Jump to a section
The short answer

Turn off 'Improve the model for everyone' in ChatGPT, 'Keep activity' in Gemini, or 'Help Improve our AI models' in Claude. Use the control in your account settings before starting a new chat. Training, saved history and memory are separate choices, and sending feedback can have different rules. For a one-off conversation, choose a temporary or incognito chat and check how long a copy is kept.

Choose the control that matches your aim

Training is how a company improves its AI using examples. Chat history is the record you can reopen. Memory is information used to personalise later answers. Start with what you want to change.

This guide covers personal consumer accounts. Use the settings available to your account; parental controls can limit some choices. You can start with the free app you already use.

Three different privacy choices
Your aimWhat to use
Stop eligible new chats helping train the modelThe app's model-improvement or activity setting
Have a one-off conversation outside saved historyTemporary Chat in ChatGPT or Gemini; Incognito in Claude
Remove a conversation already in your historyThe app's delete-chat or delete-activity control

ChatGPT: keep your history while opting out

OpenAI's current help calls the switch 'Improve the model for everyone'. Turning it off keeps your normal chat history. When you're signed in, the choice applies across web and mobile; signed-out choices are saved in that browser.

Dan checked the iPhone app on 9 November 2026, when its label read 'Improve model for everyone'. The web and mobile paths below are from OpenAI's current documentation.

Sending thumbs-up or thumbs-down feedback is a separate choice: OpenAI says the associated conversation may be used for training even after you opt out.

  1. Open Settings: use your account menu on the web, or your profile icon in the mobile sidebar.
  2. Choose Data controls.
  3. Turn off 'Improve the model for everyone'; select Done if the web dialog asks you to.
  4. If you use voice, review its separate data-sharing controls too.

Is ChatGPT safe? An honest answer, and five settings to check

Gemini: turn off Keep activity

Gemini links the training choice to 'Keep activity'. With it off, future chats stay out of your activity and aren't used to improve Google's AI models unless you send feedback.

Google still keeps these chats with your account for 72 hours to provide the service and handle safety. Human review can still happen for protection purposes. Some connected features become unavailable.

If you send feedback, Google can include the relevant chat context and uploads. Reviewed feedback and associated conversations can be retained for up to three years, separately from your account.

  1. Open Gemini Apps activity from Activity in Gemini, or use the official activity link in the help below.
  2. Near the top, select On, then Turn off.
  3. Choose 'Turn off and delete activity' only if you also want to remove stored activity; review what will be deleted before confirming.

Claude: change the model-improvement choice

In Claude's personal consumer settings, look for 'Help Improve our AI models'. Turn it off to stop previous and new chats being used in future training runs. This cannot undo training already started or completed.

The training choice does not delete your saved chats. Anthropic normally removes a deleted conversation from its systems within 30 days. That period starts after you delete the conversation, not after you switch training off.

If you allow training, de-identified training data can be kept for up to five years. Anthropic also gives feedback data a separate five-year retention period.

  1. Open your name or profile menu and choose Settings.
  2. Choose Privacy.
  3. Turn off 'Help Improve our AI models'.

For a one-off chat, choose temporary or incognito

Retention means how long a company keeps a copy. These modes keep a conversation out of normal history, but a copy can remain for a time. The table gives the ordinary periods for personal accounts.

In ChatGPT, choose Personalized or Unpersonalized before the first message. Unpersonalized does not use memory, custom instructions or plugins. If you save the temporary chat, it becomes a regular chat and follows your account's training and personalisation settings.

Gemini's temporary mode is for personal accounts in the web and mobile apps. It isn't available in Gemini in Chrome or Google Messages. Claude's incognito mode starts outside projects; in the new Claude experience it opens the previous chat experience.

  • ChatGPT: start a new chat, select Temporary, then choose the personalisation option.
  • Gemini: select the temporary-chat icon next to New chat.
  • Claude: start a chat outside a project and select the ghost icon.
Ordinary retention for temporary consumer chats
ModeSaved history and memoryOrdinary retention
ChatGPT Temporary ChatNot in history unless saved; no new memories while temporaryA copy may be kept for up to 30 days for safety
Gemini temporary chatNot in recent chats or activity; no personalised responses72 hours with your account
Claude IncognitoNot in history or memory; cannot be reopened once closed30 days by default

A practical example: a packing list

Imagine you want a packing list for a rainy weekend. No name, home address or booking reference is needed. This is an illustrative choice, not a test result from a chatbot.

If you want to return to the list, use a normal chat after choosing your training setting. If it's a one-off, use a temporary or incognito chat. Keep a copy of any useful answer before closing a mode you cannot reopen.

  • Want the list to fit what ChatGPT already remembers? Check the personalisation choice before the temporary chat starts.
  • Want a fresh starting point? Choose Unpersonalized in ChatGPT; Gemini temporary chats do not provide personalised responses.
  • Already shared something you want removed? Use the deletion controls separately from the training switch.
A low-detail example to try
Help me make a packing list for a two-night trip in rainy weather. Group it into clothes, toiletries and useful extras. Do not ask for names, addresses or booking references.

Custom instructions and memory: make AI remember what matters to you

What never to tell an AI chatbot (and what's fine)

Before your next chat

Check the setting in the account you actually use. A choice in one company or account does not set the others.

Safety and legal exceptions can affect retention. The linked policies explain those conditions; the periods above are ordinary product rules, not a promise of zero storage.

  • Pick the training setting you want before sending the next message.
  • Choose normal history or a one-off temporary/incognito session.
  • Review the separate feedback choice before sending a rating or report.
  • Remove unnecessary personal details from the prompt.
  • Use deletion controls for saved chats you no longer want, and review memory separately.

AI chatbots and your privacy: what they keep, who sees it, and how to take control

Sources

These sources support the guide. Tool settings, prices and policies change, so dated source boxes show when those details were checked.

AI helps with research and drafting. Dan checks the sources and is responsible for what is published.

How we check