Which website accessibility tasks should we give AI?

Use AI to assist with accessibility foundations during development, rather than leaving accessibility until the final stage. Code assistants can help developers build interactions that are time-consuming to implement, such as a customised dropdown menu designed to work for mouse, keyboard and voice users. Language models can also analyse an image alongside the surrounding page content to draft meaningful alternative text instead of an uninformative placeholder such as "Image1.jpg". Give AI the repetitive implementation and drafting work, with alignment to the Web Content Accessibility Guidelines (WCAG) as the aim. Treat the generated code and descriptions as work to review, not confirmation that accessibility requirements have been met. Automating baseline tasks should leave designers and developers more time to consider how people experience the website.

Keep human judgement responsible for the result by having designers, developers and accessibility advocates review AI-assisted work. AI can misinterpret images, misunderstand cultural context and reproduce biases in its training data, potentially excluding the people accessibility work is intended to support. Review image descriptions for accuracy and context, and evaluate whether generated interactions meet users' needs rather than accepting faster production as a successful outcome. In Level Access's 2026 survey of professionals in the US, UK and Europe, among respondents using AI, 61% said it accelerated design and 58% planning, but only 30% said it accelerated QA and testing. The same survey found that 89% of professionals said people must remain closely involved in accessibility as AI use grows. Use AI as a collaborator for repetitive tasks while retaining human responsibility for understanding the experience.