AI & privacy

ChatGPT privacy: what OpenAI keeps, and how to protect yourself

Last updated: November 17, 2026

The short answer

ChatGPT is not private by default. Everything you type, paste or upload is sent to OpenAI, saved to your account, and — on the Free, Plus and Pro plans — may be used to train its models unless you switch that off. Settings can reduce what OpenAI keeps, but they can't stop what you send from reaching OpenAI. The only sure way to keep a detail private is to not send it.

What OpenAI collects when you use ChatGPT

  • What you send: your messages, pasted text, uploaded files, images and voice.
  • Your account: name, email, phone number and payment details if you pay.
  • Your device and usage: things like your IP address, rough location, browser, and how you use the service.
  • Memory: if Memory is on, ChatGPT can save facts about you and use past chats to shape future answers.

Who can see your chats

  • OpenAI: OpenAI says authorised staff and trusted contractors may review conversations, for example to investigate abuse or security issues.
  • Your employer: on Business, Enterprise and Edu accounts, admins may be able to access and export workspace chats.
  • Anyone with a shared link: sharing a chat lets anyone with the link read it.
  • Courts and authorities: like any company, OpenAI can be legally required to keep or hand over data. More on that in does ChatGPT store your data?

5 ChatGPT privacy settings worth changing

  • Stop training on your chats: Settings → Data controls → turn off "Improve the model for everyone". It only affects new chats.
  • Control Memory: Settings → Personalization. Turn Memory off, or review and delete what it has saved.
  • Use Temporary Chat for anything sensitive: it skips your history and Memory, though OpenAI may keep a copy for up to 30 days.
  • Clean up: delete chats and shared links you no longer need. You can also export your data from Data controls to see what's stored.
  • Make a formal request: OpenAI's Privacy Portal lets you ask it not to train on your content, or to delete your account data.

One catch: pressing thumbs up or thumbs down on an answer may let OpenAI use that whole conversation for training, even if you opted out.

Why settings aren't enough

Every setting above controls what happens after your text reaches OpenAI. None of them stop a client's name, a patient's record or an API key from being sent in the first place. For work data, the safer habit is to remove private details before you hit send — which is exactly what Omna does, automatically, on your own computer.

For a team, put the rules in writing with our free AI acceptable use policy template. See also is ChatGPT safe for confidential information? and is ChatGPT HIPAA compliant?

Sources

Note: based on OpenAI's own pages as of November 17, 2026. OpenAI changes its settings and policies often — check its current pages before relying on this. This is general information, not legal advice.