AI for kids: a parent's guide
AI products for children fall into three groups: general assistants not designed for kids, AI tutors aimed at academic subjects, and purpose-built companions with fixed scope. The safety of any of them comes down to four things — who set the boundaries, whether a parent can see in, how it makes money, and what happens to your child's data.
The three kinds of AI children encounter
- 1General assistantsChatGPT, Gemini, Copilot and similar. Powerful, open-ended, built for adults, and generally 13+ in their terms. Useful for older teenagers with supervision. A poor fit for young children, not because they are lurid but because they have no floor, no shape and no parent view.
- 2AI tutors and homework helpersAimed at maths, reading and revision. Usually well-bounded because the subject matter bounds them. Main things to check are whether it teaches or just supplies answers, and how confidently it states things that are wrong.
- 3Purpose-built companionsA fixed character with a narrow scope — storytelling, conversation practice, social-emotional learning. The safest category structurally, because an adult decided in advance what is in bounds, but quality varies enormously. This is where Wizzly sits.
The four questions worth asking about any of them
- 1Who decided what it will and will not discuss?A product built for children should have topic boundaries set deliberately by adults, and they should hold regardless of how a question is phrased. "It has a safety filter" is not the same thing as "it only does these things".
- 2Can I see what happened?For a child under about eleven, a parent view is not a nice extra. You should be able to see what was discussed, set limits and be told when something worth knowing comes up.
- 3How does it make money?If it is free and advertising-supported, your child's attention and data are the product, and every design decision will follow from that. Subscription or one-off purchase aligns the incentives far better.
- 4What happens to the data — especially voice?Look for a plain statement of what is retained, for how long, and whether it trains anything. Vagueness here is itself an answer.
What AI is genuinely good at for children
- Infinite patience. It will do the same story, the same question and the same explanation for the fortieth time without a flicker — something no tired adult can offer.
- Low-stakes practice. A child can be wrong, hesitate, mumble or take a long time with nobody watching, which is precisely the condition under which shy or anxious children will actually try.
- Responsiveness. Interactive back-and-forth is developmentally more valuable than passive consumption, which is what most children's media offers.
- Adapting to the individual child. Pace, vocabulary and interests, without a class of thirty to average across.
What to be genuinely careful about
- Confident wrongness. These systems state incorrect things in the same tone as correct ones, and children have little defence against a fluent confident voice.
- Agreeableness. Systems tuned to be helpful can end up validating things that need challenging — a distorted account of a fight, an unkind conclusion a child has drawn about themselves.
- Substitution rather than supplement. AI conversation is useful practice; it is not a relationship, and it should not be filling time that human contact would otherwise fill.
- Open-ended scope with young children. The absence of a floor is the core risk, and it is structural rather than something a filter fixes.
- Anything free with no visible business model. The money is coming from somewhere.
Rough age guidance
- Ages 3–6: purpose-built, voice-first, heavily bounded, with an adult nearby. No general assistants.
- Ages 7–10: purpose-built products, plus supervised use of tutoring tools. Start explaining that it can be wrong.
- Ages 11–13: broader tools with supervision, and an explicit conversation about verifying what it tells you.
- Ages 14+: general assistants with the same scrutiny you would apply to any information source — which by this age should be a skill you are actively teaching rather than a rule you are enforcing.
Where Wizzly fits
Wizzly is deliberately in the third category and deliberately narrow. One character — a slightly ridiculous wizard — with a fixed personality, hard content walls that do not move regardless of phrasing, and big life questions handed back to you rather than answered.
It is voice-first and screen-free, for ages 3–15, teaching conversation, emotional regulation and social skills through interactive stories. There is a parent dashboard for topics, time and progress. It carries no advertising and is funded by parent subscriptions — and it is free during early access.
It is not trying to be a general assistant, and for a six-year-old that is the entire point.
Wizzly is completely free during early access — every story, no card required. Runs in the browser — no download.
Common questions
What is the best AI app for kids?
It depends on the job. For academic support, a subject-bounded tutor. For conversation, emotional and social skills in younger children, a purpose-built companion with fixed scope and a parent dashboard — which is what Wizzly is. General assistants like ChatGPT are 13+ and are the wrong shape of tool for young children.
Is AI safe for children?
It depends entirely on whether the specific product was designed for children. Bounded, supervised, purpose-built AI can be genuinely useful. AI built for adults and handed to a young child is where the real risk sits — mostly through open-ended scope and confident errors rather than shocking content.
At what age can a child use AI?
Purpose-built children's products can be appropriate from around three with an adult nearby. General assistants set their minimum at 13, and that is a reasonable floor — younger children cannot supply the scepticism those products assume the user has.
Will AI stop my child learning to think for themselves?
The risk is real where AI supplies answers instead of prompting thinking. It is worth preferring tools that ask your child questions over tools that hand them conclusions — and worth teaching, early, that a confident answer is not automatically a correct one.