Automatic accessibility testing with LLMs
summary
Claims a workflow that pulls the Chrome accessibility tree, extracts button and link text, and sends simple yes/no queries to Chat GPT to flag duplicated or unclear copy. Tests catch all injected errors in trials and produce one false positive (a badly written link that was indeed wrong). Link distinctiveness checks work across sites; image alt detection needs site‑specific tuning and a local model for CI integration is in progress.
AI-written summary · about
chapters (6)
- 0:00Introduction
- 0:57Accessing accessibility data
- 2:02Testing for redundant text
- 2:59Evaluating link accessibility
- 6:06Assessing image alt text
- 7:42Future work and conclusions
related talks
56mAccessibility Testing 101
35mAutomated Accessibility Testing
33mAutomated Accessibility Testing
53mAutomating Accessibility Assurance
30mTesting Web Accessibility
49mAutomated & Manual Testing + AI: A Holistic Approach to Accessibility
31mDecoding web accessibility through testing
21mTesting A11y: Validating Web App Accessibility
