What is my kid doing on AI? How to actually find out
Your child is using AI — at home, at a friend's house, or on a school device — and you have very little idea what happens in those sessions. Reading the whole transcript is the answer most parents reach for, and it is the one that works worst. Here is what to check instead: four signals that take under a minute each, what each one actually tells you, and the single question that gets a real answer at ten years old.
Why you can't just read it
Start with the honest obstacle, because every plan that ignores it collapses in week two. An AI session is long. A single homework evening can run to thousands of words of back-and-forth, most of it mechanical, and you have perhaps four minutes of attention for it between dinner and bedtime. Surveillance does not fail here for ethical reasons. It fails for arithmetic ones.
And even a parent who reads every word of it ends up with the wrong thing. A transcript tells you what was typed. It does not tell you whether your child understood any of it, whether they tried before they asked, or whether the right answer on the page came from them. Two sessions can read almost identically and mean opposite things. The question "what is my kid doing on AI" is really four smaller questions, and each one has a cheaper signal than the full log.
The four checks, in order of how much they tell you
- 1. Does anything exist that didn't before? A finished worksheet, a working game, a story, a drawing, a bug they fixed. Ten seconds to check, and it separates a session with a shape from a session with none. If nothing is ever produced, you are looking at consumption — and crucially, that is a fact about the tool's design, not your child's character.
- 2. Can they explain one part of it? Pick one thing and ask how it works, or why that answer is right. Thirty seconds. This is the entire quality gate: a child who can explain it learned it, and a child who cannot collected an output. Do it warmly and routinely, or it reads as an accusation and you get a shrug.
- 3. Did they try before they asked? The hardest one to see and the one that matters most, because it is where dependence starts. Ask what they tried first. "I didn't, I just asked it" is not a disaster — it is the default behaviour of every tool that answers instantly, and it is the thing worth agreeing a rule about out loud.
- 4. What does the tool itself show you? The only signal that does not depend on your child's self-report or your own stamina. Most AI products show a parent nothing at all; some show a transcript, which is the raw material rather than the answer. What you want is a record: what they worked on, where they got stuck, how much help they needed, and whether it held a week later.
What "I don't know what they're doing" usually turns out to be
- Copying answers — The worry parents name first, and real — but it is a tool-design outcome more than a character one. A chatbot with no concept of withholding an answer will produce one every time, and a ten-year-old under time pressure takes it. This is the mode to watch for, not to punish.
- Open-ended chatting — More common than copying and far more common than parents expect. Mostly harmless, occasionally not, and the mode with the least to show for itself — no product, no end point, nothing to talk about afterwards.
- Making something — The best case, and the one you hear about unprompted, because children volunteer things they are proud of. If your child shows you work without being asked, that is a strong signal about the mode they are in.
- Stuck, then quietly giving up — The invisible one. The session shows forty minutes of activity and the homework is still not done, because the explanation that came back was pitched at an adult. Nothing in a transcript announces this, and it is the single most useful thing a parent can catch early.
What to check at 7–9 versus 10–12
The four checks stay the same at every age. What changes is who is doing the checking, how much privacy your child is owed, and which failure mode is most likely.
| What you are looking at | Ages 7–9 | Ages 10–12 |
|---|---|---|
| Where it happens | Shared room, your device or your account, every time. Not a rule to negotiate yet. | Shared room is still the single highest-value rule. Their own login is reasonable; a closed bedroom door with an AI is not. |
| What you check | Sit near it. You will overhear most of what you need without asking anything. | Ask rather than watch. At this age reading over the shoulder costs you the honest answers you are trying to get. |
| The likely failure | Treating the AI as a friend or an authority — believing it because it sounds sure. | Quiet dependence: using it for exactly the part of the task that was the point of the task. |
| The question that works | "Show me what you made." Children this age want to demonstrate. | "What did you try before you asked it?" Specific, non-accusatory, and it cannot be answered with "fine". |
| What to agree out loud | Personal details never go in. Simple, absolute, no exceptions. | Where the line is between help and doing it for you — agreed together, before the first argument about it, not during. |
The one question that gets a real answer
"What did you do on the computer?" gets you "nothing." Every parent of a ten-year-old has run that experiment. The reason is that it is a surveillance question, and children are excellent at detecting those and answering them with the minimum legal content.
The version that works is specific and asks them to be the expert: "show me the bit you're proudest of and tell me how it works." It is almost impossible to answer in one word, it flatters rather than interrogates, and it happens to extract exactly the signal you wanted — whether they can explain their own work. Ask it on the same days you ask about school, in the same tone, and it stops being an event. The parents who know most about their children's AI use are rarely the ones with the strictest controls. They are the ones who made the asking boring.
What a tool should be able to tell you
All four checks above are workarounds. They exist because the software your child uses was not built to report to you, so you are reconstructing the session from the outside. It is worth knowing what the alternative looks like, because it is the standard to hold any kid-facing AI product to.
Four things, and they are specific enough to ask a vendor about directly. What did my child work on? Where did they get stuck? How much help did they need to get through it — a nudge, or the whole method? And is the thing they got right on Monday still right on Thursday? A product that can answer those has made your evening question unnecessary. A product that offers you a transcript has handed you the raw material and called it visibility.
This is what Luchi is built to answer, and it is worth being precise about how rather than claiming more than is true. Homework help comes as hints first and a worked example only after two of them, so the thinking stays with your child — full answers are off unless you turn them on. Each question is tagged to one of 1,389 K–8 skills, so "stuck on fractions" resolves to a named skill rather than a mood. Short checks come back days apart, and a skill only counts as mastered after right answers on three separate days. You get a weekly note written from what actually happened, not from minutes logged.
How Luchi teaches — the method, and the research under each part
Common questions
- Should I read my child's AI chat history?
- Spot-checking is reasonable at 7–12, and it is more defensible if your child knows it happens rather than discovering it. But do not build your plan on it. It is long, it is dull, and it answers the wrong question — you learn what was typed, not what was understood. The explain-it-back question takes thirty seconds and tells you more.
- How can I tell if my kid is using AI to cheat on homework?
- The reliable test is not forensic, it is oral: can they defend the work out loud, away from the screen? If they can walk you through one answer and handle a follow-up, the thinking happened somewhere. If the page is right and the explanation is empty, the answer was collected. That is worth a calm conversation about where the line is, not a punishment — the tool offered and they took it.
- Is there an AI app that shows parents what their child did?
- Very few, and the distinction to look for is a record versus a transcript. A transcript is a log of messages. A record tells you which skills were practised, where help was needed and how much, and whether it stuck. Ask any vendor those four questions; most cannot answer them, which is itself informative.
- My child uses AI at school and I can't see any of it. What then?
- Ask the teacher two questions: which tool, and what it is used for. Schools are generally glad to answer, and a district with an AI policy will have it in writing. Then run the explain-it-back check at home on schoolwork, which covers the sessions you cannot see at all.
- My child says "nothing" every time I ask. What do I do?
- Change the question, not the pressure. "Nothing" is the correct answer to a surveillance question. Ask to be shown the part they are proudest of, ask them to explain one piece of it, and ask on ordinary days rather than only when you are worried — a question that only appears when something is wrong teaches them that answering it is risky.
Know what happened, without reading the transcript
Luchi helps with homework through hints, not answers, keeps a record of what your kid knows, and writes you a weekly note from what actually happened. One plan for the whole family, Luchi Family: $5 for your first 7 days, then $99/month ($49/month for founding families). Cancel before the 7 days end and you're not charged the monthly price.
Part of the series: Is AI safe for kids?
Related guides in this series
- Is ChatGPT safe for kids? — A straight answer by age — the risks that actually come up, and the five habits that make AI safe when you do let kids use it.
- Screen time and AI — Not all screen time is equal. Why active, creative AI use is different from passive scrolling — and how to keep your child on the right side of the line.