Yes, chatbots built for children exist, and for a child under 13 one of them is the right place to start. But none of them is safe on its own. A filter narrows what a chatbot will say. It cannot decide what your child opens next, and the chatbots most popular with children were never built for children at all. Set the account up, then ask what they actually used.
What does "kid-safe" actually mean when a company says it?
It usually means four things, and it is worth knowing which four, because none of them is a promise about your child.
A narrower model, so the chatbot refuses more topics. A blocked word list, so certain questions never get an answer. A logged-in account tied to you, so there is a record. And, in the better ones, a design that asks questions back instead of handing over finished work.
That is a real improvement on an open chatbot. It is also a fence around one app. Google says so itself on its own help page for parents: "Our filters try to limit inappropriate content from appearing, but they're not perfect." That is the company that built the thing describing its own product, and it is the most useful sentence on this subject I have read all year.
The three kinds of chatbot your child can actually reach
Parents tend to picture one decision. There are three, and your child can move between them in about ten seconds.
The walled-garden app made for children. Built for an age band, usually paid, usually with some parent visibility. This is the category most parents mean when they search for a kid-safe chatbot.
The big general chatbot with a child or teen setting. ChatGPT's own help pages state that it "is not meant for children under 13" and that 13 to 17 year olds need a parent's permission. Accounts it believes belong to under-18s get extra safeguards, study hours and quiet hours. But OpenAI is equally plain about the limit: "Parental controls do not let a parent or guardian read or monitor the teen's conversations." I have walked through what those controls do, and where they stop, in a separate guide to ChatGPT's parental controls. Google Gemini can be switched on for an under-13 through a supervised Family Link account, and a parent gets an email the first time the child opens it. One detail almost nobody mentions: as of September 2026, Google's own help page says Gemini Apps are not available for supervised accounts in the UK, the European Economic Area or Switzerland. If you are reading this in Britain, that particular route does not exist for you.
The one with no account at all. A browser tab, no sign-up, no age question, no record. This is the category no parental control reaches, because there is nothing to attach a control to.
The research points at the second and third groups, not the first. When Internet Matters studied children's chatbot use in the UK, published on 14 July 2025 with 1,000 children aged 9 to 17, the three products it put through safety testing were ChatGPT, Snapchat's My AI and Character.AI. Those are the ones children were in. Not one of them is a children's product.
Why the safe chatbot is not the one that decides
One thing here matters more than the product, and it is worth knowing before you pay for a subscription.
The Other Chatbot is the one your child opens when the kid-safe one will not do what they want: a story with something frightening in it, a character to talk to, an answer without a lecture. Reaching for it is not a sign of a sneaky child. It is a sign the first chatbot was doing its job. Which means the safety of the whole arrangement is decided by the other chatbot, and the only way you will hear about that one is if asking is ordinary in your house.
You can watch this happen in public. Type "kid friendly ai chatbot" into Google or Bing today and look at what the search box offers to finish the phrase with. Some of the suggestions are a parent shopping: safe, best, for kids. Others are plainly not: free, no sign up, roleplay. Two different people are typing the same three words, and only one of them is going to end up in the app you paid for. That is search suggestions rather than a survey, so read it as a hint and not a finding. But it is a hint you can check yourself in ten seconds, which is more than most of the pages selling you a chatbot will offer.
A kid-safe chatbot is a fence around one app, not a fence around your child. It works for exactly as long as it is the app they are in.
What I see in schools
I sit on the safeguarding side of this conversation more often than the technology side, and the gap between the two is where most of the trouble lives.
The usual sequence in a school goes like this. The school buys a filtered, walled-garden AI tool, often because it was bundled with something else. A letter goes home saying the school now has a safe AI. Staff get an email with a login. And that is the whole rollout.
What a designated safeguarding lead can actually see, in most schools I work with, is narrower than parents imagine. They can usually see the school's own tool: who logged in, and a list of flagged phrases the filter caught. They cannot see the child's phone. They cannot see the logged-out browser tab on a home laptop. They cannot see the chatbot the child reached through a game or a social app. So the record that lands on a safeguarding lead's desk is a record of one tool, in one building, on one network, for six hours a day.
The part that matters in that record is not the blocking. A filter that stops a question has told you something happened, and then done nothing about it. The useful question, and the one I would ask any school, is who reads the flagged list, how quickly, and what they do next. In the schools where this works, a named person looks at it, and somebody has a quiet conversation with the child the same week. In the schools where it does not, the list exists and nobody has opened it since the start of term.
The other pattern I see, and it is nearly universal: the training that comes with the safe tool covers how to use it, not what to do when a child types something worrying into it. Teachers are careful people who will do the right thing on instinct, but instinct is not a procedure. When I train staff on this, the question that comes up in almost every room is some version of "so what actually happens if a child tells it something serious, and do I get told?" That question usually has no answer yet, and the fact that it has no answer is the finding.
None of that makes school AI a bad thing. It makes "the school has a safe AI" a sentence that tells you almost nothing on its own, in exactly the way "we have a kid-safe chatbot at home" tells you almost nothing on its own.
What no filter can guarantee
Three things, and they are the three the sales pages skip.
It cannot judge your particular child. Common Sense Media, in guidance updated on 8 June 2026, found that chatbots consistently miss serious mental health signals, and put it bluntly: "Worse, they often encourage unhealthy thinking and risky behavior instead of pushing back." A filter blocks topics. It does not notice that this child, tonight, is asking in a different voice than usual.
It cannot stop the confiding. Vodafone's UK research with 2,000 children aged 11 to 16, fielded between 30 October 2025 and 2 February 2026 and published on 10 February 2026, found 81 percent using AI chatbots, 31 percent describing them as a friend, and a third saying they had told a chatbot something they would not tell a parent, a teacher or a friend. In the same research, 86 percent said they act on advice a chatbot gives them, while 55 percent said they struggle to judge whether information is accurate or biased. Those two numbers sitting next to each other are the actual risk, and no safety filter addresses either of them. If it is your own teenager doing the confiding, that is its own conversation, and not an emergency.
It cannot replace the conversation. The American Academy of Pediatrics, writing for parents on 27 August 2025, keeps its advice deliberately low-tech: ask your child whether they have used chatbots for fun or friendship and which ones they like, encourage them to use devices in a shared room, tell them that what they type is not guaranteed to stay private, and speak to your child's doctor if they seem withdrawn and would rather talk to a chatbot than a person.
On that last point, plainly: a chatbot is not a clinician, whatever it sounds like. If your child is struggling, the routes that help are your pediatrician or GP, the school's safeguarding lead or counselor, and, when it is urgent, 988 in the United States or Samaritans on 116 123 and Childline on 0800 1111 in the UK. None of those services can promise you what they will or will not pass on, and it is better to ask them than to assume.
What I would do at home, by age
Under 8. No chatbot of their own, and I would not rush this one. Common Sense Media's June 2026 guidance recommends no chatbots at all for children aged 5 and under, because young children cannot reliably tell an AI apart from a person, and adult company alongside them for ages 6 to 12. If a chatbot is going to be in the house, it is on your account, on the family device, with you in the room. Treat it like the oven rather than the television.
8 to 12. These are the years before most children have a phone of their own, roughly Years 4 to 8 in the UK, and the years the United States splits between elementary and middle school. This is where a purpose-built children's chatbot earns its money, and where the account question matters most. Use one, on a supervised account, and pair it with a habit rather than a rule. Once a week, ask them to show you the funniest thing it got wrong. You are teaching them to enjoy catching it out, which is the skill that survives every app they will ever use.
You could say: "You can use it. The deal is you show me anything it says that feels weird, and you don't have to explain why it felt weird."
13 to 15. They are on the general chatbots now, whatever you have installed, and the honest move is to work with that. Set up the parental controls on the account they actually use, and say out loud what the controls do not do, because they will find out anyway and you want to be the one who told them. Then ask about the other one.
You could say: "I'm not going to read your chats, and I couldn't even if I wanted to. So I'm asking instead. What else do you use besides this one?"
16 to 18. Stop managing and start comparing notes. This age is close enough to using AI for work, college applications and their own money that the useful conversation is about judgment, not permission. Ask what it is good at and what it is confidently wrong about, and answer honestly when they ask you the same.
You could say: "What's it been wrong about lately? Tell me, because mine gets things wrong and I do not always catch it."
When somebody else's rule comes first
Three cases where what I have written here does not apply.
Your child's school sets the rule for school devices and school accounts, and a home decision does not override it. If the school blocks a tool, it stays blocked, and the question to take to them is the one above about who reads the flags.
Platform age limits come first as well. ChatGPT's stated minimum is 13, Google's under-13 route runs through Family Link and is not offered in the UK or the EU, and an account set up with a false birthday is a child using an adult product with adult defaults. The limits are self-declared and weakly enforced, which is a reason to take them seriously at home rather than a reason to ignore them.
And if any of this stops being about apps and starts being about your child's wellbeing, that is a conversation for your pediatrician or GP and the school's safeguarding lead, not for a settings page.
The one thing to do tonight
Do not start with the product. Start with the question. Ask your child which chatbot they opened today, and be visibly uninterested in whether it was the approved one. The answer tells you more than any filter will, and asking it tonight makes asking it in three years unremarkable.
If your child's school has sent home a letter saying it now has a safe AI and nothing since, that gap is the work I do with schools: training staff so the policy means something in the classroom. It is worth forwarding. And I write a Sunday newsletter that keeps parents and teachers current on this without the panic, which you can join here.
Dan Fitzpatrick is a former secondary teacher and assistant headteacher, a school trustee, and the founder of The AI Educator. More about Dan.
