Does ChatGPT train on your data?

What OpenAI's policy actually says

OpenAI is explicit about it in their own help documentation: ChatGPT improves by training on the conversations people have with it, unless you opt out. The setting is on by default for consumer accounts — it's opt-out, not opt-in.

This is not a scandal or a secret. It's written down. The problem is that almost nobody reads it, and the default is the one that favours the company.

Which plans train on you?

Notice the pattern: the people who pay the most are the ones whose data isn't used. Consumers pay with data; businesses pay with money.

Can humans read your chats?

Yes. OpenAI states that access to stored conversations is limited to authorized employees and specialized third-party contractors, who may review conversations for engineering support, investigating abuse and misuse, and legal compliance.

That's a narrower group than "anyone at OpenAI," and it's bound by confidentiality obligations. But it isn't nobody — and it is emphatically not "only a machine sees it."

What happens to chats you delete?

Normally, deleting a conversation removes it. But in 2025, during the New York Times' copyright litigation against OpenAI, a court ordered OpenAI to preserve output log data that would otherwise have been deleted — including chats users had deleted themselves. OpenAI publicly objected to the order and challenged it.

The lesson isn't "OpenAI is evil." It's structural: once your words are on someone else's server, deletion is a policy, not a physical fact. A court, a subpoena, or a breach can override it.

How to stop ChatGPT training on your data

1. Open ChatGPT → your profile → Settings.
2. Go to Data Controls.
3. Find "Improve the model for everyone" and switch it off.
4. For one-off sensitive questions, use Temporary Chat — it's excluded from training and from your history.

Opting out stops future conversations being used for training. It does not delete what has already been used, and it doesn't change the fact that your chats are still stored on their servers.

Is this just an OpenAI problem?

No — and it would be dishonest to pretend otherwise. Every major cloud AI (Google Gemini, Anthropic Claude, Meta AI, Copilot) has its own retention and training policies, and each has its own defaults and opt-outs. The specifics differ; the structure doesn't. If the model runs on someone else's computer, your words travel to someone else's computer.

The only way to be certain

Opt-outs are a promise. Local software is a fact. If the AI model runs on your own machine, there is no server to store your chats, no reviewer to read them, no training set to enter, and no court order that can reach them — because they never left.

Several tools do this. Ollama and Jan are excellent free, open-source options if you're comfortable with a bit of setup. GreenCube is the one we make: a one-time purchase, no account required to use it after sign-in, designed for people who just want it to work without configuring anything.