by Matthew Deeprose, University of Southampton, UK.
As an educator, you may notice digital accessibility barriers your students face. A student with wrist pain struggles with note-taking. A colourblind learner cannot interpret diagrams. A blind student cannot access handwritten solution sheets.
These moments of friction represent opportunities. With generative AI, you can transform observed problems into working solutions, often within hours rather than months.
This approach works because you possess something invaluable: direct knowledge of the barriers your students encounter. You understand the context, the constraints, and what “good enough” looks like for your specific situation. The missing piece has traditionally been technical implementation – out of reach for those who are not software engineers. Generative AI can help to bridge that gap.
In my presentation, Building Digital Accessibility Tools with Generative AI, I showed tools I built to address digital accessibility challenges in my institution and shared follow up resources for those who wanted to try this for themselves.
Here are some tips to help you get started.
Recognising solvable problems
Not every barrier translates into a buildable tool. Before investing time, validate whether your idea fits certain criteria.
Suitable problems share these characteristics:
- Your solution runs in a web browser without storing data on servers.
- You can describe the inputs, outputs, and steps involved.
- Similar tools or libraries already exist that you could adapt or combine.
- The problem affects multiple students or recurs across cohorts.
- You can test whether the solution works by using it yourself.
Avoid problems that require:
- user authentication or login systems
- storing personal data in databases
- real-time coordination between multiple users
- payment processing or sensitive transactions
A student needing downloadable transcripts from your lecture capture system represents a solvable problem. The input exists (video captions), the desired output is clear (clean text without timing codes), and the browser handles everything.
Validating your idea before building
When you lack technical expertise, you face a particular challenge: you may not recognise whether your idea is feasible or whether you have specified it correctly. Three validation strategies help address this uncertainty.
Search for existing solutions
Someone may have already solved your problem. Spend 30 minutes searching for tools, browser extensions, or web applications that address your barrier. If you find something close, you understand the problem is solvable. If existing tools fall short, you now have reference points for what yours should do differently.
Describe your tool as a user journey
Write out the steps a person would take: upload a file, see a preview, select options, download the result. If you cannot articulate this journey, the problem needs further definition. This description becomes your starting prompt for the AI.
Identify the technical components
Your tool will likely combine existing libraries rather than inventing new techniques. Colour manipulation? Libraries like Chroma.js handle this. Document conversion? Pandoc, DocShift, and gotenberg might provide the solutions you seek. Chart generation? Chart.js can generate visualisations. You need not understand how these libraries work internally. You need only know they exist and that AI can integrate them for you.
Standing on the shoulders of giants
The most reliable approach to building accessibility tools involves combining established libraries rather than creating everything from scratch. This principle, standing on the shoulders of giants, reduces complexity and increases reliability.
Consider a tool to check whether diagram colours work for colourblind students. You could attempt to implement colour vision deficiency simulation from mathematical first principles. Alternatively, you could use existing, peer-reviewed simulation algorithms already implemented in JavaScript libraries. The second approach produces better results in less time.
When prompting an AI to build your tool, specify which libraries to use. State that you want the tool to use Chroma.js for colour contrast checking rather than describing contrast calculation from scratch. The AI will integrate the library correctly, and you benefit from code that thousands of developers have tested and refined.
Testing What You Build
Validation continues after building. You must verify that your tool functions as intended and remains accessible itself. Three testing approaches catch most problems.
Console testing surfaces JavaScript errors. Open your browser’s developer tools, load your tool, and check for red error messages. A working tool produces zero console errors. If errors appear, share them with the AI and request fixes.
Keyboard testing confirms accessibility for users who cannot use a mouse. Press the Tab key to navigate through your tool. Can you reach every button, input, and control? Can you activate them with Enter or Space? If keyboard users cannot operate your tool, it creates new barriers while attempting to remove others.
Automated accessibility testing catches common issues. Browser extensions like Accessibility Insights run checks against established guidelines. These tools identify missing labels, insufficient colour contrast, and structural problems. Share the results with your AI, and it will correct the issues.
From one tool to many
Your first tool teaches you the process. The second takes half the time. By your third or fourth tool, you recognise patterns: how to structure prompts, which libraries solve which problems, what testing catches issues early.
Document your tools and share them with colleagues. Add Creative Commons licensing so others can adapt your work. The embryology lecturer who wanted colourblind-friendly diagrams created a need for the first tool I made. The solution now helps anyone preparing scientific illustrations.
Begin with a barrier you observed this term. Describe it as a user journey. Search for existing libraries that handle the technical components. Prompt an AI to combine these elements. Test the result. Iterate until it works.
You already understand your students’ needs better than any external developer could. Generative AI provides the technical implementation. Together, you can remove barriers that previously seemed immovable.
Useful resources
I have created a web page with resources designed to accelerate your first projects. You will find a system prompt I developed through a year of building accessibility tools with AI. This prompt instructs the AI to prioritise accessibility, use semantic HTML, and follow established coding patterns. I have also shared the complete conversation that produced my Panopto transcript tool, so you can observe how prompts evolve through iteration and testing.
Whether you prefer learning from worked examples or starting with tested templates, these resources provide foundations you can build upon rather than starting from scratch.
Matthew Deeprose, University of Southampton, UK
