There is no medical diagnosis called AI addiction, and no health body publishes an hours-a-day limit for it. So the honest answer to "how much is too much" is that no clock can give it to you. What can is what the time replaced: sleep, meals, friends, exercise, boredom. And if the AI use is how your child copes with something, the hours are the symptom, not the problem.
Is "addicted" the right word?
It is the word parents type into Google, and it is doing real work. It usually means something like: this has stopped being a choice.
But it is not a diagnosis anyone can make from a usage figure, including you.
The World Health Organization classifies exactly one technology-related addictive behavior in its international disease manual: gaming disorder. Nothing about chatbots, AI companions or homework help appears there. And the criteria for the one that does exist are worth knowing, because they are not about time. WHO describes gaming disorder as "impaired control over gaming, increasing priority given to gaming over other activities" and "continuation or escalation of gaming despite the occurrence of negative consequences," resulting in "significant impairment in personal, family, social, educational, occupational or other important areas of functioning." The pattern "would normally have been evident for at least 12 months."
Read that again and notice what is missing. No hours. No daily cap. Control, priority, harm, and a year of it.
That is my reading of why the hours question feels so unsatisfying to answer: you are asking for a number from a field that deliberately stopped giving numbers. The American Academy of Pediatrics is explicit about it. Its guidance says "there isn't enough evidence demonstrating a benefit from specific screen time limitation guidelines," and recommends "considering the quality of interactions with digital media and not just the quantity, or amount of time." When the AAP updated its media policy for families on June 3, 2026, the stated aim was to "move beyond simple screen-time limits and focus on quality, context and conversation."
None of which helps at 10pm when your twelve-year-old has been talking to something for three hours. So here is what to use instead.
How much are other children actually using it?
Worth knowing before you decide your child is unusual, because most parents are comparing against a guess.
Common Sense Media surveyed 1,204 US children aged 9 to 17 between March 18 and 26, 2026. It found 86 percent had used AI, and 24 percent used it daily. Daily use rose with age: 17 percent of 9 to 12 year olds, 28 percent of 13 to 15 year olds, and 30 percent of 16 and 17 year olds.
So daily use is common and not, by itself, a red flag.
The same survey asked about something more useful than frequency. One in five children said stopping AI for a month would be very or somewhat hard. Among children who used it daily, that rose to 42 percent. Eleven percent of AI users agreed that "AI understands me better than most people do," rising to 19 percent of daily users. Taking three such questions together, Common Sense Media reported that 29 percent of children gave at least one response indicating dependency.
Two cautions on those figures, because they get quoted badly. They are self-reported, and they show daily users reporting more difficulty, which is not the same as daily use causing it. A child who already has something hard going on may well be both the heavier user and the one who finds stopping harder. The direction of that arrow is not established by a survey.
The Displacement Check
The Displacement Check is the question I would ask instead of counting hours: what did this hour replace? Sleep, a meal with other people, a friend, exercise, or the boredom that ideas come out of. Those are the hours worth defending. An hour that replaced a different screen is not the hour to have the argument about.
It works because it maps onto the harms the research actually names. The AAP's 2026 guidance for families points at unhealthy sleep, lack of healthy exercise, and less time with friends and family. Those are displacement harms. They are what going missing looks like.
So on a Tuesday evening, the question is not "was that two hours or three?" It is: did they eat with us, did they go to bed at a normal time, did they see a friend this week, and did they do the thing they used to do?
The clock tells you how long your child was on it. It never tells you what stopped happening while they were.
If nothing is missing, you are probably looking at a hobby that you find strange. That is allowed. If sleep has gone, that is the one to act on first, tonight, and it is also the easiest to fix with a setting.
The Mistake I See Parents Make
The mistake is negotiating the number.
I have spent the past few years training teachers in more than thirty countries, and the same evening scene comes back to me from every one of them. A parent sets a limit of one hour. The child argues it down to ninety minutes. Both sides now believe the argument is about ninety minutes. Nobody has asked what the ninety minutes is for, and so the answer is the same whether the child is building a game, avoiding a math assignment they cannot start, or talking to a character at midnight because the house is quiet and the character is not.
Three identical hours. Three completely different situations. One of them needs a timer, one needs help with math, and one needs you.
The second mistake follows from the first: treating the AI as the thing to remove. Take a chatbot away from a child who is using it to avoid an assignment and you still have a child who cannot start the assignment. That is my view rather than a finding, but it is the pattern I would bet on, and it is the reason I would ask what the hour replaced before I would ask how long it was.
Why some hours matter more than others
The same amount of use means different things for different children, and there is UK evidence putting numbers on it.
Internet Matters surveyed 1,000 UK children aged 9 to 17, along with 2,000 parents, between March and July 2025. Among children using AI chatbots, 35 percent said that chatting with one "feels like talking to a friend," and 15 percent said they would rather talk to an AI chatbot than a real person. One in eight said they used one because they had no one else to speak to.
Among children the report classified as vulnerable, those figures roughly doubled. Half said it felt like talking to a friend. Nearly a quarter said they had no one else to speak to. They were four times more likely than other children to say they used a chatbot because they wanted a friend, 16 percent against 4 percent.
That is the finding I would want every parent to carry, and it is why an hours limit is such a blunt instrument. For one child, two hours with a chatbot is a long evening of nonsense and world-building. For another, the same two hours is the only place they feel heard this week. The clock reads identically. What to do about it does not.
If it looks like the second, the useful thing is not a shorter limit. It is a conversation, and possibly a professional one. A worry about mood, eating, or your child's safety belongs with your family doctor or pediatrician, or with the school's pastoral or safeguarding lead, and none of it is settled by a setting. In the US, the 988 Suicide and Crisis Lifeline is available by call or text. In the UK, Samaritans is on 116 123 and Childline on 0800 1111.
What the controls can actually do about time
Less than parents expect, and the gap is specific.
On ChatGPT, a linked parent account offers quiet hours, which OpenAI describes as setting "times when your teen cannot use ChatGPT, including ChatGPT Work." There is also study hours, an optional schedule that starts eligible new conversations in study mode, and OpenAI is clear that "study hours do not block access to ChatGPT." What is not there: any total daily limit, and any report telling you how long your child spent. Quiet hours are a curfew, not a budget.
Character.AI goes further on visibility. Its Parental Insights feature sends a linked parent a weekly report with the daily average time spent across web and mobile, the characters their teen interacted with most, and the time spent on each one. It also states plainly that the report "does not include a user's chat content," and the teen has to start the process, from the Parental Insights tab in Preferences, by entering a parent's email address.
So one major platform will tell you the number and not the content, the other will tell you neither and let you close the door at bedtime. Neither knows whether your child slept. And none of it reaches a chatbot on a friend's phone, a school device, or a logged-out browser window, which is why the conversation is still the layer that covers everything else. If you want the ChatGPT side set up properly, the full walkthrough of what those controls do and where they stop is worth ten minutes.
What to do this week, by age
Under 8. Use it with them, on your account, in a shared room. At this age the amount is not the question, because they should not be on it alone for any of it. Protect the wondering: let them guess before the machine answers.
8 to 12. This is where the habit forms. Put a real bedtime boundary on the device rather than on the app, because at this age one boundary you actually enforce beats three you do not. Then ask, weekly, without any tone in it, what they used it for and what it got wrong.
You could say: "Show me the best thing you made with it this week."
13 to 15. Stop policing the total and start noticing the timing. Late-night use is the pattern worth acting on, because it eats the sleep that everything else rests on. If a chatbot is the thing they go to when they are upset, that is worth understanding rather than blocking, and what to do when a teenager talks to an AI about their feelings is its own conversation.
You could say: "I'm not asking you to use it less. I am asking you to be asleep by eleven. What do we need to change to make that work?"
16 to 18. Hand the question over and stay in it. A sixteen-year-old can run the Displacement Check themselves, and most will do it more honestly than you expect if it is not an accusation.
You could say: "What has it taken over that you used to do yourself? Anything you miss?"
When it is not about the hours at all
Three signals that the amount is not the real issue, and that the useful next step is a person rather than a limit.
The first is secrecy that is new. A child who has always closed the laptop when you walk in is a teenager. A child who started doing it last month has something specific behind it.
The second is when the AI is doing a job a person used to do. Comfort, company, reassurance at bedtime. That is the situation the Internet Matters figures describe, and it is a signal about the gap, not about the app.
The third is when stopping is genuinely distressing rather than annoying. Irritation at being interrupted is ordinary. Real distress is worth taking to your family doctor or the school, and it is worth doing early, because nobody will think you are overreacting.
I would not diagnose any of these from a screen time chart, and neither should anyone selling you an app that promises to.
What none of this can guarantee
No limit, setting or check makes AI safe for a child, and none of them substitutes for knowing your own child. Platform age rules still apply regardless of anything here: most major chatbots set a minimum age of 13, and school rules on what may be used for schoolwork take precedence over anything a parent decides at home. If you are weighing up a professional's view against something you read here, including this, take the professional's.
And if the honest answer to "what did this hour replace?" turns out to be "the homework they could not start," the amount was never the story. Whether AI is quietly taking the learning out of the work is a different question, with a different answer.
One more thing worth doing this term. Much of what lands at your kitchen table is set somewhere else: homework that assumes a chatbot, devices that come home, a school with a view on AI that nobody has told parents about. If you want to know whether your child's school has actually thought this through, this is the readiness work I do with schools, and it is a fair thing to forward to a head of year. And if you want one short email a week on what is actually changing for families, the newsletter is here.
Dan Fitzpatrick is The AI Educator, a former secondary teacher and assistant headteacher who now helps schools and families make sense of AI. More about Dan.
