Your kid has been getting solid marks on homework for the last few weeks. Their essay drafts come back with high scores. Their algebra sets are getting turned in without the usual drama. And then they bomb the test.
If that pattern feels familiar, this is worth ten minutes of your time.
Most middle- and high-school students now have ChatGPT (or Claude, or Gemini, or Copilot — the specific tool doesn't matter) sitting on their phone. Some use it to actually study. Many use it to skip the studying. This post is about how to tell which one is happening in your house, without turning it into an interrogation.
What to look for in the work
The vocabulary doesn't sound like them
Every kid has a writing voice — the sentence length they default to, their favorite phrases, the words they reach for when they're bored. If a paragraph in their essay uses "leverage," "myriad," "furthermore," or "in conclusion" and your kid has never said any of those words at the dinner table, that paragraph probably wasn't written by them. The specific tells that AI leaves behind change every few months as the models change — but a mismatch between the voice in the assignment and the voice you know is durable.
Perfect final answers, no scratch work
Real math solving is messy. Kids cross things out. They start over. They write "= 12 wait no" in the margin. If the problem set has clean, correct final answers with zero visible thinking — and this is a new pattern — the thinking happened somewhere else.
Homework grades and test grades are decoupled
This is the loudest signal. A student who consistently gets A's on assignments and C's on quizzes on the same material isn't struggling with "test anxiety." They can't do the work under their own power. We wrote a separate post about that specific pattern: why homework grades don't predict test grades anymore.
The quality jumped, but their confidence didn't
Kids who genuinely level up in a subject get a little cocky about it. They start telling you what they learned. They correct you at dinner. If the work is getting noticeably better but their vibe hasn't changed — still shrugging, still "I don't know, I just did it" — the improvement isn't coming from them.
What to look for in their behavior
Homework speed changed
If assignments that used to take 90 minutes are now taking 15, something changed. Either they got dramatically better across an entire subject in a few weeks (unlikely), or a machine is doing the work.
They can't explain what they wrote or solved
Ask them to talk you through one paragraph or one problem out loud. Not "explain the concept" — that's too abstract and easy to fake. Ask "why did you subtract 4 there?" or "what were you actually arguing in this sentence?" A student who did their own work can answer, even if fumblingly. A student who pasted it in will scramble, and you'll see it.
Screen behavior changed
Laptop tipped away when you walk in. Tab closed fast. Not proof by itself — teenagers do that anyway — but a change from a few months ago matters.
Two things you can check tonight
You don't need to audit every assignment. Pick one recent one. Then:
- Ask them to redo one problem or rewrite one paragraph, out loud, without their laptop. Don't frame it as a test. Say "I'm curious how you're thinking about this — walk me through one." A student who did their own work can trace the steps, even fumblingly. A student who pasted it in will stall.
- For writing assignments, compare voices. Read one paragraph of the essay next to a text message they sent you last week. Kids code-switch between registers, so allow for some difference. But a Sunday-night text and a Sunday-night essay from the same brain shouldn't sound like they came from different people.
Neither is proof on its own. Either one, clearly positive, is a conversation worth having.
A conversation that won't blow up in your face
Don't lead with "did you use ChatGPT?" You'll get "no," and you'll be stuck there.
Lead with the outcome you actually care about: "I noticed your homework grades and your test grades are pretty different. That tells me something in how you're studying isn't working. What's going on?"
You've reframed it. You're not the enforcer. You're an ally trying to figure out why the plan isn't producing test scores. That framing tends to get an honest answer, because it's about their outcome, not their behavior.
If they cop to it, don't lose it. This is the moment where you decide whether they'll ever tell you the truth again. What they need to hear is: "OK. I get why you did that. It doesn't solve the test-score problem. Let's figure out what does."
What not to do
Banning it doesn't work. They have friends. They have their own phone. Bans just move the behavior somewhere you can't see it.
"Explaining it to them" every night doesn't work either. That's you doing the thinking. It's the same shortcut, just with a parent instead of a chatbot. The student still can't do it alone in a test room.
The real problem is that they need help
Kids don't paste homework into ChatGPT because they're lazy. They do it because the work is hard, they don't know where to start, and it's 10 PM. The chatbot solves the immediate crisis. It also skips the part where they build the skill — and the test always finds out.
If you want to go deeper on the underlying signals of a kid who's actually stuck (as opposed to a kid who just discovered a shortcut), we wrote about that here: signs your kid is struggling in math.
The fix isn't more supervision. The fix is giving them a way to work through problems that produces the same immediate relief — "I'm not stuck anymore" — without doing the thinking for them.
An alternative worth trying
We built Lernos because our own family had exactly this problem. It's an AI tutor with one hard rule: it never gives a student the answer. When a student pastes in a homework problem, the tutor responds with a question that moves them to the next step. If they're stuck, it hints. It scaffolds. It never types out the solution. The parent dashboard shows what they worked on, where they got stuck, and what they figured out on their own — the receipt that a ChatGPT screenshot can never produce.
If any of the signs above described your evening tonight, this is the page to read next: See how it works →