. An example audit table is laid out with rows for each URL, and columns containing associated purpose, audience, quality and recommendations.

Another screen showing the output from Open AI after being prompted for help content auditing. An example audit table is laid out with rows for each URL, and columns containing associated purpose, audience, quality and recommendations.

Another screen showing the output from Open AI after being prompted for help content auditing. Content recommendations are laid out with colour coding for 'Keep', 'Improve' and 'Review'.

Another screen showing the output from Open AI after being prompted for help content auditing. Content recommendations are laid out with colour coding for ‘Keep’, ‘Improve’ and ‘Review’.

Claude presented interactives to actively take me through an audit process

Anthropic’s Claude interface took a different approach to the other AI products, immediately asking about the main goal of the audit with options to pick. I picked ‘full inventory’. It then asked the desired breadth of the audit with another set of options, from which I picked ‘full site crawl’. The next screen asked me about the output format, and I picked ‘spreadsheet’.

From top to bottom, the interactive screens produced by Claude asking about the main goal of the audit, the breadth of the audit and the format for the audit output. Each has multiple choice answers to select

From top to bottom, the interactive screens produced by Claude asking about the main goal of the audit, the breadth of the audit and the format for the audit output. Each has multiple choice answers to select

After approximately 2-3 minutes, with a running commentary of what was being done, Claude presented a list of ‘headline findings’ of bugs, orphaned content, out-of-date content and broken content (dead links, incomplete calls-to-actions etc). It also produced a downloadable Google Sheet with two tabs – the first containing a summary of the issues to be fixed, their locations and a reason why they needed to be fixed, the second with a full inventory, with one row for every URL in the site and columns logging the page title, content type, date, status (with colour-coded options: ‘Keep’ (green), ‘Update’ (amber), ‘Rewrite’ (dark amber), ‘Review’ (blue) and ‘Critical’ (red)) and recommended actions for each page.

Screenshot of the audit output spreadsheet produced by Claude, showing one row per page URL and different columns logging content types, date, status and recommended actions.

Screenshot of the audit output spreadsheet produced by Claude, showing one row per page URL and different columns logging content types, date, status and recommended actions.

 

Reviewing this response, it was by far the most effective at making me feel I was starting the content auditing process and supporting me through it. Being presented with the opportunity to answer questions upfront meant I could control how the audit went, and having a spreadsheet created for me to review saved a lot of effort making this from scratch by scraping the site for a sitemap and page metadata.

Likelihood to convince me to audit: 9/10

Conclusion: If you’re putting off auditing your website, try getting AI to help

If you’re looking to improve your website, doing a content audit is one of the best steps you can take – working out what you have, what’s working and what’s not will put you in a strong place to move forward and make positive changes. Depending on how well you know your site, and how much time and resource you have to spend on it, you may appreciate some help to get started auditing its content.

From my quick experimentation with the different AI tools available I learned that, with the exception of the small non-reasoning model, each AI product offered something to help initiate the content auditing process of a website. At a basic level, asking AI to look at your website will pull out something you may not find yourself such as an out-of-date page, an irrelevant CTA or an accessibility glitch. AI can also take a zoomed-out look at your site that you may be too close to adopt yourself – such as reviewing your navigation pathways, appraising your audience segmentation or analysing the consistency of your content voice and tone.

If you’ve never audited a website before, AI can offer a lot of guidance to get you started, and whether you choose to do it yourself taking recommendations from the AI on tools and/or processes or whether you prefer to hand the heavy-lifting to AI to get your audit spreadsheet started – leaving you to deal with the more nuanced decisions that require specialist knowledge and experience, these tools can save you time and effort and may just be the nudge you need to do this all-important website maintenance task.

 

Using scenario-based research to understand how colleagues find inclusive language guidance

Earlier this year, the User Experience (UX) Service researched how colleagues find and use the University’s Inclusive Language Guide. This post explores why we used scenario-based research, what it revealed about colleagues’ behaviour and how the findings are improving the guide. 

The Inclusive Language Guide is only useful if people can find it 

The Inclusive Language Guide sits within the University’s Editorial Style Guide and provides practical guidance when writing about disability, race and ethnicity, and sex, sexuality and gender. The guide was developed in 2022 through a co-design process involving colleagues from across the University.

Inclusive Language Guide 

Having useful guidance is only part of the challenge. Colleagues also need to know it exists and be able to find it when they need it. 

More recently, we reviewed the Inclusive Language Guide to understand how discoverable it was for colleagues creating digital content. Emma Horrell has written about how this work prompted improvements to the guide’s visibility and signposting across the University. 

You don’t know what you don’t know: Improving the way we position inclusive language at the University

Why we used scenario-based research 

We wanted to understand more than whether colleagues were aware of the Inclusive Language Guide. Awareness alone does not tell us what people do when they need support, so we used realistic scenarios to observe how colleagues responded to situations where they might need guidance. 

We focused our research questions on four areas: 

  • How colleagues approached decisions about inclusive language. 
  • Where colleagues looked for support and whether they could find relevant guidance. 
  • Colleagues’ awareness of the Inclusive Language Guide and what they expected it to help with. 
  • Any gaps, ambiguities or areas of confusion. 

These questions shaped the structure of the interview and activities we designed. 

Researching colleagues approach to inclusive design decisions 

We combined semi-structured interviews with scenario-based activities to understand both colleagues’ existing experiences and how they approached real content decisions. 

We spoke to nine colleagues who create digital content 

We carried out nine interviews with colleagues in different roles across the University. Participants had a range of experiences creating digital content, including two student interns who brought perspectives of those newer to the University’s digital landscape. 

We began by exploring participants’ existing experiences 

Before looking at the scenarios or the guide itself, we asked participants about the types of content they create and whether they had ever needed to check terminology or wording before publishing. For example, we asked whether they had been in a position where they needed to consider how to word something to make sure the language was not excluding anyone. 

This helped us understand the approaches colleagues already used and the sources they relied on when dealing with these topics. 

Realistic scenarios showed how colleagues look for guidance 

We introduced a series of realistic scenarios based on everyday content tasks. 

Rather than asking whether participants knew about the guide, the scenarios allowed us to observe how they approached these situations in practice.

Example scenarios included: 

  • writing promotional content about a Pride initiative and wanting to ensure the language used was fair and respectful 
  • writing about someone who had experienced trauma and wanting to avoid misrepresenting them 
  • checking the preferred terminology for the chair of a company board before publishing a news article 

In posing the scenarios, we gained insight into where participants would usually look for this type of guidance, how confident they felt making language decisions and what they understood by the term ‘inclusive language’.

We adapted our scenarios as we learned more 

One of the original scenarios asked participants how they would write about someone elected to lead a company board. Rather than focusing on terminology, several participants described how they would research the individual or organisation before publishing. 

While this reflected realistic editorial practice, it did not help us answer the research question we were exploring. 

For later sessions, we refined the scenario so that it focused specifically on checking the preferred terminology for the role before publication. This encouraged participants to think about language choices and produced richer insights into where they expected to find guidance. 

What we learned: Four themes around finding and using inclusive language guidance

Across interviews, there were four themes that emerged consistently. 

Colleagues valued inclusive language even when they had not heard of the guide 

Across the interviews, participants consistently demonstrated that they valued using respectful and inclusive language. They were already thinking carefully about inclusive language, but they did not always know where to access University specific guidance.  

Many described that they would check terminology with colleagues or seek advice from specialist teams when writing about unfamiliar topics. They wanted confidence that they were using appropriate language. 

Awareness of the importance of inclusive language was much higher than awareness of the Inclusive Language Guide itself. 

Colleagues expected to find guidance through different routes 

The scenario-based activities helped us understand where colleagues naturally looked for support. 

Participants frequently described turning to colleagues, managers, Equality, Diversity and Inclusion (EDI) resources, staff networks and external websites. A few participants did eventually land on the guide, but they had not recalled it or looked there first. 

Several participants also relied on previous examples when writing content. Others spoke about balancing official guidance with lived experience and perspectives from the communities being represented. 

This highlighted an important consideration. Colleagues expected inclusive language guidance to be connected to the places they already look for support, linked from elsewhere rather than existing solely within the Editorial Style Guide.  

Improving signposting therefore remains just as important as the guide itself. 

Colleagues understood the phrase ‘inclusive language’ in different ways 

Participants interpreted the phrase ‘inclusive language’ in different ways. 

Some described it as “using the correct terminology”, while others framed it more broadly as “not leaving anyone out”. Those with accessibility backgrounds often drew direct connections between inclusive language and accessibility. 

Despite these differences, the term ‘inclusive’ was recognisable, with several participants using it unprompted when working through the scenarios.

Colleagues looked for practical guidance first 

After exploring how participants would approach different situations, we asked them to navigate the Inclusive Language Guide. This helped us understand whether the structure matched their expectations when looking for support. 

Several participants were unsure where the practical guidance began. Some mistook the introductory blog post for the guide itself, while others expected the examples and practical advice to appear much earlier on the page. 

Although participants valued the context and principles behind the guidance, they prioritised practical information they could apply immediately. The examples within the guide were consistently highlighted as the most useful content. 

The research continues to inform improvements to the Inclusive Language Guide 

Our findings have allowed us to identify some key areas for improvements to the Inclusive Language Guide.

Making practical guidance easier to find 

One area of focus is the guide’s landing page. We are reviewing its structure so that practical guidance is easier to find while retaining the contextual information that participants found valuable. 

The findings highlighted that the current structure does not always support the way colleagues use the guide. Participants wanted to move quickly to practical guidance while still being able to access the wider context when needed.

We are also reviewing page titles, navigation labels and the ordering of content to make it clearer where the guide begins and how it relates to the wider Editorial Style Guide. We’re considering how to better support both first-time and repeat use, making it easier for people to quickly find practical guidance while still providing the context and principles that many participants found valuable.

We’ll also consider how best to surface practical examples and support the different ways colleagues use the guide, whether they are reading it for the first time or returning to quickly check terminology and writing conventions.

Improving connections with related guidance 

The research also prompted us to consider how the guide connects with related guidance across the University’s web estate. 

Participants frequently expected to find inclusive language guidance within EDI resources or other specialist guidance. Improving discoverability means not only improving the guide itself, but also making sure it is connected to the places colleagues naturally go when seeking support.

As outlined in Emma’s blog, since completing the research, we have worked with colleagues responsible for complementary guidance across the University’s web estate to improve signposting between services.

This work will help ensure colleagues can find appropriate support regardless of where they begin their journey. 

Scenario-based research helped us understand behaviours 

The research reinforced an important principle in content design: guidance needs to be easy to find as well as useful.  

By combining colleagues’ existing experiences with scenario-based activities, we gained a better understanding of how people approach inclusive language decisions. These insights have already informed improvements to the Inclusive Language Guide and will continue to shape how we approach future content research.

QuillMark: The Technical Details Behind My Drupal Style-Guide Assistant

I previously introduced QuillMark, a tool I built to help web publishers check content against the University’s editorial style guide. In this post, I explain the technical design and methodology behind the prototype.

Introduction

My name is Shlok, and this summer I am working as an AI and UX Innovation Intern in the Information Services Group (ISG). During my internship, I have been developing QuillMark, a prototype tool designed to help web publishers check their content against the style guide. This post outlines the system design, methodology and AI pipeline behind the prototype, including how deterministic checks, language models and a MapReduce approach work together to improve reliability while keeping publishers in control.

 

QuillMark highlights violations in your existing text rather than generating an entirely new version. This:

  1. avoids the risk of AI making unnecessary changes to unrelated parts of the page, which would require a manual audit each time
  2. allows publishers to clearly see each violation and accept or reject suggested changes
  3. reduces output tokens, lowering both hallucination risk and cost

 

Problem: The style guide contains a large number of rules, around 60–70. If all rules are sent to the LLM at once, the model may not check the text against every rule reliably.

 

Research: https://openreview.net/pdf?id=R6q67CDBCH

 

Solution → MapReduce (Divide and Conquer: Map → Collapse → Reduce)

 

System Design and Methodology

 

Pipeline:

  1. Deterministic checks to flag blatant violations
  2. Fine-tuned small model flags likely obvious violations
  3. Large model audits the final page deeply for complex deeper violations

 

The deterministic checks are used to flag “obvious” violations that are directly detectable using regex or code. For example, this could include the use of a forbidden word, or a spelling convention such as using “benefited” instead of “benefitted”. These issues can be detected using a direct search of the page and do not require AI.

 

The second stage uses a smaller fine-tuned model. This model would be fine-tuned on the Style Guide rules and training examples. It would again be used to detect relatively “obvious” violations, but ones that are not simple enough to be detected reliably through deterministic checks. These issues are not highly complex, so they should still be visible to a smaller model.

 

After the publisher fixes the deterministic violations and the violations found by the fine-tuned small model, there are likely to be remaining deeper and more complex issues that both previous approaches did not pick up. The final large model can then focus mainly on these more complex issues. This means it spends less attention on obvious violations and more attention on issues that are easier to miss.

 

This improves the process in two ways.

 

First, it allows the final large model to focus on deeper issues. If the whole page was passed directly to the large model from the beginning, it would likely flag many obvious violations first and might miss some subtler issues. By fixing the most obvious issues earlier, the large model can focus more on smaller, judgement-based, or easily missed problems.

 

Second, it may reduce cost compared with running the large model multiple times. One alternative would be to run the large model once, fix the obvious issues + some deeper ones it finds, and then run the large model again to find the remaining deeper issues. In the first run, the large model may spend much of its output on obvious violations. In the second run, after those obvious issues are fixed, it may be able to look more deeply. The proposed approach tries to achieve a similar effect more cheaply by using deterministic checks and a smaller fine-tuned model before the final large-model review.

 

This pipeline is shown in Figure 1.

 

Style Guide Violation Detection Pipeline
Figure 1: Style Guide Violation Detection Pipeline

 

Using MapReduce to reduce instruction-following degradation

Based on personal experience with large language models, as well as existing research such as the Curse of Instructions, we cannot rely on a large language model to follow a very large number of rules at once. In this case, the style guide contains around 60–70 rules. If all of these rules are passed to the model in a single request, the model may not check the page against every rule properly. It may follow some rules, ignore others, or miss violations because there are too many instructions to apply at the same time.

 

To solve this, I propose using a MapReduce approach, which is a divide-and-conquer technique: Map → Collapse → Reduce.

 

The rules would be broken into smaller chunks, with each chunk containing a fixed number of rules. Each chunk would then be sent as a separate request, asking the model to check the page only against that smaller set of rules. This allows the model to focus on fewer rules at a time and check them with more attention.

 

After all the chunks have been processed, the responses would be gathered together and combined into one report. This is the collapse stage. The combined findings would then be passed through another large language model request in the reduce stage. This final request would clean up the output by removing duplicate findings, resolving overlapping issues between chunks, and filtering out possible false positives or hallucinated violations.

 

This approach makes the checking process more reliable because it avoids asking the model to apply all 60–70 rules at once. Instead, each group of rules is checked more carefully, and the final report is produced by combining and cleaning the results. This helps ensure that all rules are checked with more equal rigor, rather than relying on the model to remember and apply every rule in one large prompt.

 

The MapReduce approach is shown in Figure 2.

 

MapReduce Approach for Style-Guide Rule Checking
Figure 2: MapReduce Approach for Style-Guide Rule Checking

 

Update: Large-Model Cleanup from the Reduce stage was excluded from the prototype as the probability of duplicates occurring is low and the cost of such a cleanup is relatively high with minimal benefit.

 

Read more about the technical system plan: https://blogs.ed.ac.uk/website-communications/building-quillmark-testing-a-drupal-style-guide-assistant-with-web-publishers/

Watch a video demo: https://media.ed.ac.uk/media/t/1_yozad8ie

 

Future enhancements:

  • Strengthen Deterministic Regex-matching with more training data (web pages)
  • Audit deterministic findings using a cheap, local AI

Building QuillMark: testing a Drupal style-guide assistant with web publishers

I introduce QuillMark, a Drupal prototype designed to help web publishers check content against the University’s editorial style guide. I reflect on how I developed and tested the tool with publishers, focusing on what their feedback revealed about usability and how it will shape the next version.

Introduction

My name is Shlok, and this summer I am working as an AI and UX Innovation Intern in the Information Services Group (ISG). During my internship, I have been developing QuillMark, a prototype tool designed to help web publishers check their content against the style guide. This blog post explains the problem I was trying to solve, how I built and tested the prototype, what I learned from publishers and what I plan to do next.

 

The problem

Our style guide contains more than 60 rules covering areas such as spelling, punctuation, tone, formatting and terminology. Previous research showed that web publishers found it difficult to remember and apply every rule while writing and editing content.

 

This is understandable. Publishers are often working under time pressure, and checking a page manually against dozens of rules adds a significant cognitive burden. Even experienced publishers can miss small details, particularly when they are concentrating on the meaning and accuracy of the content.

 

I began exploring whether a tool could make this process easier. The aim was not to replace publishers’ judgement or rewrite their content automatically. Instead, the tool would identify possible style-guide violations, explain the relevant rule and allow the publisher to decide whether to apply or dismiss each suggestion.

 

This became QuillMark.

 

Read more about the Style Guide

 

Why I built the prototype in Drupal

Drupal is where publishers already create and edit web content, so it made sense to build the prototype directly into that environment rather than create a separate tool.

Before starting development, I attended Drupal in a Day to understand the platform, its content-editing interface and how a custom tool could fit into an existing publishing workflow.

 

Building QuillMark within Drupal meant publishers could check their content without copying it into another system. It also allowed me to test the tool in a realistic environment, using the same types of fields, buttons and interactions that publishers encounter in their day-to-day work.

 

Read more about Drupal in a Day

 

Planning the checking process

Before writing the prototype, I planned the system in detail.

 

One of the main design decisions was that QuillMark should highlight individual issues in the existing content instead of generating a completely rewritten version. Rewriting an entire page could introduce unnecessary changes and would require the publisher to audit every sentence. Showing individual findings makes it clearer what the tool has identified and keeps the publisher in control. This also prioritises the HITL (Human-in-the-Loop) framework, which I intend to use in my AI-based projects to ensure that people remain involved in reviewing decisions made by AI.

 

The system design proposed three types of checking:

  1. Deterministic checks for clear violations that can be identified using code or regular expressions, such as prohibited words or spelling conventions.
  2. A smaller AI model for relatively straightforward issues that require more context than a simple text search.
  3. A larger language model for more complex, judgement-based rules involving areas such as tone, structure or plain language.

 

The plan also used a MapReduce-style approach. Rather than asking a language model to apply all 60–70 style-guide rules in one prompt, the rules would be divided into smaller groups. Each prompt could then concentrate on a limited number of rules before the results were combined. This was intended to reduce the risk of the model overlooking instructions because it had been given too many at once.

 

The original design included an additional AI step to remove duplicate findings and resolve overlaps. I left this out of the prototype because duplicates were expected to be relatively uncommon, while the extra model request would increase cost and complexity.

 

Read more about the technical system plan: https://blogs.ed.ac.uk/website-communications/quillmark-the-technical-details-behind-my-drupal-style-guide-assistant/

Coding the prototype

I developed the prototype iteratively, beginning with a small number of style-guide rules and expanding the checks once the basic workflow was working. For development, I used Codex. I gave it my plan and asked it to design the architecture. After reviewing it, I asked it to proceed with the implementation.

 

The first stage used deterministic checks for issues that could be found reliably without AI. These checks searched the content for recognisable patterns and returned the location of the issue, the relevant style-guide rule and, where appropriate, a suggested correction.

 

The second stage used a large language model to identify issues that depended more heavily on context. I divided the rules into smaller prompt groups so that the model could focus on a manageable set of instructions during each request.

 

The prompts asked the model to return structured findings rather than a rewritten page. Each finding needed to include enough information for QuillMark to:

  • identify the relevant content;
  • explain the problem;
  • show the related style-guide rule; and
  • suggest a possible correction.

 

Publishers still had to approve or dismiss each finding. This human-in-the-loop approach was intentional. Style rules can depend on context, and an automated suggestion will not always be appropriate. QuillMark was designed as a decision-support tool, not an automatic editor.

 

Watch a video demo: https://media.ed.ac.uk/media/t/1_yozad8ie

 

Designing the usability test

Once the first iteration was complete, I adapted an existing usability-testing template to create a test for QuillMark.

 

I conducted three sessions with web publishers. Participants were asked to work through a series of tasks covering the main parts of the prototype, including running the checks, understanding the two stages, locating an issue in the editor, reviewing a rule and applying or reverting a suggested fix.

 

I observed how participants used the interface, where they hesitated and whether the information on screen matched their expectations. I also asked how they might use the tool as part of their normal publishing process.

 

The purpose was not only to find technical bugs. I wanted to understand whether the concept itself was useful and whether the interface communicated how the tool was intended to work.

 

What publishers told me

The publishers thought the tool would be useful as a final check before publishing. They said they would still use their own judgement and would not automatically accept every suggestion.

 

Most of the feedback was about the usability of the tool. Some parts were unclear, such as the difference between the two stages and some of the button labels. There was also too much information on the screen, and some issues were repeated. The publishers also wanted to see what a suggested fix would change before applying it.

 

What I learned

The testing suggested that the core idea is useful, but the user experience needs to become simpler.

 

Publishers do not necessarily need to understand which checks use regular expressions and which use an AI model. They need to know what issue has been found, why it matters, what the proposed change is and what action they can take.

 

The sessions also reinforced the importance of designing for scanning. More explanation does not always create more clarity. In a publishing workflow, concise labels, clear states and well-grouped findings may be more valuable than displaying every piece of supporting information at once.

 

Next steps

My next step is to refactor and review the prototype code before making changes based on the usability feedback. This will include reviewing the AI-generated code to check whether each section is needed, whether the same result could be achieved more simply, and whether any parts repeat code that already exists. I will compare the generated code with the rest of the codebase so that it follows the same structure and does not add extra functions or files without a clear reason.

 

Where the code is longer than needed, I will simplify it by removing repeated checks, combining similar sections, and reusing existing functions. I will also keep individual files to a manageable size, generally no more than 1,000 to 3,000 lines depending on the file. Larger files will be split where this makes the code easier to follow, and hardcoded rules will be moved into the database where possible.

 

The next iteration will focus on:

  • clarifying or simplifying the two-stage workflow;
  • reducing repeated and overwhelming findings;
  • improving button labels, status messages and colour distinctions;
  • previewing changes before they are applied; and
  • making the interface more concise and easier to scan.

 

June Content Improvement Club: Is your content AI ready?

Our June Content Improvement Club session focused on AI. In particular we discussed the changes in search behaviour, how that’s impacting the way people consume content and some of the things you can do to make your content ready for this new search landscape. Attendees also got the chance to work in groups to put some of our top tips into practice.

This topic clearly resonated with with our community, as it was fully booked shortly after being advertised, so we ran a second session to enable more people to attend.

Changes in search behaviour in the wake of AI overviews

We started off the session by providing a bit of context around what has led to the change in the way people now search for information.

AI overviews help answer your question on the search results page itself

In 2024 AI overviews were launched by Google to answer search queries directly on the search results page. AI overviews take snippets of content from relevant webpages (often combining multiple sources) using Google’s existing search index and the same ranking systems as regular search.

Since their introduction, AI overviews have rapidly increased and are now a prominent feature on the majority of search results pages. U.S data from the end of 2025 shows that overviews appear in over 60% of all search queries, doubling from 2024.

New Data: Google AI Overviews Now Appear in 60% of Searches

The information source is attributed using tags, which contain a link to the website itself. However, we are seeing a trend of people finding the information they need and relying on the overview without visiting the website where the information originated. This is known as ‘zero click behaviour’.

Screenshot of an AI search summary about how to apply to the University of Edinburgh, with the University logo displayed. There is a circle around the source tag for emphasis.

Example of an AI overview

 

 

 

 

 

 

 

 

 

The rise in ‘zero click behaviour’

The introduction of AI overviews has meant that people have started to become accustomed to getting an instant answer on the search results page. People are more often relying on overviews, rather than clicking a link to a relevant webpage to locate the information themselves. This means fewer people are making it to your webpage and if they do it may well be later on in their search journey.

In Q1 2026 studies report that 65% of searches on Google did not result in a click to anywhere.

Zero-Click Crisis Worsens – SuperPrompt

This figure is predicted to be significantly higher for those using mobile when searching.

Also statistics from 2025 show that 80% of people used answers directly on search pages and 50% relied on AI summaries for answers.

Zero-click searches reshape marketing strategies in 2025

These changes in search behaviour do require you to think a bit differently about the way your content is being used. This is summed up well by Guus Goorts, an education marketing coach, who says:

LLMs [Large Language Models] increasingly visit  your website on your audiences’ behalf.

Ways to make your content ready for AI

In the session we focused on the things that you can do to optimise your content for AI. Hopefully you will be glad to know that much of the evergreen content design advice we provide still stands strong in the face of the new AI search landscape.

Clear, self-contained content gives AI better material to work with

As AI can sometimes take just snippets of your page or grab sentences out of the wider context, one step you can take is to try to make sure that content makes sense even outwith its surrounding context.

The more self-contained and specific your content is, the more likely it is that AI systems will interpret it correctly and present accurate information to users. You can achieve this by avoiding:

  • non-specific or generic headings that rely on page context, but do not make sense when lifted off the page – for example ‘Funding’ rather than ‘Postgraduate funding options for international students’
  • references to content elsewhere on a page such as ‘see below’ or  ‘as mentioned above’ as AI may retrieve the sentence without the content it refers to
  • vague descriptors in the form of unclear or missing subjects such as ‘This must be submitted with your application’ as if the surrounding text is missing, neither the reader nor AI knows what ‘this’ refers to
  • videos without transcripts, captions or summaries as AI tools cannot watch video content, they process text far more reliably than video

Answer questions directly and avoid unnecessary preamble or woolly language

AI predicts likely answers from the content it finds. It can potentially miss the answer if it’s buried in long winded sentences, or not clearly explained.

In the session we gave the example of a prospective student asking: “What accommodation does the University offer?”

We provided two example bits of content that the AI could draw from.

Example 1 : ‘Our accommodation includes single rooms, shared flats and accessible housing.’ This contains the answer directly, it names the accommodation types and it can be lifted into an overview without losing meaning.

Example 2: ‘There are lots of options of different accommodation with various numbers of rooms depending on your needs.’ Whilst this sounds friendly it doesn’t provide much useful information. Phases such as ‘lots of options’ and ‘to suit your needs’ take up space without answering the question.

Content that is well structured and accessible is still key

Making sure your content is both well-structured and accessible is still key when we are looking at how AI will use and interpret it.

Frequent, meaningful headings add structure

Headings are one of the main ways people navigate content, they act as signposts, helping users quickly identify relevant sections and understand what information sits underneath them at a glance.

Frequent, meaningful headings also play an extra role in supporting the AI-generated summaries we know are becoming increasingly common. This is because headings help AI understand the structure of a page by breaking content into logical, retrievable chunks.

Headings also create a hierarchy that shows the relationship between different pieces of information on a page. Therefore, you should always mark up headings with the correct heading level, rather than using bold, underline or larger font sizes. AI also uses the heading hierarchy to:

  • understand, summarise and cite your information
  • work out which topics are primary, which are subordinate, and how information is grouped together

You can read more about using headings effectively in the style guide, our previous blog post and also on Caroline Jarrett’s ‘Editing that works’ website.

Accessible content and AI overlap

Content that has been made accessible is a big step towards being AI-ready as screen readers and AI systems rely on many of the same signals, although the outputs are different. Therefore many of the things you hopefully already do to make content accessible will also help AI understand it. We can demonstrate the overlap by looking at headings, link text and alternative text.

Headings: screen reader users use headings to navigate the page structure, whilst AI tools will use them to understand and prioritise your content.

Link text: descriptive link text provides context and helps screen reader users decide where to go next, whilst AI tools will use this link text to help interpret page meaning and relationships between content.

Alternative text: well written alternative text helps users who cannot see images extract meaning from them, whilst it also provides text that AI can use to understand image content.

You can read more about how to make your content accessible in Mel’s blog posts from previous Content Improvement Club sessions:

Descriptive link text supports users and AI

Descriptive link text is important for users as it helps them understand where the link will take them at a key decision point. This is increasingly important for AI too as AI search and retrieval tools use links and their surrounding context as signals to understand the relationship between different pieces of content.

Therefore, a link labelled ‘Accommodation options for postgraduate students’ provides a much stronger signal than ‘Learn more’ as it helps AI understand both the destination content and how it relates to the current page.

Descriptive link text also helps AI-generated summaries and citations. When AI retrieves content to answer a question, it attributes the source(s). Clear link text can reinforce what a linked page is about and strengthen the connections between related topics.

You can read more about writing effective link text in the style guide and Caroline Jarrett’s ‘Editing that works’ website:

Write for humans, using plain language

In the age of AI, it’s still important that we write for humans first and foremost. This means using language that your users understand within your content and explaining terms they might not know. Short paragraphs and sentences will also help.

Google’s advice for content to be successful with Google Search is to focus on creating ‘people-first content’, rather than content made ‘primarily to gain search engine rankings’.

Creating Helpful, Reliable, People-First Content | Google Search Central 

You may find Nick’s blog post from a previous Content Improvement Club useful on this topic:

Now you’re speaking my language – what we covered in the February Content Improvement Club session

We worked in small groups to put the theory into practice

After our whistlestop tour of advice to help make content AI-ready, we then worked in small groups, using pages that attendees had brought along to the session. We each picked one or two areas to focus on and shared suggestions to improve the content and talked about what was perhaps already working well. This created some interesting discussion and allowed attendees to gain feedback and new perspectives on their content from colleagues they may not usually work alongside.

Frequently Asked Questions – the debate continues

The debate around the effectiveness of Frequently Asked Questions (FAQs) is one which has been circulating for some time within the content design world. It has reared it’s head again in relation to AI, given the rise of the ‘answer economy’ where people are typically asking more focused questions and looking for a quick answer.

The main point of contention is whether or not we should be writing more content in a question and answer format. There are different views on this topic, we’ve seen advice to use questions in your content that aim to match the questions that users are asking AI tools.

On some interpretations of this advice, you could read this as a call for more FAQs on websites. We’d like to urge caution before you take this approach though as FAQs aren’t without their problems.

FAQs can often:

  • duplicate content that you have elsewhere on the site – this causes problems for site maintenance and search engines don’t tend to like it
  • answer questions that aren’t really frequently asked
  • make content hard to scan because they push keywords away from the left of the screen as they often start with ‘Can, What, How, Where’

So in summary, don’t rush to change all your headings to questions and create more FAQ sections. If you do use questions, answer them without waffle.

The University of Bath have written a helpful article about this topic.

Why AI doesn’t need Frequently Asked Questions (FAQs) | Digital Content and Development

Content maintenance is more important than ever

The final point we covered was the fact that content maintenance is more important than ever. This is because all published content can be cited by AI to generate an answer regardless of when it was last reviewed or updated. Therefore ensuring that you know what live content you have and whether it’s up to date is key if you want AI to bring back accurate information in response to search queries.

Duplicate content can also confuse AI, so regularly auditing your content to make sure there is effective cross referencing between pages and that the same information doesn’t appear on multiple pages, will again make it easier for AI to interpret and find what it needs.

How to hear about future sessions

We promote these sessions via our mailing list. If you’re interested, please sign up:

Join the UX and Content Design mailing list (University login required)

Suggest a topic for a future session

We’re keen to continue covering topics that colleagues across the University would find useful. It would be really helpful if you could let us know any ideas you have using this form:

Suggest a topic for Content Improvement Club (University login required)

Other training that we offer

More training is listed on the User Experience Service website:

Training | User Experience Service

❌