“How old should my child be before they use ChatGPT?” is the wrong first question. The right first question is: which product, under whose account, doing what? A 10-year-old asking a caregiver-supervised assistant to explain photosynthesis is a different situation from a 10-year-old with an unsupervised account generating images or talking to a companion app at 11pm. Age matters, but it is the second variable, not the first.
This article gives you a working mental model organized around four rough age bands — roughly 6, roughly 10, roughly 13, and roughly 16 — plus the actual minimum ages that AI products enforce today. The bands are a starting point for a conversation with your specific child, not a developmental assessment. Your 8-year-old might read at the roughly-10 level and still need roughly-6 supervision for anything involving strangers or money. Use the matrix to calibrate, not to diagnose.
The common misconception
Parents often assume “age-appropriate AI” means a special kids’ version of a chatbot with the sharp edges filed off. In practice, most general-purpose AI products are not built for children at all, and the ones marketed as safe for kids are usually one of three things: a supervised mode bolted onto an adult product (Google’s Gemini for supervised accounts), an adult product with a teen content filter (OpenAI’s teen experience), or an unrelated product entirely, like an educational app that happens to use AI. None of these give a young child the judgment a chatbot lacks — supervision does that.
The second misconception is that a child who can operate the interface understands what is happening. A 7-year-old can type a question into a chat box before they can reliably tell the difference between a fact and a confident-sounding guess. Interface fluency and epistemic judgment develop on different timelines, and the gap between them is exactly where adult supervision needs to sit.
What the age bands actually track
Each band below tracks six things: the mental model a child at that stage can realistically hold, how much supervision the interaction needs, how privacy should be handled, how claims should be verified, what kind of content creation is reasonable, and what should be off the table entirely.
Read the band that matches your child rather than all four — they are cumulative, so the later ones assume the earlier habits are already in place.
Around 6
| Mental model | ”It’s a machine that guesses words. It can guess wrong.” |
| Supervision | Adult present and driving; child watches and asks |
| Privacy | Adult account only; no child data entered |
| Verification | Adult checks every factual claim aloud |
| Creation | Adult-led only (a story told together) |
| Off the table | No solo account, no images of the child, no chat about people the child knows |
Around 10
| Mental model | ”It learned patterns from lots of text. It doesn’t know things the way a person knows things.” |
| Supervision | Adult nearby, on a shared or supervised device |
| Privacy | No school records, addresses, or photos of other children; first name only if any |
| Verification | Child names one way to check a claim (search, textbook, ask an adult) |
| Creation | Child drafts with AI, adult reviews before sharing anywhere |
| Off the table | No unsupervised account, no unsupervised image generation, no talking to companion-style apps |
Around 13
| Mental model | ”It’s a statistical system with training-data cutoffs and known failure modes I can name.” |
| Supervision | Own account possible where the product allows it (see minimum ages below); check-ins, not shoulder-surfing |
| Privacy | Understands why account data and chat logs are not private conversations |
| Verification | Independently cross-checks anything used for schoolwork or decisions |
| Creation | Independent creative and study use, with disclosure rules understood |
| Off the table | No companion-style “AI friend” products (why); no use for identity documents; no covert surveillance of others; no bypassing school AI-use rules |
Around 16
| Mental model | Can explain hallucination, training bias, and sycophancy in their own words |
| Supervision | Occasional conversation about use, not monitoring |
| Privacy | Manages own privacy settings; understands data retention and deletion |
| Verification | Applies verification habits without prompting |
| Creation | Full independent use within platform terms |
| Off the table | Still no companion-style “AI friend” products; same other legal and safety lines as an adult, plus platform minimum-age terms |
The full side-by-side grid, if you want to compare bands at a glance, is in the parent AI literacy age matrix.
Two things these bands deliberately do not do: they do not tell you your child’s exact number, and they do not replace a product’s own terms. They are a conversation starter you adapt after watching your specific child use a specific tool once, with you next to them.
These bands are starting points, not a clinical or developmental assessment, and they do not override a product’s actual minimum age or your school’s acceptable-use policy. When the two disagree, the stricter rule wins.
The real numbers (verified 2026-07-30)
Developmental readiness is one constraint. Product terms are a separate, harder constraint, and they vary more than most parents expect:
- OpenAI (ChatGPT). The minimum age is 13, or the minimum digital-consent age in your country if higher. Anyone under 18 needs a parent or guardian’s permission to use the service, and OpenAI’s systems try to detect likely-under-18 accounts and apply a “teen experience” with extra content safeguards. Parents and teens can optionally link accounts so a parent can manage some settings and get alerts on serious safety signals — but a linked parent does not see the teen’s actual conversations (OpenAI Terms of Use; OpenAI age prediction).
- Anthropic (Claude). The consumer product requires users to be 18 or older, full stop — there is no supervised or teen mode. Anthropic states this explicitly as a wellbeing decision, not just a legal one, and runs detection systems to catch and disable accounts it believes belong to minors (Anthropic, “Protecting the wellbeing of our users”; Claude age assurance). If your teenager is using Claude directly under their own account, they are outside the product’s intended audience regardless of your household rules.
- Google (Gemini). Children under 13 (or the locally applicable age) can get access only through a parent-enabled Google Family Link supervised account — it is off by default, and a parent has to explicitly switch it on per child. Some features, including voice interaction and activity history, stay restricted even once enabled, and the whole supervised-access path is currently unavailable in the European Economic Area (EEA), the UK, and Switzerland (Google Family Link, Gemini Apps guidance).
None of these numbers is a judgment about whether your child is “ready.” They are the terms you are actually bound by if your child is using an account, and they are worth checking again at the point your child’s age crosses one of these lines, because the products change these policies more often than once a year.
A concrete example: the same question, three ages
Take one question — “why is the sky blue?” — across three children using the same supervised chat session.
A 6-year-old gets the answer read aloud by the adult, who adds: “It made that up from patterns in physics writing it read. Let’s check it matches what we know from school.” The adult supplies the verification; the child supplies the curiosity.
A 10-year-old reads the answer themselves, then is asked: “How would you check if this is actually right?” They suggest looking in a science book or asking their teacher. The adult’s job shifts from providing the check to prompting the child to produce one.
A 13-year-old asks the follow-up question themselves — “does this explanation hold for sunsets too?” — notices the model’s answer gets shakier, and says so unprompted. That noticing is the actual literacy skill this whole framework is building toward: not “can use a chatbot,” but “can tell when a chatbot’s confidence has stopped tracking its correctness.”
Privacy holds regardless of age
One rule does not change across any band: no child’s name tied to a photo, no school ID numbers, no home address, and no identifying details about other children (classmates, friends, siblings) go into a prompt, ever, at any age — because you cannot get that data back out once submitted, and you do not control what it is used for downstream. This is a data-hygiene issue as much as a safety one; see what ChatGPT remembers, sees, and shares for how these systems actually retain input.
Try it today
Pick your child’s closest band above, then run one supervised session where you say the mental-model sentence out loud before opening the app — “remember, it’s a machine that guesses words” or whatever fits their band — and watch whether they can repeat it back in their own words afterward. If they can, they are ready for the next step in the matrix. If they cannot yet, stay at the current supervision level for a few more sessions rather than moving up on a fixed schedule.
The parent AI-literacy age matrix expands every cell in the table above with sample phrases for each age band and a running log for which product minimum ages currently apply in your household. Once your child is ready for a real first conversation rather than a one-off question, a child’s first AI conversation walks through a facilitated script built for exactly that moment. If your child is closer to the top of this range, talking to teenagers about AI covers the two-way conversation that replaces close supervision once a child has an account of their own.
