Here is the sentence most ChatGPT users have never read, even though it applies to them: on personal plans — Free, Plus, and Pro — your chats can be used to train OpenAI's models, and that sharing is enabled by default. I did not know that when I started using ChatGPT either. Most people do not. And once you know it, the next question is obvious: what exactly happens to my chats, and what can I do about it?
I spent an afternoon going through OpenAI's own help pages and privacy controls, and I locked down my account properly. This article is the result: what ChatGPT actually does with your data in late 2026, the three settings that matter, the gaps nobody tells you about, and the five-minute lockdown I now do on every account I touch. No scare stories. Just the facts, and the fixes.
The default you never agreed to
Let us start with the big one. OpenAI says plainly in its Data Controls help page that on personal workspaces with ChatGPT Free, Plus, or Pro, data sharing is enabled by default. That means your conversations can be used to improve OpenAI's models unless you turn it off. Business, Enterprise, Edu, and Healthcare workspaces are different — OpenAI does not use content from those workspaces to train its models by default. The API is also excluded by default.
So if you use a personal account, which is what almost everyone reading this has, the default is: your chats help train the models. That is not a scandal — it is the deal, and it is written down — but it is a deal most people never consciously accepted, because nobody reads the data controls page on day one. I did not. You probably did not either.
The important nuance is that opting out only works going forward. It does not pull back anything already used. Data already absorbed into a trained model cannot be removed, because there is no single copy of your chat sitting inside the model to delete. So the sooner you change the setting, the better — and if you have already pasted something sensitive, changing the setting does not undo it.
Switch one: turn off training on your chats
The first and most important switch is called "Improve the model for everyone." It lives in Settings, under Data Controls, on the web and in the mobile apps. Open your account menu, go to Settings, open Data Controls, find "Improve the model for everyone," turn it off, and confirm. It takes under a minute.
There is a second route that does the same thing: OpenAI's Privacy Portal lets you select "Do not train on my content," and OpenAI says either option is sufficient — you do not need to do both. Pick whichever you find first and move on.
Here is what turning it off actually does: after you opt out, OpenAI will not use your new conversations to improve its models. Here is what it does not do, and this is where people get caught out. It does not delete anything. Your saved chats stay exactly where they are. It does not hide them, it does not archive them, and it does not stop OpenAI keeping them for the normal retention period. It also does not cover everything: voice mode conversations and the thumbs-up or thumbs-down feedback you give can still be used for training even after you opt out. If you tap thumbs-down on a response, that whole conversation may be used to train the model. That is written in OpenAI's own pages, and almost nobody knows it.
Switch two: Temporary Chat for anything sensitive
For a single sensitive conversation, the right tool is Temporary Chat. A temporary chat stays out of your chat history, does not appear in your sidebar, and is not used to improve OpenAI's models while it remains temporary. It is the closest thing ChatGPT has to an incognito mode.
Use it whenever you are pasting something you would not want saved: a personal health question, a financial detail, a work document with private information, anything about someone else. The trade-off is that temporary chats do not get memory or follow-up context, and OpenAI still keeps a copy for up to 30 days for safety reasons. It is not zero-retention. But it is the best option for one-off sensitive questions, and I use it far more often than I used to.
The habit I built is simple: if I pause before pasting something and think "should I?", the answer is Temporary Chat. That pause has saved me from sharing things I would have regretted. Make the pause a rule, not a debate.
Switch three: Locked Chats, and why Memory is separate
Locked Chats are new — OpenAI introduced them in October 2026 — and they solve a different problem: privacy from other people, not from OpenAI. A locked chat stays saved but is hidden from your regular chat history, pinned chats, and conversation search, protected by a PIN. It is for the chats you do not want someone looking over your shoulder, or borrowing your phone, to stumble across.
There is a serious catch: if you reset your PIN, it permanently deletes every locked chat, along with any remembered information stored only within locked chats and files protected only by your old PIN. That cannot be undone. So write the PIN down somewhere safe, or do not lock chats you cannot afford to lose.
Then there is Memory, which is the setting that surprises people most. Memory and model training are separate settings. ChatGPT can remember facts about you — your preferences, your projects, your details — across chats, and deleting all your chats does not delete those memories. They live in a different place. To clear them, go to Settings, then Personalization, and review or wipe Memory there. I check mine monthly and delete anything stale or wrong. And remember: if you paste something sensitive in a normal chat, memory can pick it up, so check what got saved afterward.
What deletion actually means
Deleting a chat removes it from your view right away, and OpenAI schedules it for permanent deletion within 30 days, with exceptions for security and legal reasons. That is the normal window. Archiving a chat, by contrast, only hides it from your sidebar — it stays in your account, and archiving does not change whether it can be used for training.
If you want a clean slate, "Delete all chats" is in Settings. And if you want to leave entirely, deleting your account is permanent, with data deleted within 30 days, subject to limited exceptions. You cannot undo it, and OpenAI Support cannot restore a deleted chat, so export anything you need first. The export option is in the same settings area, and it gives you a copy of your data before you wipe it.
One more honest note: even with training off and chats deleted, OpenAI may keep some data for safety and abuse monitoring, and human reviewers may see flagged content. That is standard for every major AI company. Privacy settings reduce your exposure a lot. They do not make it zero. If something is truly sensitive — medical records, legal documents, other people's private information — the safest choice is not to paste it into any chatbot at all.
The five-minute lockdown
Here is the whole thing as a checklist, in the order I do it. One: Settings, Data Controls, turn off "Improve the model for everyone." Two: Settings, Personalization, review Memory and delete anything you do not want kept. Three: learn where Temporary Chat is, and use it for sensitive questions. Four: if you share a device, set up Locked Chats with a PIN you will not forget. Five: delete old chats you no longer need, and export anything worth keeping first.
That is five minutes, once, and then a monthly thirty-second check of Memory and old chats. It will not make you invisible, and anyone who promises that is selling something. But it takes you from the default — where your chats train models, your memories pile up, and your history sits open — to a setup where you actually chose what happens to your words. That is worth five minutes of anyone's time.
Conclusion
ChatGPT's privacy problem is not that it is secretly stealing your data. It is that the defaults are generous to OpenAI and most users never look at them. Your chats can train models by default on personal plans. Deleting a chat does not delete your memories. Feedback you give can still be used for training. Temporary chats are kept for 30 days. None of this is hidden — it is all in the help pages — but nobody reads help pages until something goes wrong.
So read this instead: turn off training, clean up Memory, use Temporary Chat for the sensitive stuff, and think before you paste. Five minutes now, thirty seconds a month after that, and you are in control of your own words. That is all privacy really is — not secrecy, but choice. And the choice is sitting in your settings right now.
FAQ
Does ChatGPT train on my chats?
On personal plans — Free, Plus, and Pro — yes, by default, unless you turn off "Improve the model for everyone" in Settings, under Data Controls. Business, Enterprise, Edu, and API traffic are not used for training by default.
How do I stop ChatGPT from using my chats for training?
Open your account menu, go to Settings, then Data Controls, and turn off "Improve the model for everyone." You can also choose "Do not train on my content" in OpenAI's Privacy Portal. Either option is enough, and it only stops future use — it does not remove data already used.
Does deleting a chat delete it from OpenAI?
It removes it from your view immediately, and OpenAI schedules it for permanent deletion within 30 days, with exceptions for safety and legal reasons. Deleting chats does not delete your saved Memories — those have to be cleared separately under Personalization.
What is a Temporary Chat?
A chat that stays out of your history and is not used for training while it remains temporary. OpenAI still keeps a copy for up to 30 days for safety. Use it for sensitive one-off questions.
What are Locked Chats?
Chats hidden from your regular history and search, protected by a PIN. They were introduced in October 2026. Warning: resetting your PIN permanently deletes all locked chats and anything remembered only inside them.
Does turning off training delete my chat history?
No. Turning off "Improve the model for everyone" changes future training eligibility only. Your history stays put. To remove chats, delete them individually or use "Delete all chats" — and remember that archiving only hides a chat, it does not delete it or stop it being used for training.
