Does ChatGPT train on your prompts?

On a personal account (Free, Plus, or Pro), yes, by default. On a business account or the API, no, by default. You can change the personal setting, and it is worth knowing exactly what it does and what it does not.

Personal accounts vs business accounts

The rule follows the type of account. On a Free, Plus, or Pro plan, OpenAI may use your conversations to help train future models, and that is switched on unless you turn it off. On ChatGPT Team, Enterprise, or Edu, or through the API, your content is not used for training by default. That difference is contractual, not a matter of being careful.

One exception is worth remembering. If you give a thumbs up or thumbs down on a reply, that specific conversation can be shared with OpenAI for review, even on a business plan. Feedback is a deliberate signal, so it is handled differently from an ordinary chat.

How to turn training off

On a personal account, open Settings, go to Data Controls, and turn off "Improve the model for everyone." From then on, new conversations are not used to train the models. It does not touch old conversations that were already used, and a training run that has finished cannot be unwound.

There is also Temporary Chat, which starts a conversation with no history. A temporary chat is not saved to your history, not used to build memory, and not used for training. OpenAI still keeps it for up to 30 days to check for abuse, then deletes it.

Training is not the only thing that happens to your text

Turning off training is not the same as your chats staying private. Those are two different things. With training off, your conversations still go to OpenAI's servers, get stored, and in some cases get looked at. Automated systems scan for abuse, and anything flagged can be routed to a human reviewer. Separately, trainers may read short excerpts to improve the system.

Retention is its own story, and 2025 made that clear. In the copyright case brought by The New York Times, a court ordered OpenAI to preserve chat logs, including ones users had deleted, across consumer and standard API accounts. That order was narrowed in late September 2025, so deleted and temporary chats clear out again after about 30 days. But the logs captured during that window are still held, and in January 2026 the court confirmed that 20 million de-identified conversations must be handed to the plaintiffs. For a while, deleted did not mean gone.

What this means in practice

If you are asking ChatGPT to explain an error or reword an email, none of this is worth losing sleep over. The setting to know about is the training toggle on personal accounts, and turning it off is a reasonable default. But if the text you are about to send holds something you would not want stored, reviewed, or pulled into a lawsuit, the setting does not save you. The only thing that does is not sending it.

Where Riidact fits

That last point is the idea behind Riidact. It watches the box you type in and checks what you are about to send for keys, credentials, personal details, and anything on your own watchlist, then flags it before it leaves. The check runs on your device, so the text being checked never reaches us or anyone else. You can keep the training setting wherever you like. Riidact is about the things that should not be in the message to begin with.

So: personal accounts train on your prompts by default, business and API accounts do not, and you can switch the personal setting off under Data Controls. Just remember what off actually means. It means not used for training. It does not mean never stored and never seen. For anything sensitive, keep it out of the box in the first place.