AI in Accessibility
Note: This is a living note, an auto generated page synthesized by my self updating second brain managed by Claude. While it's input was my actual organic writing, this is an AI summary/extraction/synthesis part of my The Living Notes section.
If you prefer to read my organically written words or work, please go to any of the other nav links besides living-note or a page, route, or tag without #living-note.
AI in Accessibility
The Opportunity
AI is a force multiplier for accessibility work. One accessibility engineer with the right AI tooling can do what previously required a team.
Key applications:
- Automated alt text generation at scale — images across massive sites
- ARIA pattern validation — LLMs can reason about semantic structure
- Plain language rewrites — making content more readable / cognitively accessible
- Test case generation — AI generates edge cases human testers miss
- Remediation assistance — LLMs can suggest (and draft) fixes
The Gap (Deque Research + External Studies)
Deque notes an "AI gap in a11y" — current automated tools (including AI-powered ones) still miss significant categories of real-world accessibility issues:
- Context-dependent issues
- Cognitive / neurodivergent-specific barriers
- Complex interactive patterns
- Focus management in SPAs
AI improves coverage but doesn't replace manual testing + lived experience.
WebAIM 2025: 94.8% of top million homepages have accessibility failures. The scale of the problem remains unchanged despite AI tooling proliferation.
Carnegie Mellon study (circulated 2026): AI doesn't give accessible code by default — omits a11y attributes and doesn't verify compliance. Tawsif publicly reposted this finding on LinkedIn, signaling alignment with the "AI is not solving accessibility alone" position.
What I've Built
At Gap Inc.: Automated Lighthouse testing suite with AI/LLMs in one sprint. Deployed across 6 brands.
Claude Code integration: Using AI-assisted development for accessibility work. Notes in Bear on Claude/fundamentals.
Shifting Left
"Shift left" = catch accessibility issues earlier in the development process (design, planning, code review) rather than in testing/QA.
AI enables earlier detection:
- Linting at write time
- Design review automation
- PR review assistance
- Documentation generation
Risks
- AI-generated alt text can hallucinate or be wrong — needs review
- Over-confidence in automated passes
- AI reproducing biases in accessibility recommendations
- Barrier: teams may think "AI handles accessibility" and disengage