Screen reader: no heading structure in conversation view (NVDA)
Resolved 💬 0 comments Opened Jun 22, 2026 by JoshMiele Closed Jun 25, 2026
Part of umbrella accessibility issue: #69996
Problem
The conversation view does not use heading elements to mark the boundaries between user prompts and assistant responses. Screen reader users rely on heading navigation (H key in NVDA) to move quickly through long documents. Without headings, the only way to jump to the most recent prompt or response is to use text search — for example, searching backward for the string "branch from here," which happens to appear reliably after each user prompt.
Recommendation
Each assistant response should follow this heading structure:
- H1 — The first sentence of the response, marked as H1. This gives screen reader users a reliable top-of-response landmark reachable with a single keypress.
- H2 and below — Used as appropriate within the response to mark major sections and subsections.
- H5 — The final line of every response should be the string "End of response" wrapped in H5, providing a reliable bottom-of-response landmark confirming generation is complete.
User prompts should similarly be preceded by a heading element (H2 recommended) so users can jump backward through conversation history to review their own input.
Environment
- Claude Code desktop app, Windows 11
- NVDA screen reader (latest stable)