UNDERSTAND AI · FOR TEENS
What should you not tell an AI?
Talking to a chatbot feels private. It’s just you and a screen, no one’s watching, and it answers in that calm, patient voice that makes it easy to say more than you meant to. But it isn’t a diary, and it isn’t a sealed confessional. What you type can be stored, can be read by people at the company, can be used to help train future versions of the AI, and — if a “memory” feature is on — can be carried from one chat into the next. None of that means stop using it. It means there are a few things worth keeping out of the box, the same way you wouldn’t hand a stranger your house keys just because they were nice to you.
Why it feels private when it isn’t
The feeling is real and it’s built in. You’re alone with your phone, the AI never judges, never interrupts, never tells anyone at school. So it’s natural to treat the chat like a journal or a late-night text to a best friend. But the screen is a window, not a wall. On the other side, your messages travel to a company’s computers, get saved there, and can be looked at — sometimes by automated systems, sometimes by actual employees reviewing conversations to check safety or improve the product. “No one’s watching” is a feeling. “This is being stored” is the fact underneath it.
This isn’t a scandal and it doesn’t make the AI sneaky. Storing conversations is how the thing gets built and kept safe. The point is just to match your habits to the reality instead of to the feeling.
The short list: what to keep out
Most of what you do with an AI — homework help, ideas, questions you’re embarrassed to ask a person, drafting a message — is completely fine. This list is small on purpose. A simple test runs under all of it: would I be okay if a stranger, a parent, or a future college or employer read this exact message someday? If the answer is no, keep it out.
- The pieces that identify exactly where you are. Your full name plus your home address plus the name of your school is enough for someone to find you in the real world. Any one alone is usually harmless; the combination is the thing to avoid.
- Passwords, logins, and codes. No password, no PIN, no two-factor code, no “don’t tell anyone but my login is…” A real service will never need you to type these into a chatbot to help you.
- Photos of yourself or other people. Once an image is uploaded it’s out of your hands — you can’t un-send it. Be especially careful with photos of anyone who didn’t agree to be shared.
- Other people’s private business. Your friend’s secret, your family’s situation, a classmate’s name and what they did — that’s not yours to put into a system that stores it. If it’s about someone else, get their say-so or leave it out.
- Secrets you’d hate to see surface later. The deeply personal stuff — the things you wouldn’t want read aloud. Not because the AI will gossip, but because once it’s stored, you’re trusting it won’t ever be exposed, leaked, or read by the wrong person. That’s a bet, and you don’t have to make it.
One more: if you’re in real trouble — being hurt, thinking about hurting yourself, in danger — an AI is not the safe place for that, and not because of privacy. Tell a person you trust, or a real crisis line. A chatbot can sound caring and still not be able to actually help.
The “memory” feature, plainly
Most chatbots used to start every conversation from a blank page — close the chat and it forgot you completely. Many now have an extra feature, usually called memory, and it’s often switched on without you doing anything. When it’s on, the AI quietly writes little notes about you — your name, your interests, things you’ve mentioned — and slips those notes into future chats so it can pick up where you left off. That’s genuinely handy. It also means something you typed once, maybe without thinking, can keep traveling with you from chat to chat.
The good news is you’re allowed to look. In most products you can open Settings and find a section called something like Personalization, Memory, or Data controls, and there you can usually do three things: see the actual list of notes it’s saved about you, delete any note or wipe all of them, and turn the whole feature off if you’d rather it start fresh each time. The exact wording differs by app and changes when they redesign things, so that’s the durable version rather than a menu path that’ll be stale next month. Reading that list once is worth doing — it’s the clearest picture you’ll get of what the AI is keeping.
Go deeper: what “used to train future models” actually means
When a company says your chats may be “used to improve our models,” here’s the rough mechanism. AI models learn from huge piles of example text (more on that on how AI is trained). Some companies add real user conversations to that pile so the next version learns from how people actually talk and what answers helped. Your message usually isn’t stored word-for-word inside the finished model like a file you could open — it gets blended into the statistical patterns the model picks up. But “blended in” is not the same as “guaranteed gone.” Researchers have shown models can sometimes repeat unusual chunks of their training text, and before any training happens your conversation still sits on the company’s servers where it can be reviewed. Two practical notes: many products offer a setting to opt out of having your chats used for training (often the same Data controls menu as memory), and the safest rule doesn’t depend on any of these policies — if it’s something you’d never want repeated, the move is simply not to type it.
Don’t overcorrect
It would be easy to read all that and decide the AI is creepy and you should never tell it anything. That’s an overcorrection, and it’s not the point. These tools are useful, often genuinely fun, and most of what you’ll ever type into one is completely fine to type. The whole job of this page fits on a sticky note: a handful of things stay out, you know the memory switch exists and you’ve looked at it once, and you run the “would I mind if this were read later” test when something feels personal. Do that and you get the good parts without the regret.
The one-line version: a chatbot feels private but isn’t — what you type can be stored, reviewed, used to train future AIs, and (with memory on) carried across chats. Keep out your full identity, passwords, photos, other people’s business, and secrets you’d hate read later. Everything else is fair game.
Where to go next
Does it remember you?
The fuller story on memory — how the feature works, why it’s a useful tool and a double-edged one, and where the switch lives.
How memory works →What one chat can hold
The other kind of memory — the “context window” that holds a single conversation, and why even that has a limit.
The context window →Is an AI your friend?
It listens, it never judges, it’s always there. What that’s actually doing — and where the friend feeling stops being true.
Friend or mirror? →Check your AI
A copy-and-paste prompt that makes your chatbot show you exactly what it’s saved about you, in its own words.
Check your AI →Want the plain-language definitions behind any of this — memory feature, training data, context window? They’re all on the glossary, and the resources page collects the practical guides in one place.