Is ChatGPT safe for kids? What I allow at each age

Short answer: ChatGPT is safe enough for a child to use with you from about nine, and on their own from the early teens, provided it is used to help them think and not to think for them. It is not safe for a child of any age to use as a machine that does the homework. The danger is not in the app. It is in the habit.

That distinction is the whole book in one sentence, so let me earn it.

The rules first, because you will be asked

I checked OpenAI's own pages on the date at the top of this post. Here is where they stand.

Minimum age. ChatGPT is not meant for children under 13, and OpenAI requires that 13 to 17-year-olds have a parent's consent to use it. If a teacher uses it with a class of under-13s, the company says an adult must be the one operating it.

ChatGPT for Teens. From August 2026 OpenAI has been rolling out a teen version of ChatGPT that switches itself on. If the account's stated age, a verified age, or the company's own age-prediction system says the user is under 18, the account gets the teen experience automatically. It is the same ChatGPT with age-appropriate safeguards around self-harm, violence, eating disorders, dangerous activities and sexual content, plus study tools: a study mode that gives hints and asks questions instead of handing over answers, homework reminders, quizzes, and reminders to take a break. An adult who is wrongly flagged can verify their age to switch it off. A teenager who lied about their birthday may still be caught by the prediction, which I find reassuring, though no such system is perfect.

Parental controls. You can link your account to your teenager's from Settings, by phone number or email, and either of you can start the invitation. Once linked, you can turn these on or off: reduce sensitive content (on by default), whether their chats are used to improve the model, saved memories, voice mode, image generation, a study mode for new chats, study hours, and quiet hours when ChatGPT cannot be used at all. You cannot read their conversations. In limited situations, if trained reviewers see a serious self-harm concern or signs of disordered eating, or if the account is suspended over violence, you get a safety notification. Either of you can unlink at any time.

Those controls are worth ten minutes of your evening. But I want to be honest about what an age limit is. It is a box a child ticks, and a prediction system is a guess. The limit that protects your child is the one you set at home, and it needs to be about how the tool is used, not whether.

The three risks that are real

Parents worry about the wrong things with ChatGPT. Nobody is going to be kidnapped through it, and it is not going to tell your ten-year-old how to build a bomb. The real risks are quieter and they arrive over months.

It does the thinking, and the thinking is the point. The most talked-about study on this is a 2025 MIT Media Lab experiment, reported everywhere as "Your Brain on ChatGPT". People wrote essays over several months while researchers measured brain activity. The group using a chatbot showed the weakest engagement, remembered least about what they had written, felt least ownership of it, and did worst when later asked to write without the tool. The researchers called it cognitive debt. It is one modest study and the authors themselves warned against "AI damages brains" headlines. But its mechanism is the one I have watched at my own kitchen table: writing is thinking, and if you outsource the writing you skip the thinking.

It is confidently wrong. ChatGPT does not know things. It predicts words. Most of the time that produces something accurate, and some of the time it produces something fluent, plausible and false, delivered in exactly the same tone. A child who has not been taught to check will not check.

It is a stranger you tell things to. Whatever your child types in goes to a company. The rule I use for every AI tool is the same: only put in what you would happily show a stranger in the street. The teen version now nudges before some photo uploads, which is welcome, but a nudge is not a rule.

What I allow, by age

I set different rules for a five-year-old and a fifteen-year-old about crossing the road. AI is no different.

Ages 5 to 8: a toy we play with together. No account, no solo use, and OpenAI agrees. But do not hide it either. Sit together on your account and make a story about a dragon who is scared of toast. Ask it to draw the dragon. Laugh at what it gets wrong. The lesson at this age is conceptual and it is simple: this is a clever machine that makes things up, it is not a person, and it is not always right. Children who learn that early are far harder to fool later.

Ages 9 to 12: the supervised apprentice. Still on a parent's account or alongside you, because the rules say 13. This is where the 20-Minute Rule starts, and it is the single most useful thing in this post.

At this age the AI lives on a shared device in a shared room, and you are nearby and curious, not hovering. Start saying the sentence that runs through the whole book: "Is this helping you think, or thinking for you?" Say it lightly and often. It becomes a voice in their head.

Ages 13 to 18: the independent user. Your teenager will use ChatGPT without you, for school and for everything else, whether or not you allow it. So the job changes from permission to judgement. Link the accounts, switch on study mode and quiet hours, leave memory off, and then hand over the two tools that carry the real load.

The first is the V.E.T. rule for anything an AI produces. Verify it against a second source. Explain it back in your own words. Transform it into your own work. If your child cannot explain it, they have not learned it, and it does not get handed in.

The second is the "Could I do this without it?" test. Before submitting anything AI-assisted, your child asks themselves: "If someone asked me to explain this, could I?" If the answer is no, the AI did their thinking for them. This is not policing. It is the honest self-awareness they will need in every job they ever have.

And at this age, write the rules down together. The Family AI Agreement in Chapter 10 is a one-page template: which tools, for what, what never gets typed in, and what happens when something goes wrong. The builder on this site is free and takes five minutes. The agreement matters less than the conversation you have while filling it in.

The mistake I see most

The parent who bans ChatGPT because they are worried about cheating has not stopped the cheating. They have made sure it happens at a friend's house, on a phone, with no guidance at all. I understand the instinct. I felt it. But the evidence points the other way: the problem in almost every case is not AI use, it is unguided AI use. Guide it.

Questions parents ask

How old do you have to be to use ChatGPT?

OpenAI says ChatGPT is not meant for children under 13, and that 13 to 17-year-olds need a parent's consent. Accounts identified as under 18, whether by stated age or by OpenAI's age prediction, are switched to ChatGPT for Teens automatically. Check the current pages, because they change. My own rule is that solo use starts in the early teens, and use with a parent can start around nine.

What can ChatGPT parental controls actually do?

Once your accounts are linked you can reduce sensitive content, stop their chats being used to train the model, turn off saved memories, voice mode and image generation, make new chats start in study mode, set study hours, and set quiet hours when ChatGPT cannot be used. You cannot read their conversations. You may be notified if reviewers see a serious self-harm concern or signs of disordered eating.

Will my child's teacher know they used ChatGPT?

Often, yes, and not mainly through detection software, which is unreliable. Teachers notice when the writing does not sound like the child, and when the child cannot discuss what they handed in. That second one is the point. If your child can explain the work in their own words, it is their work.

Should I just turn it off completely?

Only for under-nines, and even then I would use it together. For older children a ban does not remove the tool from their lives; it removes you from the conversation about it. Set the 20-Minute Rule, teach V.E.T., write the agreement, link the accounts, and stay curious.

Sources

• OpenAI help centre, "Is ChatGPT safe for all ages?", "Managing parental controls in ChatGPT" and "ChatGPT for Teens", all checked on the date at the top of this post.
• MIT Media Lab, 2025, the essay-writing and brain-activity study widely reported as "Your Brain on ChatGPT".
• Harvard University randomised controlled trial on AI tutoring in an undergraduate physics course.
• Barna Group, 2024, survey of parents' concerns about AI and children.