Is ChatGPT safe for kids? A straight answer by age

What ChatGPT is, in one paragraph

ChatGPT is a chat program made by OpenAI that answers in fluent sentences. You type a question or an instruction, and it writes a reply — an explanation, a story, a piece of code, a summary. It does that by predicting likely next words from patterns in an enormous amount of text it was trained on, which is why it sounds authoritative whether or not it happens to be right. There is no fact-checking step between what it predicts and what you read.

Two things follow from that, and they are the whole reason this page exists. It was built for adults, so its knowledge, tone, and safety tuning all assume an adult reader. And it has no idea whether any particular answer is true — only whether it reads like the kind of thing that is usually said. Neither of those is a scandal. Both of them matter a great deal when the person typing is nine.

Start with what the product itself says

Before any opinion about safety, there's a fact worth knowing: OpenAI's usage terms set a minimum age of 13 for ChatGPT, and require parental permission for users under 18. Most other general-purpose assistants draw a similar line. That isn't a formality — it's the company telling you the product was not designed with a nine-year-old in mind.

This matters because it reframes the question. You're not evaluating a children's product that might have flaws. You're deciding whether to hand a child an adult tool, which is a normal parenting decision you already make about films, phones, and YouTube — with the same range of reasonable answers depending on the child, the supervision, and the purpose.

What actually goes wrong — in rough order of frequency

• It's confidently wrong — By far the most common issue, and the least dramatic. The model produces a fluent, well-structured, incorrect answer — a wrong date, an invented book, a mangled explanation. An adult notices. A ten-year-old with no reason to doubt it usually doesn't.
• Homework does itself — The second most common. It's not that the child sets out to cheat; it's that the path of least resistance now produces a finished-looking answer, and the thinking the assignment was designed to cause quietly doesn't happen.
• Adult-shaped content leaks through — Safety filters are good and not perfect, and they're tuned for adult users. Curious kids probe, and topics a nine-year-old shouldn't meet unaccompanied are reachable — not on every session, but often enough to matter.
• It feels like a friend — The least discussed and the most worth watching. A tireless, agreeable, always-available conversational partner is genuinely appealing to a lonely child. It is not a relationship, and a child at this age can't easily tell the difference.

A straight answer, by age

Ages 7–9: no solo use. At this age children haven't yet built the habit of doubting a confident voice, and open-ended chat gives them nothing to do and nothing to finish. If you want them exploring AI — and there are good reasons to — use a tool built for the age, or sit with them.

Ages 10–12: still under the stated minimum age, and still better served by a purpose-built tool. If you do allow supervised use for a specific task, treat it like the first weeks of a bike on the road: your account, shared space, a defined purpose, and a conversation about what came back.

Ages 13+: now within the terms, and a reasonable time for supervised independence — with the habits already in place. Teenagers who learned to check AI's work at nine handle this transition very differently from those meeting it cold.

Is ChatGPT safe for a 7, 8, 9, 10, 11 or 12 year old?

Parents rarely search by age band; they search by their own child's age. This is the guidance above, resolved age by age. Every row assumes the standard ChatGPT product, not a restricted school deployment.

Safe use of ChatGPT with children: five habits that do the work

1. 1. Use your account, not theirs It keeps the age terms honest, keeps the history visible, and quietly signals that this is a shared tool rather than a private one.
2. 2. Give the session a purpose and an end "Help me understand why volcanoes erupt" ends. "Chat with me" doesn't. The difference in what the child gets out of it is larger than it sounds.
3. 3. Check one thing, every time Pick a single claim from the answer and verify it together. It takes a minute and it teaches the most valuable lesson available here: fluent does not mean true.
4. 4. Never enter personal details No full name, school, address, photos, or friends' names. State it as a flat rule, the same way you'd state it about a stranger asking questions in a park.
5. 5. Ask them to explain it back If they can't put the answer into their own words, they haven't learned it — they've collected it. This is also the fastest way to spot homework that did itself.

The better alternative, honestly stated

For kids, the fix isn't a stricter chatbot; it's a different shape of tool. What a child at this age needs from AI is a project to finish, guardrails on the topic space rather than a filter bolted on afterwards, and an adult who can see what happened without reading every line. An open chat box provides none of those, not because it's badly made, but because it was made for someone else.

That's the design Luchi takes: missions with a beginning and an end, Luchi, whom the child directs and checks rather than just chats with, and a parent update that shows what the child actually practised. Not because chatbots are dangerous in a lurid way — but because "talk to an adult tool for an unbounded amount of time" is a poor default for a nine-year-old, and there's a better one.

Quick answers

What is the minimum age for ChatGPT?

13 under OpenAI's own terms, with parental permission expected for users under 18. Age isn't verified at signup, which is exactly why the decision lands on parents.

My 10-year-old already uses it. Have I done damage?

Almost certainly not. Move it onto your account, into shared space, give sessions a purpose, and start the habit of checking one claim each time. The habits matter far more than the past few weeks.

Is ChatGPT's kids or teen mode enough?

Restricted modes reduce content risk and are worth using, but they don't address the two bigger issues for this age group — dependence and unverified answers. Those are habit problems, not filter problems.

Should I read every conversation?

Not necessary and usually counterproductive. Proximity plus one honest question — "what did you ask it, and did it get anything wrong?" — gets you more than surveillance, and keeps the child telling you things.

What about Claude, Gemini, Copilot and the rest?

The analysis barely changes. The major general-purpose assistants all set a minimum age in their terms — typically 13, sometimes higher without parental consent — because they were built for adults and trained on adult material. They differ in tone, in how cautious their refusals are, and in what they'll do when a user pushes; they do not differ in the thing that matters here, which is who they were designed for.

It's worth being specific about what "safer" means when you read comparisons. One assistant might refuse more topics than another, which reduces content risk. None of them reduce the two risks that actually shape a child: that fluent answers replace the child's own thinking, and that a confident wrong answer goes unchallenged. Those are properties of how the tool is used, not which brand it is — which is why picking the "safest" chatbot is solving the smaller half of the problem.

The same goes for assistants bundled into things you already own — the AI in a search engine, an office suite, a phone keyboard, a school laptop. Your child will meet several of these without ever choosing to, and the habits you build transfer across all of them. That's another argument for teaching the habits early rather than trying to control which product they encounter.

Signs it's going well, and signs it isn't

Going well: your child tells you unprompted when the AI got something wrong. They try a thing before asking. They can explain what they made. They stop when the thing is finished rather than when you say so. They're mildly rude about it — "it's not that clever" is a healthy nine-year-old's attitude to a language model.

Worth a conversation: homework arrives finished but unexplainable. They get defensive when you ask what they asked it. Sessions have no output and no end. They describe it as a friend, or say it understands them. None of these are emergencies; all of them are easier to redirect at nine than at fourteen, which is the entire argument for starting the conversation now.

Why the age limit is 13, and why that number is everywhere

Parents notice that essentially every general-purpose AI product sets its minimum age at 13, and reasonably assume it reflects a judgement about maturity. It does not. It is a legal boundary, and understanding where it comes from tells you more about these products than any safety feature list.

The Children’s Online Privacy Protection Act — 15 U.S.C. 6501–6505, implemented by the FTC at 16 CFR Part 312 — applies to online services directed to children under 13, and to any service with actual knowledge that it is collecting personal information from a child under 13. If you are in that category, you must give parents notice, obtain verifiable parental consent before collecting, allow parents to review and delete what was collected, and allow them to refuse further collection.

Now consider that from the perspective of a company building a general-purpose AI assistant. Building verifiable parental consent, a parental review portal, a deletion path and a revocation mechanism is months of engineering that serves a tiny fraction of your users. Writing "you must be at least 13 to use this service" in the terms takes an afternoon. Almost everyone writes the sentence.

The consequence matters for you. That age line is not a claim that the product becomes appropriate at 13. It is a statement that the company has chosen not to build the under-13 apparatus — which means when your ten-year-old uses it, they are using a product whose maker has explicitly said it is not for them, with none of the parental controls the law would otherwise require. That is a different thing from "risky," and it is more useful to know.

It also explains why "kids mode" on an adult AI product is usually a content filter rather than a consent system. The filter is real work and may help. It does not change the age line in the terms, and it does not give you review, deletion or revocation rights.

What changed in the COPPA Rule for 2026

• The deadline — The FTC published amendments to the COPPA Rule in the Federal Register on 22 April 2025. The amended Rule took effect 23 June 2025, and regulated entities have until 22 April 2026 to comply in full. If you are evaluating an AI product for a child during the 2026–27 school year, you are evaluating it against a stricter standard than existed when most of these products launched.
• Biometric identifiers now count — The amendments expand the definition of personal information to include biometric identifiers. For an AI product this is not abstract — voice input is increasingly standard, and a child’s voice is exactly the kind of data this reaches. If a product listens, the question of what it keeps is now a covered question.
• Third-party sharing needs its own consent — Disclosing a child’s personal information to a third party for purposes not integral to the service now requires separate verifiable parental consent — not bundled into the original sign-up. In practice this means a compliant product has to ask you twice, and a product that never asks you a second time should not be sharing.
• Written security and retention policies — Operators must establish and maintain a written information-security program and a written data-retention policy. This is the most useful change for a parent, because it turns "how long do you keep my child’s conversations" from a question a company can shrug at into a question it must have a documented answer to. Ask it.

School rules and home rules are not the same rules

American parents increasingly get an email from a district about AI, and then apply it at home, or assume it applies at home. Neither is quite right, and being clear about the difference is worth ten minutes.

A district AI policy governs school-issued devices, school-managed accounts, and schoolwork. Most of what such a policy is about is academic integrity — what counts as your own work — and student-records privacy, which in the US runs under FERPA and the contracts the district signs with vendors. When a school adopts a tool, that arrangement does much of the consent work that would otherwise fall to you.

None of that governs your living room. What your child may do on your account, on your device, at home, is your decision as the parent, and the district has no view on it.

The practical move is to read the actual policy document rather than the email summarising it. As of an August 2026 count, 35 states and Puerto Rico have official K–12 AI guidance, and districts have layered their own on top; these documents are generally public and generally more moderate than the announcement. Then be explicit with your child about which rule applies where. A ten-year-old will not draw that line unprompted, and the cost of getting it wrong at school — an integrity violation on an assignment — lands entirely on them.

Questions US parents ask about the age limit

Is it illegal for my under-13 to use ChatGPT?

No. COPPA places obligations on the operator collecting the data, not on you or your child, and it is not a criminal statute aimed at families. What the age line means is that your child is using a service whose terms exclude them and which has not built the parental review, deletion and revocation machinery the law would require for their age group. That is a reason to sit beside them and use your own account, not a reason to panic.

Does a school-approved AI tool mean it is fine at home?

Not automatically. School adoption runs through district contracts and student-records privacy under FERPA, with the school doing much of the consent work in a defined educational context. That tells you the district negotiated terms it found acceptable for classroom use. It does not tell you the product is appropriate unsupervised on a Saturday, and the district is not making that claim.

What should I actually look for instead of an age number?

Four things you can check from outside the product: does it state a specific age range rather than "for kids"; does an adult have to consent before a child account exists; can you see and delete what was collected; and does it say in writing how long it keeps your child’s conversations. Those four map directly onto what COPPA requires of a genuine under-13 service, which is why they are a better filter than any age number in a terms page.

Give them an AI tool built for their age

Luchi is designed for kids from the ground up — guided missions, restricted topics, and parent-visible learning instead of an open chat box.

Get on the list

Back to AI safety

Part of the series: Is AI safe for kids?

Related guides in this series

• 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.
• What is my kid doing on AI? — You can’t read every transcript. Four checks that take under a minute each, what each one tells you, and the one question a ten-year-old actually answers.
• AI apps a parent can actually monitor — Alert, transcript, or record — three different things sold under one word. What seven products say they show a parent, quoted from their own sites.