How to stop your chats being used to train AI, on every major assistant

All three companies use consumer chats to train models unless you turn it off. Business and enterprise plans are the exception: those are not trained on by default. If you pay for a personal plan, you are on the consumer side.

Turning the switch off stops future chats going into training. It does not pull back anything already used. It does not stop the company storing your chats, and it does not stop a human from reviewing a chat flagged for safety. And it does not cover feedback: if you press thumbs up or thumbs down, that conversation can be used anyway. So the switch is worth flipping, and it is not privacy on its own.

Open your profile picture, Settings, then Data controls. The switch is Improve the model for everyone. Turn it off. Your chats stay in your history; they just stop going to training. The choice follows your account across devices, so you do it once.

If you use Codex on a personal plan, OpenAI says the same switch covers your Codex tasks. There is also a separate request form at OpenAI's privacy portal called Do not train on my content; either one is enough, you do not need both.

I could not load OpenAI's help page directly from this machine in October 2026, so this comes from the page as quoted by others. The switch name has been stable since it was renamed from "Chat History & Training".

Open your name at the bottom left, then Settings, then Privacy. The switch is Help Improve our AI models. Turn it off. The same path works on the phone app.

Anthropic moved consumer accounts to training-by-default in late 2025, with a longer retention period for people who leave it on. Turning it off applies to new and resumed chats from that moment. Chats you had before and never reopen are not pulled into training after you switch off.

Google's switch is not in the Gemini app. Go to myactivity.google.com/product/gemini and turn off Keep Activity. With it off, your future chats are not saved to your activity and are not used to train Google's models, unless you send feedback. They are kept for 72 hours so Gemini can respond, then deleted.

One detail from Google's own page: chats that a human reviewer has already looked at are kept for up to three years, disconnected from your account, even after you delete them. That is the clearest statement any of the three makes about the limit of the switch.

Reload the settings page. Switches sometimes do not save on a slow connection. Then do the same on the phone app, because some settings are per account and sync, and some, like signed-out browser choices in ChatGPT, are stored only in that browser.

If you have a work account on a Business, Team or Enterprise plan, those are not used for training by the vendors' own terms. Your personal account is the one that needs this.