Stop AI Chat Training Without Deleting History

To stop future consumer chats from being used for model improvement, turn off Improve the model for everyone in ChatGPT, Help Improve our AI models in Claude, and Keep Activity in Gemini. Check the account you actually use before starting another conversation; a personal setting does not establish the policy for a workplace account.

These controls do different things to saved conversations. ChatGPT and Claude can leave existing chats visible after you opt out, while Gemini’s Keep Activity setting also determines whether future chats appear in Activity. Deleting old records, using a temporary mode, submitting feedback and managing memory are separate decisions.

ChatGPT: switch off model improvement

OpenAI’s ChatGPT data controls state that switching off Improve the model for everyone excludes new conversations from model training but leaves regular chats in history; Temporary Chats may still be retained for up to 30 days for safety.

While signed in on the web, open your account menu, then Settings and Data controls.
Select Improve the model for everyone, turn it off and select Done. In the mobile app, open the sidebar, select your profile icon, then open Data controls to turn it off.
Reopen Data controls and confirm the setting remains off before sending another chat.

The signed-in choice applies across devices. If you use ChatGPT without signing in, open Settings and turn off the same control in each browser where you chat; that choice is stored in the browser. Regular chats remain in history, archived chats remain archived, and project chats remain in their projects. Delete a chat separately if you want it removed from your visible history.

Temporary Chat is a different option for a conversation you also want kept out of history and excluded from model improvement. While it remains temporary, it does not create or update memories; a personalized temporary chat can still use existing memories unless you turn personalization off when starting it. Saving the chat converts it to a regular conversation governed by your account settings. Thumbs-up or thumbs-down feedback is another exception: the associated conversation may be used for training even after the model-improvement control is off.

Claude: change the consumer privacy setting

Anthropic’s consumer privacy update states that turning off model training excludes new and previously stored chats and coding sessions from future training runs, while training already underway or completed cannot be undone; feedback may be retained for up to five years. This policy covers consumer Claude plans and Claude Code used through those accounts, rather than Claude for Work or the API.

In Claude on the web, desktop or mobile, select your name, then open Settings and Privacy.
Under Help Improve our AI models, turn the control off.
Reopen Privacy and confirm it remains off before continuing a chat or starting a coding session.

Turning the control off does not itself erase conversations from your account. If you want a visible conversation removed, use Claude’s conversation deletion control separately. Claude Incognito chats are excluded from model improvement even when the account setting is on, making that mode useful for an individual chat you do not want included.

Feedback and safety review follow their own paths. A thumbs rating can include the related conversation in feedback handling, and a conversation flagged by safety systems may still be analyzed to improve safeguards. Treat the training switch as a choice about ordinary consumer model improvement, not as a promise that no one can review content for another stated purpose.

Gemini: turn off Keep Activity

Google’s Gemini Apps Privacy Hub states that, with Keep Activity off and no feedback submitted, future chats do not appear in Activity or train its AI models; those chats are still retained with the account for 72 hours, and chats already reviewed by people may remain disconnected from the account for up to three years after Activity is deleted.

While signed in to a personal Google Account, open Gemini Apps Activity from Gemini’s settings or menu.
Near the top of the Activity page, open the Keep Activity control and choose Turn off. If you also want to delete existing account-linked Activity, choose Turn off and delete activity and follow the prompts.
Return to Gemini Apps Activity and confirm Keep Activity is off.

Unlike the ChatGPT and Claude switches, this choice also stops new chats from appearing in Gemini Apps Activity. Older Activity needs its own deletion action. Google can still process chats to provide Gemini and protect its services, including with human review for safety. Feedback can bring the associated conversation into review and improvement processes even when Keep Activity is off.

Some Connected Apps become unavailable with Keep Activity off, including Google Workspace connections, while certain Android phone and messaging integrations remain available. The switch does not change Web & App Activity or remove data saved by other Google services or third-party apps. A work or school account may have administrator-controlled settings instead of the personal controls described here.

What each switch changes

ChatGPT — Improve the model for everyone: Excludes new conversations from model training. Regular saved history remains; Temporary Chat is a separate choice for keeping a new chat out of history.
Claude — Help Improve our AI models: Excludes new consumer chats and coding sessions, and previously stored material, from future training runs. Visible conversations require separate deletion; feedback and safety review have separate rules.
Gemini — Keep Activity: Excludes future chats from Activity and, without feedback, model training. Earlier Activity requires a separate deletion choice, while short-term retention and previously reviewed copies can remain.

Memory and personalization govern how a service tailors replies; they are not substitutes for these training settings. A training opt-out does not automatically clear information a service has saved for personalization. Likewise, deleting a conversation from the visible list should not be read as proof that every safety, feedback or connected-service copy has disappeared.

Verify the setting and its visible effect

Before changing a control, note which account or workspace is open, whether the switch is on, and whether older chats appear in history or Activity. After changing it, reopen the same setting and confirm the off state. This checks the choice you made without relying on a brief confirmation message.

If you want to check the history effect, use a nonsensitive test conversation. A regular ChatGPT or Claude chat may still appear in history after its training switch is off; a new Gemini chat with Keep Activity off should stay out of Activity. The visible check cannot inspect a provider’s internal training pipeline, but it can catch a mistaken account, an unchanged setting or an assumption that opting out deleted earlier chats.