We’re running into a UX issue related to tool calls and full-screen widgets.
In our app, we display a map in a full-screen widget. While the map is open, the user can continue prompting. When they do, we handle the intent with a tool that updates GPS coordinates (e.g., to change the map location). This tool returns text-only output and is meant purely to update the widget state.
The issue occurs when the user exits full-screen mode. Instead of returning to the widget in context, ChatGPT scrolls to the bottom of the conversation. This happens because the tool call triggers a standard assistant response, which gets appended to the thread.
From a UX perspective, we’d prefer one of the following:
-
A way to prevent ChatGPT from auto-scrolling to the bottom after a tool call in this scenario.
-
A way for a tool to indicate that no assistant message should be generated in response.
-
Some mechanism to mark a tool invocation as “UI-only” or “state update only,” so it doesn’t produce a visible conversational turn.
Is there currently a supported pattern for this? If not, is there a recommended workaround to avoid the scroll behavior when the tool is being used strictly to update widget state?
Any guidance would be appreciated.