Stop ChatGPT using new chats for training without losing history
Change ChatGPT’s model-improvement setting, distinguish it from memory and Temporary Chat, and understand what saving or sharing a conversation changes.
In brief
You do not have to delete your entire history to turn off model improvement for new conversations. In a personal account, review the Data controls setting first. Memory, chat retention and data sent to connected services are separate questions.
The useful privacy decision is specific: whether you want a conversation saved, whether you want personalization, and whether eligible content may improve models. One switch does not answer all three.
Keep your history while changing the training choice
On the web, open your account settings and Data controls, then turn off Improve the model for everyone where offered. This controls eligible new conversations; it does not erase saved chats. The signed-in account choice applies across supported web and mobile use.
An ordinary example is a personal writing exercise you want to revisit tomorrow. Turn off model improvement if that is your preference, keep the useful chat, and separately decide whether its content belongs in memory. Do not delete the exercise just to find the training switch.
Documentation for these steps: OpenAI — Data controls ↗
Temporary does not mean every setting is identical
When starting a Temporary Chat, review whether personalization is available and selected. Current documentation distinguishes a personalized temporary conversation from an unpersonalized one. Neither creates new memories while it remains temporary, but existing personalization can still matter.
Temporary chats are not used for model improvement while temporary; a safety copy may be retained for up to 30 days. If you choose to save one, it becomes a regular chat governed by the applicable account settings. Read the save prompt rather than assuming the temporary rules stay attached forever.
Documentation for these steps: OpenAI — Temporary chat ↗
Think about where the information goes
Suppose you are drafting a message from a private document. You can usually explain the writing goal with invented names and a short sample instead of uploading the full record. The best setting is still no substitute for leaving unnecessary personal details out of the prompt.
If the task sends information to another service, review that action before approving it. The receiving service does not inherit every ChatGPT retention or privacy setting. Keep the draft and the external action separate until you know what will be sent.
Documentation for these steps: OpenAI — Temporary chat ↗
A quick check before your next session
Confirm the account, then review model improvement, memory and the chat type as separate controls. If you are on a work or school workspace, ask the administrator which rules apply instead of copying personal-account instructions blindly.
Archiving only changes how a saved chat is organized. Deletion and retention follow their own process. Also read the notice when submitting feedback: opting out of routine training does not mean every voluntary feedback action has the same treatment.
Documentation for these steps: OpenAI — Data controls ↗ · OpenAI — Chat and file retention ↗
Before you act
Controls vary by account and managed-workspace policy. Turning off training is not a promise that nothing is retained or processed. A connected third-party service has its own data practices.
What to do next
- Open Settings → Data controls in the account you intend to use.
- Turn off Improve the model for everyone where the control is available.
- For a temporary conversation, review its personalization choice before entering private context.
- Recheck the implications before saving a temporary chat, submitting feedback or sending data through a connected service.
