Where your words go after you hit send, what "training" really means, the off switches that are already in the menu, and the short list of things to never type.
← back to the front coverHere's the honest version, no scare tactics and no hand-waving: talking to an AI is more like writing a postcard than whispering in a closet. It's usually fine — as long as you don't write your secrets on the postcard.
When you type something into Claude, ChatGPT, or Gemini, your words travel over the internet to the company's computers, which think about them and send an answer back. That part has to happen — it's how the whole thing works. The privacy questions are all about what happens to your words after that: whether they're saved, who might see them, and whether they're used to teach the next version of the machine. Let's take those one at a time, in plain English.
Postcards to the robot are fine. The padlocked envelope stays in your pocket.
By default, most AI services save your conversations to your account — partly so you can scroll back to that recipe from last Tuesday, partly to run their service. Think of it like your email's sent folder: it's there unless you clear it. The company's staff don't sit around reading your chats for fun, but conversations may be looked at in limited ways — for example, to check for abuse or to fix a problem — and they're stored on the company's computers, not just your phone.
The practical takeaway is simple: assume your chats are written down somewhere, the same way you'd assume an email you sent still exists. That's not sinister. It just means you treat the chat box like a postcard, not a diary with a lock.
You'll hear that AI "trains on your data." Here's the plain version. The machine was built by reading an enormous library of text before you ever showed up. Separately, some companies may also use your conversations to help improve future versions — and some don't, or only do it if you say it's okay. The rules differ by company and by whether you're on a free or paid plan, and they change over time.
Two things are true and worth holding onto. First: it's not as if a person is reading your chat and memorizing it — "training" is the machine adjusting itself from huge piles of text. Second: even so, you don't want your private details in that pile, just on principle. Which leads to the one setting worth finding, and the one short list worth memorizing.
Every major service puts the controls right in the Settings menu — usually under a heading like "Data controls," "Privacy," or "Personalization." You don't need to be technical to use them. Here's what to look for:
Can't find these? Ask the AI itself: "Where in your settings do I turn off using my chats for training, and how do I delete my history?" It'll walk you to the exact menu. (Then do it yourself — don't let anyone do it "for" you over the phone.)
This is the heart of the chapter, and it's blessedly short. No matter how convenient it seems, never type any of these into an AI chat:
The trick for all of these is the same and it's easy: ask the question without the secret in it. "Explain this kind of insurance letter" works just as well as pasting the whole letter with your policy number on it. Strip the identifying details, ask the general question, and fill the specifics back in yourself with pen and paper.
Put it all together and the safe way to use AI is short enough to remember: use the real service (the addresses in Chapter Three, typed by you — not a link in an email); turn off training in Settings if you'd rather; clear your history when a chat had personal stuff in it; and above all, keep the short list in your pocket — no SSNs, passwords, full account numbers, or named medical and financial records, ever.
The whole chapter in one line: talk to it like a helpful stranger on the phone. Tell it plenty about your question, and nothing a thief could use about you. Do that and you can use AI all day long without a worry.