Start with judgment, not tools. Before a child learns to use AI, they need to notice when a machine is answering them, say in their own words what it is doing, and check what it says. Age changes the words you use, not the goal. Two things still come first: your school's rules, and the platform age limits, which most children are under.
What does it actually mean to teach a child about AI?
It means four things, and using the tools is only one of them.
In 2024 UNESCO, the education arm of the United Nations, published an AI competency framework for students: a list of what schools around the world should cover. It names four aspects. A human-centred mindset. Ethics of AI. AI techniques and applications. AI system design. UNESCO describes that first aspect as students' "values, beliefs and critical thinking skills, applied to the examination of whether AI is fit for purpose, whether its use is justified, how humans should interact with it."
Look at the order. The techniques come third.
Here is that list at a kitchen table, which is my reading of it rather than UNESCO's wording:
- Should this be a machine's job at all, and what is my part in it?
- Who does this help, who does it harm, and whose work went into it?
- What is it, how do I use it, and why does it get things wrong?
- How is it built? Only if your child wants to go that far.
Almost everything sold to parents as teaching kids about AI is the third one, taught as prompting. Prompting is a keyboard skill with a short shelf life. The first two are the ones that will still be worth something when your child is twenty-five.
Can my child already spot AI? Probably not
Probably not, and neither can most adults. That second half is the part that surprises parents.
Psychologists at Temple University in Philadelphia ran a study published in the journal Technology in Society in 2026. Thirty-seven children aged 6 to 10 were introduced to a "SmartBot," then shown short texts and images and asked, each time, whether a person or the bot had made it. The children performed at or below chance. They were significantly worse than a comparison group of 49 adults aged 18 to 34. Children who spent more time on devices at home did worse rather than better, though that is an association in a small study and not evidence that screen time causes it.
The UK numbers say something similar about the older ones. Ofcom's Children and Parents: Media Use and Attitudes report, published on 21 May 2026 and covering 8 to 17 year olds, found that 59% correctly picked out an AI-generated image of a celebrity. Another 13% thought it was real, and the rest were unsure. Among teenagers, a quarter said they were confident they could tell, and then got it wrong.
The number that worries me is not the 59%. It is the confidence. A child who knows they cannot tell will ask. A child who is sure they can tell will not.
Your child will meet AI as an answer. The work at home is to make sure they meet it as a claim.
What I See in Schools
Schools are further behind on this than most parents assume, and it is not because teachers do not care.
When I read a school's AI policy, it is nearly always about two things: staff use, and cheating. Both matter. Neither is a lesson. The question I give parents to ask instead is the Where Question: 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.
England has now committed to fixing the teaching side. The government's response to the Curriculum and Assessment Review, published on 5 November 2025, makes citizenship compulsory so that every pupil covers media literacy, and promises AI literacy inside a rebalanced computing curriculum. The review's own finding, quoted in Computing at School's summary for teachers, is the sentence I would put in front of every school leader: young people are not picking up digital literacy on their own, and assuming they will is a mistake. A new national curriculum takes years to reach an actual classroom, and this one has not arrived yet.
In the United States it is a patchwork. ExcelinEd's review of state K-12 AI policy, published on 26 May 2026, counted seven states with AI laws in place, around ten with guidance or a task force only, and close to a hundred bills in play. Most of that activity is about protecting children and vetting products. Very little of it says what a student should be taught. Utah's middle school AI literacy bill is one of the few that does.
So the honest position, in both countries, is that "the school has an AI policy" and "my child is being taught to judge AI" are usually a few years apart. That gap is yours for now.
What to teach at each age
Under 8: notice when it is a machine
The goal is not understanding. It is noticing.
Say it out loud when it happens, in the ordinary moments: the summary at the top of a Google results page, the voice in the smart speaker, the toy that answers back.
You could say: "That answer was written by a computer program. It made a guess. Should we check whether it is right?"
None of the big chatbots is built for this age. As of September 2026, OpenAI's own help pages state that ChatGPT "is not meant for children under 13," and that where it is used with younger children the interaction should be carried out by an adult. Toys that talk back are a separate decision, and the testing on those is worth reading before you buy one.
8 to 12: the guessing machine, and three questions
The goal is one explanation and one habit. Give them the explanation I use with every age from five to fifteen:
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.
Then the three checking questions, in this order, every time: Who says? How could we check? What would it look like if this were wrong? The longer version of that conversation is here.
This is also the band where the age rules bite. ChatGPT's minimum age is 13. As of September 2026, Google requires 13 or over for Gemini on a personal or school account, and an under-13 can only reach it through a supervised Google Family Link account that a parent switches on, which Google states is not available for supervised accounts in the UK, the European Economic Area or Switzerland. Those limits are self-declared and weakly enforced, so they are a line rather than a lock. Google's own guidance to parents is worth quoting to your child: Gemini "should not be treated as a friend or used to discuss personal problems or secrets."
13 to 15: where help ends and doing begins
The goal is the line, and they should be able to draw it themselves. 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.
Add the two things a teenager can now understand and a nine-year-old cannot: that the model learned from other people's writing, and that what they type into it goes somewhere.
You could say: "Show me the part you would have gotten wrong without it."
That question does more than a rule, because it makes them look at their own work rather than at you. What teenagers actually use AI for is usually broader than parents expect, and less alarming.
16 to 18: declare it, verify it, notice the pitch
They are about to use this on applications, at work, and in a first college or university term where somebody else writes the rules. Three things carry over: say when you used it, open the source rather than trusting the summary of it, and notice when a fluent answer is doing persuasion rather than reporting.
You could say: "What did it leave out?"
The habit that does most of the work
Children do not learn to doubt a machine by being told to doubt it. They learn it by watching an adult do it.
So: the Out Loud Rule. Whenever you check something an AI told you, do the checking out loud, in front of your child, including the part where you turn out to have been wrong. Thirty seconds at the kitchen table beats a lecture. Open the link. Say "that does not sound right." Say "I was wrong about that." A child who has heard you be wrong in public has permission to be wrong in public, and that permission is the whole skill.
One thing to do tonight: pick one AI answer your family met today, and run the three questions on it out loud. Once. That is the lesson.
What none of this guarantees
No amount of teaching makes a child immune, and it would be dishonest to suggest otherwise. Detection tools for AI-generated text and images are unreliable, so "we will just check it with a detector" is not a plan. A child who can spot a fake in a calm moment can still be fooled by one that flatters them or frightens them.
Some things also outrank anything in this post. Your child's school sets the rules for their homework, and those rules win. Platform age limits are the company's own line, not a safety guarantee, but they are still the line. And if your worry is really about your child's mood, their sleep or their safety rather than their schoolwork, that belongs with the school's designated safeguarding lead or school counselor, and with your GP or pediatrician, not with a lesson about AI.
The principle underneath all of it is the one I give schools, said for families: let AI help with the doing; keep the thinking with your child.
If your child's school has not yet told parents what it actually teaches children about AI, that is a fair question to ask, and it is the work I do with schools: training staff so the policy turns into lessons. It is worth sending on. If you want the wider picture first, the parent's guide to AI covers what you need and what you can skip, and I write a Sunday newsletter that keeps teachers and parents up to date without the panic. Parents are very welcome.
Dan Fitzpatrick is a former teacher and assistant headteacher who now speaks on stages in more than 30 countries and has advised the UK Department for Education, KHDA Dubai and the Ministry of Education in Kazakhstan on AI in education. More about Dan.
