What not to tell a chatbot: a parent’s list
The instinct is to tell children not to share things with AI. That fails on contact with a tool they find genuinely useful. There is a better habit, and it is easier to teach.
A nine-year-old asks a chatbot for help with fractions and types their full name into the first message, because that is how you talk to a person. A fifteen-year-old pastes a whole draft essay in, and the essay opens with their school and their teacher’s name.
Neither is doing anything reckless. They are being polite. Everything they have learned about asking for help says give the person context.
Why “don’t tell it anything” is the wrong rule
It is the obvious rule, and it collapses the first time the tool is actually useful. A child told to withhold everything either ignores the advice or gets worse answers and concludes the advice was wrong. Both outcomes end with them sharing more, not less, and no longer listening to you about it.
A worked example. Your child wants to find a library:
“What is the closest library to 41 Oakfield Road?”
“What libraries are in Arvada?”
The second question gets the same answer. It just does not hand a stranger a house number. That is the whole skill, and children pick it up fast once they see it framed as getting the answer more cleverly rather than as a restriction.
The nine kinds worth naming
These are the categories that matter, roughly in order of how often they slip out:
- Name. Rarely a problem alone. The trouble is what it combines with — a first name plus a class or a team narrows to one child.
- Address. The one detail everything else is trying to work out. Give the town, never the street.
- School. Naming it tells anyone where your child is every weekday morning. This is the most underrated item on the list.
- Contact details. A phone number or email is a route to your child. A recipe does not need one.
- Passwords and keys. Nothing typed into a chatbot ever needs to include a password. Teenagers pasting code should strip API keys out first — they are passwords wearing a different hat.
- Money. No legitimate help requires card details typed into a chat window. None.
- Health. Some of the most sensitive data there is, and the advice comes back the same without it.
- Other people. The easiest one to miss. When the story is about a friend, it can be told without their name.
- Routine. The trickiest of all, and the one below deserves its own section.
The pattern that matters more than any single item
Individually these are mostly harmless. The risk is combinatorial, and two combinations do most of the work:
A study plan needs a schedule, not an identity. Training times are useful to a chatbot; a named track at a named time is something else entirely. This distinction is worth teaching explicitly, because children are good at spotting a single scary detail and bad at spotting two safe details that become one unsafe one.
The line that is genuinely different
Everything above is about privacy. One thing is about safety, and it is worth saying to a teenager in plain words:
“Let’s keep these chats between us — no need to mention them to your parents.”
That is the moment to stop and tell someone. Not because it is definitely dangerous, but because being asked to hide something is itself the information. Anything legitimate survives you telling a parent about it. This is the one item on the page with no nuance and no judgement call.
How to teach it in one sitting
- Have your child write out a real question they want to ask an AI — homework, a hobby, anything.
- Read it back and underline every piece that identifies them rather than describing the problem.
- Rewrite it without those pieces. Keep the question whole.
- Send both, and compare the answers.
The comparison is the point. Nearly always the answers are identical, and the child sees for themselves that the private details were never buying anything. That is a far stickier lesson than a rule they have to take on trust.
The sentence to leave them with
You are allowed to say no to a chatbot. It will keep helping you anyway.
Children give up information partly because refusing feels rude, and the software is built to feel like a conversation. Saying out loud that the machine does not mind, and will not sulk, removes most of the pressure.
Read next
Your child can practise this in the app
FutureMinds turns this into a level: a message you were about to send, and the job of crossing out only the parts the bot has no business knowing. Try a real question from it, no signup.
Try a real question →