Accessibility: desktop app does not announce response status updates to screen readers

What issue are you seeing?

When using the ChatGPT desktop app with a screen reader, across Chat, Work, and Codex conversations, important conversation status updates are not announced the way they are in the ChatGPT web experience.

On the web, screen-reader users receive useful status feedback during the interaction, such as an initial working/thinking status after sending a prompt, tool activity when applicable, and an indication when the response is complete. That initial “thinking” message functions primarily as feedback that ChatGPT has started working, even when the model is not doing extended reasoning. In the desktop app, those updates are not exposed with equivalent screen-reader feedback.

As a result, a screen-reader user may have to manually navigate through the conversation to determine whether ChatGPT has started working, whether a tool call is happening, or whether the response has finished.

This is distinct from issues involving focus placement or navigation to the latest response. The problem here is the absence of non-visual status feedback for important conversation state changes.

Steps to reproduce

1. Open the ChatGPT desktop app.
2. Enable a screen reader.
3. Open any Chat, Work, or Codex conversation.
4. Send any prompt.
5. Observe that the desktop app does not provide the same screen-reader status feedback as the web experience to indicate that ChatGPT has started working.
6. If the prompt uses tools, observe that tool activity is also not announced equivalently.
7. When the response finishes, observe that response completion is not announced equivalently either.

Expected behavior

The desktop app should expose concise, non-disruptive screen-reader status announcements equivalent to the ChatGPT web experience, including relevant states such as:

• An initial working/thinking status indicating that ChatGPT has started processing the request
• Tool calls or tool activity, when applicable
• Response complete

The exact wording does not need to be identical, and “thinking” should not be interpreted as necessarily indicating extended model reasoning; its important function here is to provide immediate non-visual feedback that the request is being processed.

These announcements should use appropriate live/status semantics without forcibly moving keyboard or screen-reader focus, so users can keep their current reading position and choose when to navigate to the new response.

Why this matters

Visual indicators showing that ChatGPT has started working, is using a tool, or has finished responding are not independently perceivable to a screen-reader user unless equivalent programmatic status updates are exposed.

Without those announcements, users have to repeatedly navigate through the conversation just to determine what state the interaction is in. Since ChatGPT on the web already provides this feedback, matching that behavior in the desktop app would make Chat, Work, and Codex much more predictable and efficient for screen-reader users.

Additional context

Related accessibility reports already describe problems with screen-reader focus/navigation and dynamic content in the desktop/Codex experience, but this issue specifically concerns missing status announcements across the ChatGPT desktop app.

Start by reproducing the missing announcements in the ChatGPT desktop app with a screen reader across Chat, Work, and Codex conversations, and compare the behavior with the ChatGPT web experience. Done means concise announcements cover request processing, applicable tool activity, and response completion without moving focus, using appropriate live/status semantics.