ChatGPT, Claude and Gemini privacy settings: how to stop AI training on your chats
All three major AI chatbots can use your conversations to improve their models, and all three let you switch that off. It takes about two minutes per tool. Below are the steps for each, plus the part most guides skip: what these settings don't protect.
Menus change often. If a label below doesn't match exactly, look for the same words in your settings, or check the official help pages linked at the end.
ChatGPT (OpenAI)
Stop training on your chats
- Click your profile picture, then Settings.
- Open Data controls.
- Turn off Improve the model for everyone.
Good to know
- Temporary Chat doesn't appear in your history, but OpenAI may keep a copy for up to 30 days for safety purposes.
- Feedback counts. Even with training off, giving a thumbs up or down can allow that whole conversation to be used for training.
- Deleting a chat schedules it for deletion within 30 days, unless OpenAI must keep it longer for security or legal reasons.
- You can also opt out through OpenAI's Privacy Portal by choosing Do not train on my content.
Claude (Anthropic)
Stop training on your chats
- Click your profile name, then Settings.
- Open Privacy.
- Turn off Help improve Claude.
Good to know
- Since late 2025, Anthropic uses chats from consumer plans (Free, Pro and Max, including coding sessions) for training unless you turn this off.
- Business, government and education accounts aren't affected.
- If you allow training, Anthropic can keep your chats for up to five years. Older chats are only used if you reopen them after the change.
Gemini (Google)
Stop training on your chats
- Open Gemini and go to Settings & help, then Activity (called Gemini Apps Activity or Keep Activity).
- Turn it off.
Good to know
- With activity off, Google still keeps your conversations for 72 hours to respond and keep the service safe.
- Human reviewers read a subset of conversations. Reviewed conversations are disconnected from your account but kept for up to three years, even if you delete your activity.
- Google's own advice: don't enter information you wouldn't want a reviewer to see.
Side-by-side
| ChatGPT | Claude | Gemini | |
|---|---|---|---|
| Training on by default (consumer) | Yes | Yes (since late 2025) | Yes, while Activity is on |
| Setting to turn off | Improve the model for everyone | Help improve Claude | Gemini Apps Activity |
| Kept after you opt out | Stored in your account until you delete it | Stored in your account until you delete it | 72 hours |
| Human review | For safety and abuse | For safety and abuse | Yes, a subset; kept up to 3 years |
What these settings don't protect
Turning off training is worth doing, but notice what stays the same:
- Your prompt is still sent. The company still receives every name, number and detail you type.
- It's still stored, in your account history, and sometimes for safety review.
- It's still in your account if someone gets access to it.
- Business plans are different, not magic. ChatGPT's business and enterprise plans, Claude's business plans and Gemini in Google Workspace are generally not used for training by default, but your employer can still see and keep what you type.
The only way to make sure a detail isn't stored anywhere is not to send it. That's what anonymizing a prompt does: replace names and numbers with placeholders, get the answer, then put the real details back on your side.
SafePrompt does that automatically in Chrome, on all three of these tools plus Grok, Copilot and DeepSeek. It's a good pair with the settings above: settings limit what happens to your data, and SafePrompt keeps the sensitive part from being sent at all.