Continuous AI for accessibility: How GitHub transforms feedback into inclusion

GitHub built an internal AI-powered workflow using GitHub Actions, GitHub Copilot, and GitHub Models to transform how accessibility feedback is triaged and resolved. The system captures feedback from any source, uses Copilot to auto-populate ~80% of issue metadata including WCAG mapping, severity scoring, and affected user groups, then routes issues through submitter and accessibility team review before closing the loop with the original reporter. Results after one year: 89% of issues now close within 90 days (up from 21%), average resolution time dropped 62% from 118 to 45 days, and a 70% reduction in manual administrative work. The workflow is event-driven, continuously improves via prompt file updates through pull requests, and keeps humans in control of final decisions while AI handles repetitive triage work.