Uses of artificial Intelligence for accessibility

As artificial intelligence (AI) is becoming increasingly popular in workplaces, how can it help us with accessibility, and how can it hinder us?

With accessibility AI should be seen as a tool to help you, but not a tool to rely on. Accessibility requires a human touch to understand how a disabled user uses something. AI can guide you but not necessarily complete the task. AI can support accessibility work, but it cannot replace the complex understanding that comes from lived experience, user testing and real-world context.

And I’m going to share a few examples of why.  And then how it can help.

AI generated alternative text

You can ask AI to generate alternative text (alt text) for an image. But how accurate is it?

I asked Copilot to generate alt text for an image of one of my tattoos. And rather surprisingly it gave me a pretty accurate description of it.

It’s description was “a tattoo on a persons lower leg showing a styalised black snake coiled around an ornate frame, with text inside reading “only you can decide what breaks you”.

And that is very close to how I would describe it. I would maybe split it into 2 sentences rather than one.

I then asked Microsoft Word to autogenerate alt text and it said “a tattoo on a persons elbow with a snake and writing”. Which is significanlty different. And wrong.

That was just 2 examples. But my advice here is ask with caution.

On a website, it may not be necessary to add alt text to an image. And AI probably will not be able to advise you on this. Some images are purely decorative. Therefore they do not need alt text, as a user who cannot see that image is not missing context. However, images that are informative always require alt text.

AI tools often struggle with context. They can dexcribe what is visually present but not if the image is meaningful, decorative, or functional.

As with alt text, context matters. See [how to write alt text article] for more information and help with writing alt text.

AI generated captions

Some AIs can autogenerate captions in your videos.

Now again, this comes with a proceed with caution warning. Always check your captions are accurate.

Some automated caption tools can misunderstand people’s accents. If someone is speaking quickly it can misunderstand or miss words. Sometimes it adds fillers in, such as urm, and err. Which can add cognitive load to the user.

In general, they are not always accurate enough for people who rely on captions to understand content.

AI captioning tools are not a person. They cannot capture context and tone. And this means sometimes meaning is not conveyed.

Automated captions do not meet accessibility standards unless they have been reviewed and mistakes corrected.

Uses of AI

That is not to say that AI cannot help you.

You can ask AI a multitude of questions relating to accessibility. And as always with AI, read the output before using it.

Summaries

Often, long documents can overwhelm people. You can use AI to create a summary that is easier to read than the long document. Always ask AI to use plain English and keep it simple. This reduces the cognitive load on colleagues. And it’s also useful for busy colleagues who need a quick summary.

As an accessibility checker

You can ask AI to look through a document and ask for its recommendations to make it more accessible. Always check what it recommends, and that it has not edited meaning in your document.

You can put Excel spreadsheets, Word documents, PowerPoint presentations, Google documents, and more through AI.

AI can only highlight potential issues. It cannot perform an accessibility audit.

Agents

You can build an accessibility agent that can answer colleagues’ questions.

For example, you can get it to answer questions such as:

  • How do I make my PowerPoint presentations accessible?
  • What is the best way to check colour contrast?
  • What are the key principles of web accessibility?
  • How do I make sure my digital content is accessible for everyone?
  • How do I write in plain English?

I built mine in Copilot studio. And I gave it information from internal resources, such as our accessibility guidelines and principles. Content from Scope for Business. Content from W3C. And documents the accessibility team have produced internally.

This helps support colleagues with accessibility, independently, and helps create confident colleagues.

Directly supporting disabled colleagues

AI can directly help support colleagues. This is not an alternative to proper adjustments though.

Wording

AI can proofread and edit the tone of documents or emails. This is useful for neurodivergent colleagues, people who have not slept well, or those who worry that they come across differently.

AI can also simplify complex language or jargon. Explain concepts to make sure they are clear and understood.

Breaking down tasks

You can also use AI to break down tasks. So, asking “please give me the steps required to complete [task]”.

  • Tell me how long each step should take
  • My deadline is in 6 weeks
  • Be specific in how I can complete each step of the task
  • Suggest websites or software that can help me

If the response gives you too much detail, or not enough detail. You can go back and ask for additional information.

Cognitive and neurodivergent support

Visual structuring of information

AI can convert text-heavy content into:

  • bullet points
  • tables
  • flowcharts

This is to help colleagues process information.
Prompt: “Please turn this document into a bullet-point summary with clear headings.”

Reducing cognitive load
Ask your AI to simplify instructions or policies into step-by-step guides.
Prompt: “Break down this onboarding guide into 5 simple steps with estimated time for each.”

Creating checklists
It can also generate checklists for tasks, meetings, or projects.
Prompt: “Create a checklist for preparing a quarterly accessibility audit.”