Search should be a composer-level toggle, not an inline @mention inserted at the caret

[kraysir]

I’d like to report a UX regression in the current ChatGPT web composer.

When I select Search from the tools menu, ChatGPT inserts an @Search mention directly into the editable prompt at the current caret position.

This produces situations like:

Can you che@Searchck whether this is true?

if I remember to enable Search while the cursor is in the middle of a sentence or even in the middle of a word.

The UI also allows the same Search mention to be inserted multiple times into a single prompt.

Why this is a problem

Search is not semantically part of the user’s sentence. It is a request-level execution/tool setting.

Its exact textual position doesn’t appear to have any meaningful purpose, yet it:

• visually breaks the prompt;

• can be accidentally inserted inside words;

• makes prompts harder to read and edit;

• allows meaningless duplicate Search mentions;

• forces the user to manage an implementation detail that should be handled by the composer UI.

This is especially annoying when I finish writing a prompt and then remember that I want to enable web search.

Suggested behavior

Ideally, Search should be represented as a persistent composer-level chip/toggle outside of the editable prompt, similar to other execution settings such as reasoning/thinking controls.

For example:

[Search ✓] [Thinking: High]

followed by the normal text field.

If Search must remain represented inside the prompt internally, the UI could at least:

1. always normalize it to the beginning or end of the prompt;

2. separate it visually from the user’s text;

3. prevent duplicate Search mentions;

4. never insert it at the current caret position inside the user’s prose.

The previous approach where Search behaved as a separate mode/tool felt significantly cleaner.

This isn’t an issue with having to click Search manually. The problem is specifically that selecting a tool modifies the user’s prose at an arbitrary cursor position.

Please consider treating Search as composer state rather than editable message content.

I can provide a screen recording if useful.

[OpenAI_Support]

Hey! I can see why having Search inserted inside your text is frustrating.

We’ll log your request for a separate Search toggle, along with preventing duplicate mentions and insertion inside words. We can’t promise implementation or a timeline, but thanks for explaining the changes you’d like to see!

-Bhaviyn