AI Chatbot Privacy Settings: Stop Training, Use Temporary Chats and Delete Your History

To keep your AI chats private, turn off the setting that lets the company use your conversations to improve its models, use a temporary or incognito chat for anything sensitive, and delete history you don’t need. In ChatGPT that’s Settings > Data controls; in Gemini it’s Keep Activity; in Claude it’s Settings > Privacy; in Copilot it’s Privacy > Model training. Meta AI has fewer controls, so the main protection there is what you choose not to type.

Menus change often; if a label doesn’t match, look in the app’s settings for “privacy” or “data”.

Quick reference

ChatGPT

1. Open ChatGPT, click your profile, and go to Settings > Data controls.
2. Turn off Improve the model for everyone. OpenAI says this applies to your whole account on every device. Existing chats stay in your history but won’t be used for training.
3. For a one-off private conversation, start a Temporary Chat from the top of the chat window. OpenAI’s Data Controls FAQ says temporary chats don’t appear in your history, don’t create memories, aren’t used for training, and are deleted from its systems after 30 days.
4. To download everything, choose Export data in the same menu; you’ll get an email with a download link.
5. To clear history, use Delete all chats in Settings, or delete individual chats from the sidebar. Check Settings > Personalization > Memory too, since saved memories are managed separately.

Gemini

1. Go to gemini.google.com (or open the Gemini app and tap your profile picture).
2. Open Settings & help > Activity. The setting is now called Keep Activity (formerly Gemini Apps Activity).
3. Click On, then Turn off or Turn off and delete activity. You can also set an auto-delete period if you’d rather keep recent chats.

According to Google’s Gemini privacy help, chats are still kept for up to 72 hours when the setting is off, and turning it off disables connections to apps like Gmail and Drive. Google also says some conversations may be reviewed by people to improve its services, so don’t enter anything you wouldn’t want a reviewer to see. To export, use Google Takeout and select Gemini or My Activity data; chats you’ve already deleted won’t be included.

Claude

1. Open Claude and go to Settings > Privacy.
2. Find the model-improvement setting (Help improve Claude) and switch it off if you don’t want your chats used for training.
3. For sensitive conversations, start an incognito chat. Anthropic says incognito chats aren’t used to improve its models, even if the training setting is on.
4. To download your data, use Export data on the Privacy page; you’ll get an email link.

Anthropic’s help pages say that if you allow training, your data may be kept in de-identified form for up to five years in its training pipeline; if you don’t, chats are deleted from its back-end systems within about 30 days after you delete them. Changes apply going forward, not to training that has already happened.

Microsoft Copilot

1. On the web, click your profile and choose Privacy. In the Windows or Mac app, go to Settings > Privacy; on mobile, Account > Privacy.
2. Turn off Model training on text and Model training on voice. Microsoft says this can take up to 30 days to take effect.
3. To export or delete Copilot history, use Export or delete history, which opens the Microsoft privacy dashboard for your account.

Microsoft notes that opting out of training doesn’t stop other uses such as safety, security and product improvement. On Windows, see also how to turn off Copilot in Windows 11.

Meta AI

Meta AI has the fewest controls. In most regions, including the US, Meta uses your Meta AI interactions to personalize ads and content, and there’s no general switch to stop that. You can delete your Meta AI chats, and people in the EU and UK can use objection forms in Meta’s Privacy Center. Details are in our guide to what you can turn off in Meta AI.

What never to paste into any chatbot

• Passwords, PINs, verification codes and recovery codes. No legitimate AI task needs them.
• Social Security, passport, bank or card numbers.
• Other people’s private information, including screenshots of their messages, without permission.
• Medical records or test results with identifying details. Remove names and dates of birth if you need help understanding something.
• Confidential work material, unless your employer provides an approved business plan.
• API keys, access tokens and private code secrets.

Even with training off, your chats are stored for some period and may be reviewed for abuse or safety. Treat a chatbot like an email to a company, not a private diary.

Bottom line

Spend five minutes on each assistant you use: turn off training if you prefer, use temporary or incognito chats for anything sensitive, and clear old history. Then keep identifying details out of your prompts. If you’ve ever pasted a password into a chat, change it now and turn on two-factor authentication.

FAQ

Does turning off training delete my chats?

No. It stops future use for training. Delete chats separately if you don’t want them stored in your account.

Are temporary chats completely private?

No. They’re kept out of your history and training, but the company may still hold them for a limited period, such as 30 days for ChatGPT, to monitor for abuse.

Do paid plans have better privacy?

Personal paid plans usually use the same settings as free ones. Business, team and education plans often exclude data from training by default, so check which kind of account you’re on.

Can I get my data back after deleting it?

Generally no. Export your data first if you might want it later.