AI chatbot privacy settings are not one switch. Training, chat history, temporary conversations and human review are separate controls, and ChatGPT, Claude and Gemini label them differently.
This guide uses the providers’ consumer-account documentation checked on 9 October 2026. Work, school, API and enterprise accounts may follow different data rules.
Before changing anything
Open the official settings page while signed into the account you actually use. A control on a personal account does not necessarily change a team workspace. Export or save conversations you need before deleting history.
Also remember that turning off model improvement does not automatically erase existing chats. Look for a separate delete or retention control.
ChatGPT: Data Controls
OpenAI puts consumer choices under Settings → Data Controls. Its Data Controls FAQ explains the model-improvement setting and Temporary Chat.
Turning off Improve the model for everyone stops new conversations from being used to improve OpenAI’s models, according to the help page. Your ordinary conversations can still remain in chat history unless you delete them.
Use Temporary Chat for a conversation that should not appear in history or create memories. Temporary mode is still subject to OpenAI’s stated safety retention, so it is not a promise that no server-side copy exists.
Claude: model improvement and incognito chats
Anthropic’s consumer control is described in its privacy settings help. The model-improvement choice applies to consumer Claude accounts.
Anthropic says incognito chats are not used for model improvement even if the account-level improvement setting is enabled. That makes incognito useful for an isolated conversation, while the account setting handles the default for ordinary chats.
A private window in your browser is not the same as a Claude incognito chat. The first changes local browser history; the second is a product data mode.
Gemini: Keep Activity and Temporary Chat
Google uses Gemini Apps Activity, also shown as Keep Activity, to control whether conversations are saved to the account activity record. Google’s Gemini privacy help says the default auto-delete period can be changed and Keep Activity can be turned off.
When Keep Activity is off, Google says conversations can still be held for a limited period so the service can respond, maintain safety and process feedback. Temporary Chat gives a clearer one-conversation mode.
Google also warns that feedback may include the associated conversation. Review what is attached before submitting feedback on a sensitive exchange.
What each control changes
| Goal | Look for |
|---|---|
| Stop new chats helping train models | Model improvement or activity setting |
| Keep a chat out of visible history | Temporary or incognito chat |
| Remove an old conversation | Delete chat or activity |
| Clear remembered preferences | Memory or personalisation controls |
| Avoid sharing a sensitive chat with reviewers | Do not submit feedback; use the provider’s private mode |
The labels may move as apps update. Search the provider’s own help centre if a menu name no longer matches.
A safer routine
Do not paste passwords, one-time codes, recovery codes, full card details, confidential client files or health records into a consumer chatbot. Privacy settings reduce some uses of data; they do not turn a public consumer service into your organisation’s approved confidential system.
For occasional sensitive planning, minimise the data first. Replace names with roles, remove account numbers and quote only the passage needed for the question.
If you use several assistants, audit all of them. Changing ChatGPT does nothing to Claude or Gemini. Put a calendar reminder beside major account reviews and record the date you checked each provider.
For product differences beyond privacy, see ChatGPT vs Claude vs Gemini.
