How to do it
Privacy with AI doesn't require being a computer expert. It requires three repeatable actions, always the same on every platform.
Turn off training, once per account. It's the switch that decides whether your chats help train future models. On ChatGPT it's in the data controls, on Claude in Privacy, on Gemini in the apps activity. From browser or app it's the same path. You turn it off and forget about it.
Use the temporary chat for delicate topics. It's a mode (an icon at the top, label like "Temporary chat") in which the conversation isn't kept in history and isn't used for training. Think of it as a board that erases itself: you write the confidential things on it and they disappear at the end of the session.
Anonymize before pasting. Where there are names, amounts, addresses, put placeholders: "the client," "the figure," "the city." The AI understands the problem just the same, and you haven't given away anyone's data.
Check what's left, now and then. Open the service's memory (in the settings) and take a look at what it has saved about you. Delete what you don't like.
If you can't find a button because the interface has changed, there's the shortcut that always works: ask the AI itself in the text box. "Where do I find the setting to not use my chats for training?" and it gives you the updated path.
The operational syntax to set up a confidential conversation from the start:
In this conversation I'm dealing with a personal topic.
I don't need you to memorize anything about me.
Work only on what I write here, without asking me for identifying data
and without trying to guess it. Here's the situation: [describe without real names].
A concrete example
Anna, a retiree, wants the AI to explain a letter from the bank she doesn't understand, but it has her IBAN. She knows nothing about computers. She does three things: in the settings she turns off the "Improve the model" switch, opens a temporary chat, and when she pastes the letter she deletes the IBAN and the amount by hand, leaving "my account" and "a certain figure." The AI explains what the bank is asking. No bank data went out, and Anna didn't have to learn anything technical.
When it does NOT work (and how to fix it)
If it all seems too complicated to remember
Four actions are a lot to keep in mind every time. Fix: two you do only once and they stay (turn off training, check the memory). That leaves two gestures per conversation: temporary chat and placeholders. They become automatic in a week.
If you're not sure you've removed all the sensitive data
It happens to forget a detail. Fix: before pressing enter, reread the text once looking for proper names and numbers. If one remains, replace it. For the really delicate cases, make the anonymized version on a sheet of paper first and then paste it.
If the AI keeps reminding you of things you wanted private
It means the memory is active and has saved something. Fix: go into the memory settings, open the list of what it has noted, delete the entries you don't want and, if you prefer, disable the memory entirely.
A tip from someone who actually uses it
Don't think of privacy as a subject to study: think of it as locking your front door. You don't have to know how the lock works, you just have to get into the habit of turning the key. Turn off training once, and for anything personal open the board that erases itself. Two gestures, always the same ones.
Frequently asked questions
Do I have to understand how AI works to protect myself?
No. The protections are switches and habits, not technical concepts. Turning off training, using the temporary chat, putting placeholders: none of these actions requires knowing what a model is or how it's trained. They're done with two fingers.
Is there a single setting that keeps me safe?
No, and it's right to know that so as not to fool yourself. Protection comes from several gestures that cover different risks: training, saving, the data you write. No single switch does everything. But they're few and easy, and together they're enough.
Without technical skills am I more exposed than the experts anyway?
That's the fear that blocks many, and it doesn't hold. The same switches an expert uses, you use too: they're designed for anyone, they're not hidden behind jargon. The difference isn't made by technical skill, it's made by habit: those who turn the key every time are as protected as those who designed the lock.