Unclear retention settings are a reason to stop before entering a sensitive prompt when the reader cannot establish whether continued storage is acceptable. The help documentation says available data controls depend on whether the user is signed in, the current plan, and workspace settings. It also says that turning off “Improve the model for everyone” does not delete or hide saved chats.
These are separate issues. Disabling model improvement is not evidence that saved chats have been removed, and a control available in one account or workspace should not automatically be assumed to apply in another.
How to check before entering a sensitive prompt
-
Identify the unresolved risk. If the prompt contains information that should not be exposed or retained without approval, uncertainty about storage or deletion is enough to pause.
-
Confirm the current account context. Check whether the user is signed in, which plan applies, and which workspace is active. Available controls can differ across those conditions.
-
Separate model improvement from saved-chat controls. Turning off model improvement neither deletes nor hides saved chats. It should not be treated as a deletion setting.
-
Look for an explicit retention explanation. The reader needs a clear statement of what happens to the relevant chat—not merely a setting that changes model improvement.
-
Stop rather than infer. If the available information does not resolve the storage, retention, or deletion question, the prompt should remain unsubmitted until it can be verified.
What the reader must still confirm
The two documented points do not establish:
- whether a particular prompt will become a saved chat;
- how long any saved chat will remain;
- what deletion action covers;
- when any deletion effect would occur; or
- whether different account or workspace settings change the result.
A control seen elsewhere may provide context, but it is not proof of the behavior in the reader’s current account context.
The practical threshold is straightforward: when a prompt is sensitive and the applicable retention or deletion behavior remains unclear, stopping is justified if uncertainty about continued storage is unacceptable. Waiting for clear, applicable information is more reliable than assuming that a nearby privacy or model setting has the same effect.