3 March 2026 — PageLive Project
Summary
This report documents the difficulties a screen reader user encounters when using claude.ai. Testing was conducted using NVDA on Chrome on a Windows desktop. While the findings are based on NVDA, the barriers described are expected to affect users of other screen readers, including JAWS, in similar ways. While claude.ai is usable with a screen reader, several interactions require significant extra effort or guesswork that sighted users do not face.
1. Scope and Testing Environment
This analysis evaluates the screen reader experience on claude.ai as encountered during manual testing. The scope is limited to interactions that are part of routine use: composing and reading conversations, navigating between chats, and accessing application controls.
Findings are not evaluated against WCAG criteria. The focus is on the practical experience of a screen reader user during everyday tasks.
| Screen Reader | NVDA 2025.3.3.54605 |
| Browser | Chrome 145.0.7632.117 |
| Platform | Windows Desktop |
| Interface Tested | claude.ai web application |
Findings are classified by their impact on task completion:
- Blocker — The task cannot be completed.
- Barrier — The task can be completed, but requires significant additional effort, guesswork, or repeated keystrokes that are not required of sighted users.
- Friction — The task works as expected, but the experience is degraded in a way that compounds over extended use.
2. Findings
2.1 Response Completion Is Not Announced
Severity: Barrier
When claude.ai finishes generating a response, no announcement is made to the screen reader. The response content is not placed in a live region, so the screen reader does not notify the user that new content is available or that generation has completed.
As a result, the user has no reliable way to know when it is appropriate to begin reading. The user must manually navigate to the chat area, attempt to locate the new response, and determine through repeated keystrokes whether generation has finished or is still in progress. This interaction occurs on every single exchange, making it the most consequential barrier in routine use.
2.2 The "Keyboard Shortcuts" Dialog Is Not Accessible
Severity: Blocker
The "Keyboard Shortcuts" dialog in claude.ai presents two separate barriers when accessed with a screen reader.
2.2.1 Dialog appearance is not announced. When the dialog is opened, the screen reader does not announce that a dialog has appeared, and focus is not moved into the dialog. The user receives no indication that anything has changed on screen.
2.2.2 Dialog content is not reachable. When the user navigates to the bottom of the page using the virtual cursor, the screen reader reads only the dialog's heading — "Keyboard Shortcuts" — and does not expose the list of shortcuts contained within. The content of the dialog is effectively inaccessible.
These two barriers combine to make the Keyboard Shortcuts dialog non-functional for screen reader users. The meaningful task — reading the available keyboard shortcuts — cannot be completed. A feature that is specifically intended to support keyboard navigation cannot itself be navigated by keyboard users relying on a screen reader.
2.3 The "More Options" Menu in the Chat List Does Not Announce Its Appearance
Severity: Barrier
Each item in the chat list sidebar contains a "More Options" button. When activated, a popup menu appears containing actions such as Star, Rename, and others. The menu implements a focus trap, meaning keyboard focus is correctly contained within it once the user is inside. However, the appearance of the menu is not announced by the screen reader, and the user is not informed of how to close it.
The result is that after pressing the button, the user receives no confirmation that anything has occurred. The user must discover — through navigation — that a menu has appeared, learn its contents, and determine independently that pressing Escape will close it. A sighted user receives immediate visual confirmation of all three.
2.4 The Sidebar Lacks Sufficient Landmarks for Direct Navigation
Severity: Barrier
Navigation between chats requires access to the chat list in the sidebar. The sidebar contains a landmark that allows the user to jump to its top, but no landmark or heading structure is provided to allow direct navigation to the chat list itself.
Available landmarks during testing were limited to a jump to the top of the sidebar, a banner landmark in the chat section, and a notifications landmark. None of these allow the user to navigate directly to the chat list. A user who wishes to switch chats must traverse through the sidebar from the top, passing over controls unrelated to the chat list on each visit.
2.5 Missing Keyboard Shortcuts for Frequently Used Actions
Severity: Friction
The following actions do not have keyboard shortcuts in claude.ai:
- Jump to the chat input field
- Start a new chat
- Delete the current chat
Each of these actions is available through the interface but requires the user to locate the relevant control through navigation. For a sighted user, these controls are immediately visible. For a screen reader user, the location of a button is not always remembered between sessions, and finding it requires traversal through surrounding content.
The absence of shortcuts does not prevent these actions. However, the cumulative effect of repeated traversal — compounded with the other barriers documented above — meaningfully degrades the experience of extended use.
2.6 Chat List Items Do Not Expose Metadata
Severity: Friction
Chat list items correctly expose the title of each chat to the screen reader. No additional metadata — such as the number of responses in a conversation, the date of the last message, or the model used — is provided.
A user who cannot visually scan the chat list must rely on the title alone to identify and select the correct conversation. Additional metadata would reduce the effort required to distinguish between similarly named chats and to assess whether a conversation is worth reopening.
3. Summary of Findings
| # | Finding | Severity |
|---|---|---|
| 2.1 | Response completion not announced | Barrier |
| 2.2 | Keyboard Shortcuts dialog not accessible | Blocker |
| 2.3 | "More Options" menu appearance not announced | Barrier |
| 2.4 | Sidebar lacks landmarks for chat list navigation | Barrier |
| 2.5 | Missing keyboard shortcuts for common actions | Friction |
| 2.6 | Chat list items lack metadata | Friction |
— End of Article —
Disclaimers
This analysis was conducted as part of the PageLive project, which develops keyboard shortcut and screen reader support for AI chat interfaces. The author develops PageLive, a Chrome extension that addresses several of the barriers described in this report. A companion Screen Reader Experience Mitigation documents the approaches PageLive applies to each finding.
PageLive can be installed for free on the PageLive chrome webstore page.
The findings, severity classifications, and testing observations in this report are the author's own, based on manual testing with NVDA on Chrome. AI assistance was used solely to shape the writing style of this document.
Top comments (0)