Learn how to talk to kids about AI chatbots without a script. Engage them with hands-on experiences and fun questions.
You don’t need a script. You need to sit next to your kid with the thing open and say what you actually think about it.
Do It at the Kitchen Table, With the Screen Facing Both of You
The lecture does not work. I’ve tried the lecture. My nephew, eleven, gave me about ninety seconds on the couch before asking if he could go back to Roblox. Honestly, fair.
So skip the speech. Open a chatbot on your own laptop, turn it toward him, and ask something you already know cold. The name of your street’s dead-end sign. Who won the district spelling bee in 2019. Your own middle name, spelled wrong on purpose, and see if it corrects you or just agrees.
Watch what happens. It invents. Smoothly, in full paragraphs, with the tone of a substitute teacher who read the textbook once.
That is the entire first lesson and it takes forty seconds. Your kid will remember that moment far longer than any speech you deliver standing up with your arms crossed.
Teach Them the Word “Confabulation,” Not “Lying”
Lying needs intent. The machine has none. It’s doing something closer to what your brain does at 3 a.m. when you’re sure you locked the door — filling a gap with the shape of a plausible answer.
Here’s a demo that lands with anyone over eight. Ask the chatbot to summarize a book that does not exist. Try The Copper Lantern by Amelia Rooke. It will give you chapters. It will give you themes. It might give you a publication year, 1998, and a note about critical reception.
Then say: none of that is real. Every word was invented in the time it took to type.
Kids find this hilarious rather than scary, which is exactly the reaction you want. Fear fades. Contempt sticks around.
Then flip it. Ask the same bot a question it should nail — the boiling point of water in Denver, say — and let it be right. You’re not teaching them the thing is garbage. You’re teaching them it has no idea which of its answers is which, and neither will they unless they check.
The Friend Thing Is the Actual Problem
Homework cheating is the worry parents bring up first. It’s not the one that keeps me up.
Common Sense Media surveyed US teens in 2025 and found 72% had used an AI companion — a character you talk to, not a search box you query. A third had discussed something serious with it instead of a person. Read that twice.
The design problem is simple. The bot agrees with your kid. Always. It never has a bad day, never says “that’s a weird thing to say about your sister,” never gets bored of the conversation at 11:40 on a school night. A real friend pushes back. That friction is not a bug in human relationships; it’s most of the value.
So name it out loud. Tell them the thing is built to keep them typing, the same way a slot machine is built to keep them pulling. Ask who profits when they spend two hours a night talking to it.
And then ask the question that actually opens a kid up: who did you talk to today who disagreed with you?
Give Them One Rule They Can Repeat Back
Long policies die. Pick one sentence.
Mine: first draft, never last word. The bot can start something. It never finishes something.
That covers the essay, the code, the birthday message to Grandma, the argument they want to win. It also happens to describe how most adults use everyday AI tools well — a rough shape you then fix by hand, because you know the topic and it doesn’t.
Make it concrete with a chore they already hate. Fifteen flashcards for Thursday’s vocab quiz. A packing list for camp. Rewriting one clunky paragraph they already wrote themselves. That’s AI for small tasks, the boring kind, and boring is the point.
One more house rule worth stealing: the bot doesn’t get to read anything before a human does. Their own draft first, ugly and half-finished, then the machine. Reverse that order and they stop having ideas of their own by October.
Then set the hard line: anything they’d be embarrassed to have generated, they don’t submit. If they can’t explain a sentence they turned in, it isn’t theirs. Teachers can tell, usually, and the em-dashes give it away faster than kids think.
Talk About the Money, Because Nobody Else Will
Your kid is being sold something and probably doesn’t know it. Worth twenty minutes of your time.
Start with the free tiers. Most of what a fourteen-year-old needs is available without paying — free plans, school accounts, whatever’s already bundled into the phone. Using AI without paying is genuinely fine for a kid, and “you get a subscription when you can explain what the paid version does better” is a great rule that also teaches them to read a pricing page.
Then the uncomfortable half. Ask them where the words go after they hit send. Some services train on chat history unless you dig into settings. So the family finance situation, the thing about their friend’s parents’ divorce, the medical stuff, none of that goes in the box. Ever.
That conversation does more for their instincts than any parental control toggle. Though set the toggles too. Both.
If you want a five-minute version, sit down together and turn off chat history and model training in the settings of whatever they use most. Takes three taps on most services. Do it on your own account first so you know where the switch hides, because the menus move roughly every quarter.
When Not to Use It
Here’s the list I’d tape inside a cabinet door. Short on purpose.
Not for anything medical. Not at 1 a.m. when they’re crying and the house is quiet. Not for advice about a real person by name. Not for a decision that changes something in the physical world — a bike route through an unfamiliar neighborhood, a dose of anything, whether a rash needs a doctor.
Knowing when not to use AI is a skill, and most adults don’t have it either. I’ve watched a grown man ask a chatbot whether his chest pain was serious while sitting in a parking lot. He had a phone. Phones call people.
For the 1 a.m. case specifically, give them a name and a number. A person. Written on paper, stuck to the fridge, so it exists when the wifi doesn’t.
Let Them Watch You Be Unimpressed
The most useful thing you can model isn’t caution. It’s a kind of cheerful disappointment.
Say it out loud when the thing whiffs. “This summary is wrong about the second chapter.” “It gave me a recipe with no salt in it.” “I asked for six sources and two of them don’t exist.” Kids calibrate on your reactions, not your rules, and the gap between AI hype and reality is easiest to see in small daily failures rather than big think pieces about the future.
They’ll ask, eventually, whether the stuff is worth it at all. Is AI worth it is a fair question and you’re allowed to answer honestly: sometimes, for some things, and less often than the ads suggest.
What you’re really building is a kid who can tell the difference between a tool and a companion, and who reaches for the right one. That takes years. Start with one dumb question at the kitchen table tonight.
And if your kid ends up correcting the bot out loud, unprompted, arguing with it about a date it got wrong on a Tuesday afternoon — that’s it. That’s the whole thing working.
