How to Use AI to Build Accessible Products
summary
Demo compares an axe-core linter (0/7 then 3/13 issues) with GitHub Copilot (5/7 ≈71% then 11/13 ≈85%), plus tools like VSCode, NVDA, Wave and Lighthouse. AI finds quick, repeatable issues, suggests fixes, and can autofix PRs in CI/CD, but misses semantic, contextual problems and hallucinates (adds unnecessary ARIA or fabricates suggestions). Prompt engineering and feeding standards (WCAG) improve results, yet manual checks and user testing remain required for true accessibility and to avoid regressions.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:25Meet Carrie Fischer
- 1:43What is AI
- 4:25Why do we need to write accessible code
- 10:20Where to find accessible code
- 11:00Pros of AI Accessibility
- 16:15Writing Accessible Code
- 22:00Limitations of AI
- 23:53Boosting AI Accessibility
- 25:47Adding a Pro
- 31:05Conclusion
- 31:57QA Time
- 32:47Accessibility Overlays
- 34:53Making a strong case for accessibility
- 36:54Competing priorities
- 39:39Motivation factors
- 40:44Final thoughts
related talks
28mYou Don’t Need to Worry About Accessibility with AI (Or Do You?)
05mGitHub Accessibility Scanner
35mAutomated Accessibility Testing
49mHow Do You Make Accessibility Testing as Efficient as Possible
53mAutomating Accessibility Assurance
48mGet It Right the 1st Time: Stop 80%+ of A11Y Bugs During Development
29mProgramming for Accessibility
56mAccessibility Testing 101