Identify the signs your child relies too much on AI tools for homework and critical thinking. Can they work independently?
My bias, up front: I don’t think a chatbot is rotting your kid’s brain. I think it’s quietly doing the boring middle part of thinking. And the boring middle part is where thinking actually happens.
So the question was never whether your kid uses it. They do. Their friends do. Their teacher probably does, at 10pm, writing report-card comments.
The real question is smaller and meaner. Can they work for twenty minutes without it? That’s the test. Everything below is just a different way of running it.
They freeze at the blank page.
Watch the first ninety seconds of homework. Not the whole hour. Ninety seconds.
A kid who’s fine will produce something bad immediately. A clumsy first sentence, a wrong equation, a doodle of their history teacher with a huge nose. A kid who leans too hard on AI sits still, stares, and reaches for the laptop. Not to research. To get a starting point.
That reflex is the tell. Starting badly is a skill, and it’s the first one to go, because “give me three angles for an essay on the Dust Bowl” feels so harmless. It feels like using AI for small tasks. It isn’t a small task. Beginning is the task.
Try this tonight. Two sentences on paper before any screen opens. Bad sentences count. Ugly ones count double. If five minutes pass and the page is still blank, you’ve found your answer, and it isn’t about laziness.
They can’t explain their own homework back to you.
This takes four minutes and it’s devastating. Pick one paragraph from something they handed in. Ask three questions.
What does this word mean. Why is this bit before that bit. What did you cut?
A kid who wrote it defends choices. They’ll say “yeah I know the ending’s rushed, I was tired.” They remember the sentence they were proud of. A kid who didn’t write it will summarize the paragraph back at you in slightly different words, then tell you what it’s “basically about.” Notice the hedging. Notice the shrug.
Do this with math too. Ask them to do the second-to-last step out loud. Not the answer, the step.
The version history says something different than your kid does.
Google Docs keeps everything. File, then Version history, then See version history. You get timestamps down to the minute.
Real writing looks like wreckage. Forty minutes of edits, a deleted paragraph, a comma moved three times, the opening line rewritten at 11:38pm. Pasted writing looks like this: document created 11:04, a 600-word block appears at 11:06, one typo fixed, done.
I’m not telling you to run forensics on a 13-year-old every week. Surveillance just teaches better hiding, and then you’ve lost the thread entirely. But if you’re already lying awake about it, look once. Then talk about what you saw, out loud, instead of collecting evidence for a trial.
They’ve stopped being wrong.
Kids used to arrive at dinner with confidently incorrect facts. Octopuses have nine brains. Napoleon was four foot eleven. The Titanic sank because of a cursed mummy in the hold.
That wrongness is an engine running. A guess is a kid modeling the world with incomplete parts, which is the entire job of being twelve.
Over-reliant kids stop guessing. “I’ll check” replaces “I think.” Ask what they reckon happens if you drop a hammer and a feather on the moon, and they reach for a phone instead of a theory. That shift is subtle and it’s worse than a missed homework. You can fix a missed homework in a night.
Push back gently. Make them commit to a guess before anyone checks. Wrong first, right after.
The panic is louder than the usage.
Here’s the thing about dependence: you see it best when the supply gets cut.
The wifi drops. The free tier taps out mid-essay, because most kids are using AI without paying and the limits land right when the panic does. Watch what happens next. Mild annoyance is normal. A kid who slams a laptop shut and says the assignment is now impossible has told you something about the last six months of their homework.
Same with the sleepover test. Away from home, no account, a worksheet due Monday. Do they just do it badly? Doing it badly is health.
They’re outsourcing the stuff that isn’t schoolwork.
Read the apology text your kid sent their friend after a fight.
Does it use a semicolon? Does it say “I hear you” or “I want to acknowledge how that landed”? Your fourteen-year-old does not talk like a mediation consultant. Nobody’s fourteen-year-old does.
This is the one that worries me most, and it’s nowhere in the school’s AI policy. Rehearsing hard conversations with a bot at 1am feels safe, which is exactly the problem. Awkward apologies are how people learn to apologize. A polished one that isn’t yours teaches nothing except that the machine does feelings better.
Say it plainly to them. Homework help, fine, arguable. Your actual voice to your actual friend, no.
What I’d actually do about it.
Rules beat lectures, and vague rules beat nothing but not by much. So: specifics.
Twenty minutes unplugged first. Timer on the kitchen counter, phone face down. After twenty minutes the tool is allowed, but only as an editor. “Find three weak spots in this draft” is a fine prompt. “Write me an intro” isn’t.
Math goes on paper. Physical paper, visible working, even when it’s slower. Especially when it’s slower.
Read the final thing aloud at the table. Anything that doesn’t sound like your kid gets cut. This catches more than any detector, and school detectors are a coin flip anyway. Detectors have flagged the Declaration of Independence as machine-written. Don’t hand your judgment to software that confident and that wrong.
And do the demo. Ask a chatbot about something you know cold — your street’s one-way system, the rules of the local swim club, your grandmother’s hometown in 1954. Watch it invent an answer with total confidence. Ten minutes of that does more for their understanding of AI hype vs reality than a year of you saying “be careful online.”
The question to keep asking is “is AI worth it here”.
Not in general. On this task, tonight.
Because the honest answer changes. Formatting a bibliography, translating a recipe, turning a photo of a whiteboard into notes — sure, obviously, take the help. Those are everyday AI tools doing dull work, and refusing on principle is just theater.
But there’s a short list of when not to use AI that I’d hang on the fridge if I were less embarrassed about hanging things on fridges. First drafts. Anything you’ll be tested on later. Anything where the point was the struggle. Any message to a person who loves you.
Your kid can learn that list. Mine did, roughly, after a fight about a Steinbeck essay that I handled badly and still think about. The goal isn’t a child who never touches it. It’s a child who notices themselves reaching, and occasionally decides not to.
That noticing is the whole thing. Teach that, and the rest mostly sorts itself out.
