How should we use AI to improve website accessibility?

Use AI to assist with accessible interface code and context-aware alt text, while keeping human review and accessibility testing essential to acceptance. Evaluate time savings after review, correction and QA, not just initial generation. Treat adaptive interfaces as a design possibility, not proof that a website already works for people with different accessibility needs.

Yellow heart with a black outline and the handwritten words Good design in the center on a beige background.

Automate baseline work, not accessibility acceptance

AI can support website accessibility by helping developers build interface features and draft meaningful image descriptions during development. Code assistants can help implement custom dropdown menus intended to work with mouse, keyboard and voice input. Image analysis can use surrounding page content to propose alt text rather than leaving a filename as the description. These are opportunities to integrate accessibility into the foundation of a website, rather than add it at the end.

However, faster code generation is not the same as faster delivery of an accessible website. In his account of a high-pressure digital build, Lead Developer Sergio Agosti described rapid implementation alongside bloated, repetitive code and a continuing need for expert oversight. For an enterprise website, assess the complete workload: generation, human review, accessibility QA, correction and ongoing maintenance. Initial coding speed alone does not establish a net accessibility implementation saving.

Sources: The Prompt: Using AI to Create a More Accessible and Dynamic Digital Future, Cracking the code: How AI is transforming high-pressure digital development

Review image descriptions for accuracy and page context

AI-generated alt text needs editorial judgement because describing an image is not the same as explaining its purpose on a page. A model can analyse both an image and surrounding content to propose a contextual description, but the result still needs to be accurate, complete and concise for that use. When the same image appears in different contexts, assess each description against the message the image communicates in that location rather than assuming one description fits every page.

Human review is particularly important because AI can misinterpret visual content, misunderstand cultural context or reproduce biases from its training data. Those errors can exclude the users the description is intended to support. Keep content editors, designers and accessibility advocates involved in deciding whether the wording conveys the relevant information. Treat generated text as a draft to evaluate, not as an accessibility requirement satisfied merely because text is present.

Sources: The Prompt: Using AI to Create a More Accessible and Dynamic Digital Future, Creating connections through design

Distinguish adaptive possibilities from proven accessibility

AI could support interfaces that adapt to different accessibility needs, but adaptation should build on an accessible foundation. Potential applications include increasing button hit areas for someone struggling with fine motor control, simplifying complex navigation or multi-step forms, and providing cognitive support through summaries, simpler vocabulary or explanations of unfamiliar concepts. These are possibilities for designing flexible experiences, not demonstrated outcomes for a deployed website.

The practical distinction is between generating an adjustment and establishing that the adjustment helps a person use the service. Designers, developers and accessibility advocates should remain central to that judgement. An adaptive interface should not be treated as permission to leave the underlying experience inaccessible. Use AI as a collaborator in exploring and implementing support, while retaining human responsibility for accuracy and for understanding how people experience the resulting website.

Sources: The Prompt: Using AI to Create a More Accessible and Dynamic Digital Future

Stats

UK Government Digital Service simplified testing between January 2022 and September 2024 found keyboard issues on 63.6% of monitored websites.

Government Digital Service

UK Government Digital Service simplified testing between January 2022 and September 2024 found insufficiently visible focus on 76.3% of monitored websites.

Government Digital Service

FAQs

Which website accessibility tasks should we automate with AI?

Use AI to assist with repetitive implementation work, including accessible interface code and context-aware alt-text drafts. Code assistants can help develop custom dropdowns for different input methods, while image analysis can consider surrounding page content when proposing descriptions. Keep human review and accessibility QA in place, because generated code or text does not establish that the resulting experience is accessible.

Why does AI-generated alt text need human review?

AI-generated alt text needs human review because models can misinterpret images, misunderstand cultural context and reproduce training-data biases. A useful description must also fit the image's purpose on the particular page, rather than simply identify visible objects. Review the draft for accuracy, relevant detail and contextual meaning before publishing.

How can AI adapt websites to different accessibility needs?

AI could help websites adapt by enlarging button hit areas, simplifying navigation and multi-step forms, or offering summaries and explanations of unfamiliar language. These are potential applications for supporting motor and cognitive accessibility needs, rather than established results for a particular website. Build accessibility into the underlying experience and keep human designers, developers and accessibility advocates responsible for evaluating the adaptations.