Last updated October 2, 2026: every statistic re-checked against its primary source. Corrects what the Common Sense Media census says about how much this age group has been taught, removes one figure that could not be confirmed, and adds what now happens when a child turns 13.
For most 9 to 12 year olds the answer is yes, with you, on your account or the school's, not on one of their own. ChatGPT's rule is 13 and up. Google allows Gemini under 13 only through a supervised Family Link account. Tonight, check whose account AI is already signed in on. No setting can promise your child will not see something you would rather they had not.
Which AI, at what age, and on whose account
Three rules cover almost everything a 9, 10, 11 or 12 year old is likely to open. All three were read on the companies' own pages on October 2, 2026.
- ChatGPT. OpenAI's help center says ChatGPT "is not meant for children under 13", and that "we require that children ages 13 to 18 obtain parental consent before using ChatGPT". Consent, not permission: the company's own word. At 9 to 12, that means your account with you in the room, or not at all.
- Gemini. Google allows an under-13 to use Gemini through a supervised Family Link account, but its guidance for families says Gemini Apps "are not available for supervised accounts in the European Economic Area, Switzerland, and the United Kingdom". So for a UK family the supervised route is closed, and the school account is the one to ask about.
- Character.AI. Character.AI's own help center says under-18 users no longer get open-ended chat with characters, a change that began on November 24, 2025, though they can still make characters, scenes, videos and voices. The page was last updated on October 29, 2025. It is not a place for a 10 year old.
Every one of those is a stated rule rather than a lock, because the age is whatever the child types into the birthday box. What the age limits do and do not enforce is set out in what the AI age limits actually mean.
What to set up tonight, in about ten minutes
- Find out which account it runs on. Pick up the device your child uses most and look at whether a chatbot is signed in, and as whom. Four minutes of that tells you more than a week of reading.
- On Android or a Chromebook, set Gemini deliberately rather than by default. In the Family Link app or at familylink.google.com, select your child, then Controls, then Gemini, then Gemini Apps, and turn it on or off on purpose. On some versions the path runs through Controls, then Content restrictions, then Gemini. Google's page adds that the Keep Activity setting is not available to users under 13, so an under-13 supervised account is not building a stored history in the way an adult account does.
- If an older sibling is on ChatGPT, link the accounts. A parent links a teen from Settings, then Parental controls, then Add family member. That system is built for 13 and up, so at 9 to 12 it is the sibling's account you are setting, not your fourth grader's.
Then ask the school the question in the last section of this article, because the school account is the one door you cannot set from the kitchen.
That is the short version. The longer version matters, because at this age most parents brace for the wrong thing.
What 9 to 12 year olds are actually doing with AI
Mostly, they are playing with it.
Common Sense Media's census of AI use, published on June 8, 2026, surveyed 1,204 children aged 9 to 17 in the United States between March 18 and 26, 2026, and breaks its results down by age. Among 9 to 12 year olds, 81 percent had used AI and 17 percent used it every day. What they used it for is the part worth sitting with: 91 percent for entertainment or fun, and 74 percent to make images, videos, stories or music.
Fun comes first in this band. By 16 and 17 the order has reversed: 94 percent for schoolwork against 91 percent for fun. For the wider picture of how many children use AI at all, in the United States and the UK, see how many kids use AI.
In the UK the picture rhymes. Ofcom's Children and Parents report, published on May 21, 2026 from interviews with 3,426 children aged 8 to 17, found that 56 percent of 8 to 17 year olds had used AI, and that 35 percent of 8 to 9 year olds had. Across that whole 8 to 17 sample, 43 percent of the children who use AI say they use it simply for fun, and 20 percent use it to make images or songs. That 8 to 9 band is the youngest Ofcom breaks out, and below it the question changes shape: for a five or six year old the argument at home is not about boundaries at all, it is about whether AI can teach a child to read.
Here is what I make of that. The homework argument, the one that fills the parenting articles and the staff room, is a teenage argument. It arrives at 13 and it peaks at 16. At 9 and 10 your child is not negotiating with you about an essay. They are asking a machine to write a song about the dog, generating a picture of their teacher as a dinosaur, and finding out that it will answer absolutely anything, immediately, in a friendly voice, at eleven at night.
At nine, AI does not arrive as homework. It arrives as a joke, a picture and a voice that always answers. By the time it looks like homework, the habits are already set.
Why the account matters more than the app
Ask the account question before the app question.
Parents at this stage tend to ask me which tool is safe. It is the wrong first question, because a 9 to 12 year old almost never chooses the tool. They use whatever is already signed in on whatever they are holding.
Common Sense found that among 9 to 12 year olds who use AI, 64 percent use it on their own device, 47 percent on a school device and 37 percent on a shared family device. Those overlap, because children use more than one device. The number that should stop you is the middle one. Nearly half of 9 to 12 year olds who use AI are using it on school hardware, through an account you did not create, under rules you did not set.
So the test I would use at this age is not about the app at all.
The Whose Account Question. Before you ask whether an AI tool is safe for a 9 to 12 year old, ask whose account it runs on. At this age there are only three: yours, with you there; the school's, which the school controls and you can ask about; or one they set up alone, which is the only one you have no view of at all. The third is the one to close.
Door one: your account, with you in the room
This is the one to say yes to. A child of nine or ten using a chatbot next to a parent is a good half hour. They see an adult ask a question, get an answer, and doubt it out loud.
If you want a phrase for what AI is that will hold from 5 to 15, I use this one: AI is a guessing machine. It has read more words than any person ever could, and it uses all of them to guess what should come next. It is very good at guessing. It does not know when it is wrong, so checking is your job. There is more on how to explain AI to a child if you want the version by age.
Door two: the school's account
This door opened wider in August, and most parents have not noticed.
Google's own announcement to school administrators says that from August 10, 2026, Gemini in Classroom is available to K-12 and higher education students of all ages. The announcement, dated August 4, 2026, puts the default plainly: "Access to the Gemini tab in Classroom is On by default for teachers and students of all ages." An administrator can switch it off, and access also depends on which Workspace for Education edition the school runs and on the settings applied to the group your child's account sits in. Read that again with your nine year old in mind. If your child's school runs Google and has not changed the setting, a nine year old can meet Gemini through their school account, in class, with no decision from you at all.
That is not a scandal. Schools are allowed to make decisions about the tools in their own classrooms, and many have thought about it carefully. But it is a fact worth knowing, and it is the reason the answer to "is this safe for my child" cannot come only from you. The details of what changed in Classroom are in the post on Gemini and Google Classroom.
Door three: an account they set up alone
An 11 year old can type 2012 into a birthday box. Age limits are a promise, not a lock.
The stated rules are at the top of this article, and they are worth saying out loud to your child rather than enforcing in silence. The sentence I would quote to them is Google's own. Its guidance for families, read on October 2, 2026, says the filters built for minors "try to limit access to inappropriate content, but they're not perfect. Your child may encounter content you don't want them to see."
That is the company writing about its own product on its own help page. It is the most useful sentence on it, because it tells you exactly what a setting is for: it shortens the odds, and it does not remove the need for you.
What I See in Schools
The schools have taught this age group the rules and not the judgment.
The same Common Sense census asked children what their school had actually told them about AI. Among 9 to 12 year olds, 42 percent said their school had explained how to tell whether information from AI is accurate or trustworthy. That is the lowest of the three age bands on that question: 58 percent of 13 to 15 year olds and 58 percent of 16 and 17 year olds said the same.
What makes it worse is the measure they do score well on. Asked whether their school had told them what they should and should not use AI for in schoolwork, the 9 to 12 band reads 80 percent, level with the oldest teenagers. So the rules have reached them. The judgment has not. They have been told where the line is and not how to tell whether the machine is talking sense.
That gap matches what I see when I work with schools. The AI policy gets written for the exam years first, because that is where the cheating risk sits and where the pressure comes from. The elementary and middle years get the same policy with the word "appropriate" in it, and a staff meeting that ran out of time. Meanwhile the younger children are the ones on the school device. Across the whole study, just 35 percent of 9 to 17 year olds correctly identified as false the statement that AI can tell the difference between things that are true and things that are false.
Sixteen percent of 9 to 12 year old chatbot users in the study had come across something inappropriate. Across the whole sample, 33 percent of the children who saw something inappropriate told a trusted adult, and 53 percent said they did not. Not telling a trusted adult is not the same as telling nobody, and the gap between those two is usually a friend.
I would not read that as a reason to ban the thing. I would read it as a reason to be the adult who has already asked.
What is appropriate at 9 and 10
Together, short, and out loud.
At this age the useful setup is a shared device in a shared room, your account, and AI treated like a kitchen appliance rather than a friend. Make things with it: a story where your child supplies the characters, a quiz on the topic they are studying, a picture of an invented animal. The child does the thinking. The machine does the typing.
Three habits are worth more than any control panel:
- Check one thing out loud, every time. Ask it something you already know the answer to, and let your child watch you catch it being wrong.
- Keep the first draft human. If your child is writing a story, they write the first version. AI can suggest a better word after that, not before.
- Name the strange feeling. Tell them, once, in ordinary words, that it is built to sound like a person and is not one. Google says the same thing to parents in its own guidance: help your child understand it is "an AI tool, not a person to confide in".
You could say: "This thing is really good at sounding sure. Let's see if we can catch it out."
What changes at 11 and 12
Privacy arrives, and the school account gets busier.
An 11 year old starts wanting a closed door, and they start having homework that AI could plausibly do. This is the year to introduce the question that keeps the learning with your child.
The Homework Test is one question to ask of any piece of homework before AI touches it: if the machine did this part, what would my child lose? If the answer is nothing, let it help. If the answer is the learning, that part stays with your child. There is a fuller version of that test, and what teachers actually look for, in the post on homework and cheating.
The other thing to add at 11 or 12 is a way of asking that gets you a real answer. Do not ask "do you use AI?" That gets a shrug. Ask "what were the last three things you asked it?" A general question gets a general answer. A specific one usually gets a phone held up.
The American Academy of Pediatrics, in guidance published on June 19, 2026 through its Center of Excellence on Social Media and Youth Mental Health, suggests much the same approach: low-stakes, non-judgmental prompts such as "How are you using AI? Can you show me how it works? How are your friends using it?" The AAP also names the thing to watch for, which is not a particular app but a change in sleep, in usual activities, or time spent with an AI app instead of other interests.
You could say: "Show me the last thing you asked it. I want to see if it got it right."
What changes at 13, so you know what is coming
Thirteen is not the year the platforms stop saying no. It is the year they start saying something different.
At 13 your child meets the age floor on most mainstream chatbots, use for schoolwork overtakes use for fun, and the conversation moves from what they are allowed to use to what they are allowed to hand in.
One thing has changed since this article was first written, and it matters for anyone with a 12 year old. OpenAI announced ChatGPT for Teens on August 18, 2026. Its words: "If our system estimates someone is under 18 or they state their age is between 13 and 17, they are automatically placed into ChatGPT for Teens." So the gate does not simply open. Your child is put into a different product, with study scaffolding, break reminders and safeguards on self-harm, eating disorders and explicit content, and with the company's guess at their age deciding which version they get. The fuller verdict, age by age, is in is ChatGPT safe for kids.
There is a full guide to the teenage band in what teens actually use AI for. The habits you build at 10 are the ones that have to hold at 14.
What none of this can guarantee
No control setting covers a logged-out browser, a friend's phone, or a school account you do not administer.
The parental controls are worth switching on, and I have written a walkthrough of what ChatGPT's controls do and do not do. But be clear about the limits. OpenAI's own FAQ on the controls, read on October 2, 2026, says a safety notification to a parent "does not include your teen's conversation, chat transcript, or chat history", and that parental controls "do not let parents or guardians read or monitor a teen's conversations". Controls work at the account level and the device level. They do not work at the level of a child sitting on a bus with someone else's phone, and Google says in its own words that the filters are not perfect.
Three things override anything in this article. Your child's school rule comes first on a school device and on schoolwork. The platform's own age limit comes first when it says no. And if you are worried about your child's mood, sleep or withdrawal rather than their screen time, that is a conversation for your pediatrician or family doctor, and for the school's pastoral or safeguarding lead, not for a settings page. If your child has been using a chatbot to talk about how they feel, the post on what to do when a child confides in AI has the practical version.
What to ask the school this term
Ask one question, and ask it as a list.
The question I would put to an elementary or middle school is the same one I put to a high school. Where does AI touch my child's work, and who checks it? It works because it cannot be answered with a policy. It has to be answered with a list. If the school runs Google Workspace, the follow-up is short and specific: is Gemini in Classroom switched on for my child's grade, who decided, and were the teachers trained before it was?
You do not need to be confrontational to ask this. The pattern I see is that the most useful parents' evening question is the plainest one, asked once, in writing, with a request for a written answer. A school that has thought it through will be glad you asked. A school that has not will now know a parent is watching, which is usually how these things start moving. There is more on what is actually happening inside schools in how AI is used in schools.
The one thing to do this week
Use it together once, on your account, with the door open.
The account check at the top of this article takes four minutes. The half hour that changes more comes after it: your child beside you, one question you both already know the answer to, and the sight of an adult doubting a confident machine out loud. At 9 to 12 that is the habit you are actually building. Everything else on this page is a setting.
If your child's school has not spoken to parents about AI yet, this is the work I do with schools: training the staff so they can explain it to children and to you. It is worth sending to your head of year or principal. And if you want this kind of thing without the panic, I write a Sunday newsletter read by more than 44,000 teachers and parents. Parents are very welcome.
Dan Fitzpatrick is The AI Educator: a former teacher and assistant headteacher, a Forbes contributor, and the author of bestselling books on AI in education. He has trained more than 150,000 educators across more than 30 countries and uses these tools every day. More about Dan.
