Five Minutes in the Settings. What Your AI Keeps, and What You Can Switch Off.

Open your AI tool's settings, find the section about privacy or data, and keep it open beside this page.

Every AI tool keeps some of what you type. That includes the symptom you looked up at 2am and the client email you pasted in to make it sound gentler. It includes the message you rewrote four times and never sent.

On a personal account, the company may use some of your chats to improve its AI. That isn't anything sinister in itself. It's a choice, and right now the company is making that choice for you.

Making the choice yourself won't make the AI any worse at helping you. The switches only change what the company does with your words afterwards. You'll know what you're handing over, and you'll keep back the parts you never meant to give.

The switches have different names in Claude and ChatGPT, and they move around, so I'll describe what to look for rather than where to click. There are five switches to look at.

1. Training. Look for a line about improving the model or helping to train it, where the model is the AI itself. While this switch is on, the company can use your chat history to improve the AI. On personal accounts the switch is often on to begin with, while work and school accounts usually have it off and locked. If you're happy for your chats to help, leave it on. If not, turn it off and delete any chats you'd rather not keep. The switch doesn't work backwards, so anything already used stays used. A thumbs up or down on a reply may send that whole chat to the company, whatever the switch says. So don't rate the sensitive ones.
2. Voice and video. Recordings follow the same rules as your chats, but they have a separate switch. If you turned training off, turn this off too.
3. Memory. This is what the tool has learned about you, and every new chat you start can see it. Read it, because it's built from guesses as well as facts. Delete anything out of date, anything about health or money you'd rather it didn't keep, and any guess that's wrong. You can also turn memory off altogether. I've written a whole page on this. The short version is that the memory is yours to edit and worth a proper read every three months.
4. Shared links. Every chat you've shared is a public page that anyone with the link can read, forever. Most people have a few they'd forgotten, including the one with a client's name in it, so delete the ones you no longer need.
5. Connected apps. These are other apps you've linked to your AI tool, and each one can see whatever you gave it access to. So remove anything you tried once and never used again. If you're not sure what an app can do, remove it.

Settings decide what happens to your chats normally. For chats you'd rather left no trace, both tools have a temporary or incognito mode.

A temporary chat isn't saved to your history or used for memory, and in most cases it isn't used for training either.

One tap before you start typing turns it on, and you never have to go back and delete anything. My own rule for this is simple. If I'd hesitate to say something in an open-plan office, it goes in a temporary chat.

That covers the medical question and the message about a family row, and it also covers anything with someone else's private details in it.

Know what the training switch doesn't promise. Your chats are still stored, and most companies also keep some data for a while, for safety and legal reasons.

"Not used for training" and "nobody can see it" are different promises, and one read of your tool's privacy page tells you which one you've got.

No setting protects you from pasting in what you shouldn't, like other people's data, passwords and your card number. For those, you need to pause before you paste.

If a switch isn't where I've described, your tool's help page will show you where it is now.