AI & Accessibility
AI can assist accessibility work; its claims and proposed corrections need verification.
Overview
Treat generative AI as an assistant for explanations, code review, and possible corrections. Its confidence does not establish whether a barrier exists or whether a correction works.
Why It Matters
A useful observation can still cite the wrong criterion or recommend an incomplete correction. Verify each claim against the actual task, source, and behavior.
Course Sequence
- Observe — record what actually happened.
- Claim — identify the specific suspected problem.
- Evidence — check authoritative sources, measurements, or reproducible behavior.
- Test — choose a method that could establish the claim.
- Fix — address the actual problem.
- Retest — repeat the test and check that the task still works without new barriers.
What to Look For
- Supported: the observation and checked source support the conclusion.
- Partially supported: useful observation, but an overstated or incomplete explanation, classification, or correction.
- Unsupported: the evidence does not establish the claim.
- A role and accessible name do not supply keyboard activation. Test behavior.
Standards & Guidance
- Check the actual WCAG success criterion and level. Do not treat an AI citation as verified.
- Use fictional course samples or concise observations from an approved public-page review with an assistant. Do not share sensitive data; no paid model is required.