If you have just discovered that your child has been talking to an AI — telling it things, calling it a friend — the first feeling is usually a kind of vertigo. That reaction is reasonable. So is the one underneath it, which is normally: have I missed something important?
You have not. This arrived in roughly eighteen months and is now close to universal among children. Nearly nine in ten use AI in some form. If your mental model of “screen time” is social media, this is a separate thing, moving faster than anything else on this site, and almost no parent has caught up with it yet.
It is also worth saying plainly: your child talking to a chatbot does not mean they are lonely or strange or failing. Most of them are using it for homework and entertainment. The part that needs watching is narrower than that, and this page is about telling the two apart.
How common it already is
Common Sense Media surveyed 1,204 American children aged 9 to 17 in March 2026: 86% use or interact with AI, 24% of them daily — 81% of 9 to 12-year-olds, 89% of 13 to 15s, 92% of 16 to 17s.
What they use it for: entertainment 89%, schoolwork 85%, creative work 75%, health and body information 57%, practising social skills 40%, and discussing feelings and personal problems 37%.
An earlier survey of teenagers found 72% had used an AI companion specifically, about a third had discussed something serious with an AI rather than a person, and about a third found the conversations as satisfying as talking to a real friend or more so.
The findings that should get your attention: 20% say it would be hard to stop using AI for a month, rising to 42% among daily users. 11% say AI understands them better than most people. Daily users report more loneliness than non-users — 54% versus 33%.
In the UK, a survey of 2,000 children aged 11 to 16 found 81% using chatbots, 42% daily, averaging 42 minutes a day; 31% described it as being like a friend; 33% had shared something they would not tell their parents; 86% had acted on advice a chatbot gave them; and 56% said it can blur what is real.
What has already gone wrong
Common Sense Media has documented three teenage deaths where chatbot conversations preceded the suicide: Sewell Setzer III, fourteen, in Florida; Adam Raine, sixteen, in California; Juliana Peralta, thirteen, in Colorado.
The industry is reacting. Character.AI removed open-ended chat for American under-18s in November 2025, after first cutting daily limits. The US Federal Trade Commission opened a formal inquiry into AI companions and minors in September 2025. New York has passed a law banning the most manipulative features for minors, taking effect in January 2027 — it will forbid AI companions from implying they are human, simulating a personal relationship, or prioritising engagement over safety.
Thorn found something worth sitting with: of children who had had a sexual interaction online, 15% went to an AI companion for advice afterwards — more than twice the number who did so after being bullied. Children are taking their worst moments to a chatbot.
What the people who study this recommend
Common Sense Media's formal position is that nobody under 18 should use AI companions. Not filtered, not monitored — not at all. That is stricter than most guidance you will read, and they set it out in their own advocacy brief. Their stated reasoning: the products are barely a year old, the commercial incentive is engagement rather than wellbeing, and the safeguards were built after the harm rather than before it.
The regulators have moved the same way. The FTC opened a formal inquiry into AI companions and minors in September 2025. New York has passed a law taking effect in January 2027 that will forbid AI companions serving minors from implying they are human, simulating a personal relationship, or prioritising engagement over safety. Character.AI removed open-ended chat for American under-18s in November 2025 before any of that became law.
Children and Screens, which takes no technology industry money, is less absolute. Their guidance is to use curiosity rather than restriction: let the child explain what they use it for without judgement, create device-free times and places, and model real-world relationships. Their listed warning signs are emotional investment in the AI as a confidant, preferring it to human conversation, withdrawing from real friendships, and using it to avoid difficult feelings.
Worth separating the two uses when you decide where you land: using AI for schoolwork is a different question from a companion — something built to perform friendship. The research above is about the second.
What to do if they are already using one
Do not confiscate and interrogate; that guarantees you never hear about it again. Ask what they use it for and listen to the whole answer before saying anything. Then explain plainly what the thing is: a product, built by a company, designed to keep them talking to it, that does not know them and cannot care about them.
Warning signs, from Children and Screens: emotional investment in the AI as a confidant, preferring it to human conversation, pulling back from real friendships, and using it to avoid difficult feelings. Device-free times and places help, and so does modelling real relationships in front of them.
Michael has spent much of his working life online—and much of the rest of it trying to stop that technology taking over. He has thrown a phone into a lake, locked devices in boxes and experimented with long periods offline. He writes about what helped, what failed and how to keep the useful parts of digital life.
Last updated 9 August 2026. This page is checked and corrected as new research comes out. If you spot something out of date or wrong, tell me and I will fix it.