Claude A11y

Formats AI chat responses for screen readers with ARIA landmarks, code block labels, and keyboard navigation. Claude A11y makes AI chat sites usable with screen readers (NVDA, JAWS, VoiceOver, Orca). Works on claude.ai, ChatGPT, Gemini, and Microsoft Copilot.

Without this extension, screen readers announce code blocks as "backtick backtick backtick python print open paren hello close paren backtick backtick backtick." Tables are read as pipes and dashes. There are no landmarks to navigate between responses.

This extension transforms every AI response into accessible HTML:

• Code blocks get announcements: "[Python] print('hello') [End Python]" instead of raw backticks
• Tables are labeled with row and column counts, with proper header associations
• Headings, quotes, bullet points, and separators are announced with clear markers
• Every AI response is wrapped in an ARIA region with a label like "Response 1"
• A live region announces "Generating response..." and "Response complete." so you know when Claude is working
• Alt+Up and Alt+Down navigate between AI responses
• The input area gets a proper ARIA label and role

Toggle between the accessible format and the original output anytime using the popup button.

Built by a developer with Ehlers-Danlos Syndrome who uses screen readers. Open source under MIT license.