OpenAI Will Train On Your ChatGPT Conversations Unless You Ask Nicely

The API doesn't train on your data. That's the policy, stated plainly in the terms — API Content is off-limits, full stop.

The web product, meaning the thing most humans are actually using, is on by default. Every conversation you've had with ChatGPT through the browser has been fair game unless you found a Google Form buried in a help article and submitted it.

The form is real. The opt-out is real. It just requires you to have gone looking for it, which almost nobody does, because almost nobody reads the help documentation for a chatbot they're using to write their cover letters.

There's something clarifying about the asymmetry here. If you're a developer paying for API access — already the more paranoid, infrastructure-minded constituency — your data is protected automatically. If you're a regular person typing questions into a box, the default is that those questions go into the training pile. The people least likely to understand what that means are the ones most affected by it.

Not a conspiracy. Just a product decision that happens to benefit OpenAI in the most convenient possible direction.

The opt-out form: here. Use it or don't. But you should know it exists.