I don't understand AI myself. Can I still teach this?

Yes, and your not-knowing is useful. "I don't know, let's find out" models exactly the behaviour you want. The three core ideas — it predicts, it inherits our mistakes, people made choices about it — are all you need to hold.

What age should I start?

Around five for the conversational version, with no solo use. Around seven for hands-on with a guided tool. The concepts scale down further than most parents expect — children are used to things that guess.

Should schools be doing this instead?

Many are starting, unevenly. But the habits form at home, in the moments when your child actually reaches for the tool. That's not a gap in schooling so much as a difference in what home is good at.

How do I teach this without making them scared of AI?

Frame it as capability, not danger. "You can catch it being wrong" is a more powerful sentence to a nine-year-old than "be careful," and it produces a child who engages critically instead of one who avoids the subject.

How is teaching a 7-year-old different from a 12-year-old?

Same ideas, different delivery. Under about ten, teach AI away from the screen through pattern games and shared use on your account. From ten up, move to real tools and real projects, and add the harder topics: bias, privacy, and why a confident answer can still be invented. The table above splits it row by row.

Parent's guide

Teaching kids about AI: a parent's roadmap, no CS degree required

You don't need to understand transformers to teach your child about AI, any more than you needed to understand internal combustion to teach them about road safety. What they need from you is a handful of correct mental models, delivered at the right ages, mostly in conversations that take two minutes. Here's what to teach, in what order, and how to turn ordinary moments into the lesson.

The three ideas everything else hangs off

First: AI predicts, it doesn't know. It has read an enormous amount of text and produces what tends to come next. That's why it can write a beautiful paragraph about a book that doesn't exist — it isn't lying, and it isn't looking anything up. Once a child holds this idea, confident wrongness stops being confusing and starts being expected.

Second: it learned from us, so it inherits our mistakes. AI systems are built from human-made material, which means human blind spots and biases come along. This is where a child's natural sense of fairness becomes a genuine analytical tool.

Third: someone made choices about it. AI didn't fall out of the sky. People decided what it should do, what it should refuse, and who it's for. That idea — that technology has authors — is the one that turns a passive user into someone who asks questions.

The order to teach it, by age

1. Ages 5–7: it's a machine that guesses Keep it concrete and mostly off-screen. The autocomplete on your phone is the perfect prop: it guesses the next word, sometimes hilariously wrong. That's the whole lesson at this age, and it's a surprisingly complete one.
2. Ages 7–9: it's often right and sometimes confidently wrong Now add the check. Ask it something your child is an expert in — their favourite game, their sport, their book series — and let them catch the errors. Being the expert who corrects the machine is a formative experience at this age.
3. Ages 9–11: you can direct it, and how you ask changes what you get This is where the real skill starts. Vague request, vague result; precise request, useful result. Have them ask for the same thing three different ways and compare. They'll discover prompting without anyone using the word.
4. Ages 11–12: who made it, and what does it want? Introduce the idea that products are designed with intentions — some to help you finish, some to keep you there. Ask which one they're using and how they can tell. This is media literacy, and it transfers far beyond AI.

Ages 7–9 vs 10–12: what changes

The core ideas stay the same at every age. What changes is how your child meets them — younger kids learn AI best away from the screen, older kids are ready to use real tools with you beside them and to handle the trickier parts.

What you are teaching	Ages 7–9	Ages 10–12
The big idea	AI guesses from patterns it has seen before. Keep it concrete: it is a very good guesser, not a knower.	AI predicts the next likely words, which is why it can sound certain and still be wrong. Name the mechanism.
How they meet it	Mostly unplugged — sorting games, autocomplete, spotting the recommendation feed guessing. Any tool use is shared, on your account.	Hands-on inside a guided, kid-safe tool. Real projects they own, with you checking in rather than sitting through it.
What you say	"Let's see if it gets this wrong." Turn every error into a game rather than a warning.	"How would you check that?" Push for the source, not just the doubt.
What to practise	Describing what they want clearly enough that someone else could follow it. Precision, before prompting.	Directing, judging and revising — keep this, cut that, try again — and explaining the finished thing in their own words.
The hard conversation	Personal details never go in. Simple, absolute, no exceptions.	Bias, made-up sources, and why "the AI said so" is not a reason. They can hold the nuance now.
What to watch for	Treating AI as a friend or an authority. Redirect to "it is a tool we use together".	Quiet dependence — using it for the part of the homework that was the point. Agree out loud where the line is.

Everyday moments that are already AI lessons

The autocomplete game — Let your phone's keyboard finish a sentence by tapping the middle suggestion repeatedly. Read the nonsense aloud. That is a language model, at a scale a seven-year-old can hold.
The recommendation question — "Why do you think it showed you that video?" Your child has strong opinions about this already. It's the cleanest introduction to algorithms shaping what you see.
The expert test — Ask an AI about the thing your child knows best. Errors will appear. Their outrage is the lesson landing — and it makes scepticism feel like power rather than homework.
The two-answers trick — Ask the same question twice in fresh sessions and compare. A source that gives different answers to the same question is not a source of truth, and children work that out instantly.

What to teach after understanding: the habits

Understanding what AI is turns out to be the easy half. The half that changes outcomes is behavioural, and it's four habits: think first before asking; be specific in what you ask for; check what comes back; and be able to explain the result in your own words. Every one of those is teachable at seven, and each one is a small defence against the failure mode where AI does the thinking and the child collects the output.

The most efficient way to teach all four at once is to have your child build something with AI rather than ask it things. A project forces a plan, a specific instruction, a test, and an explanation — in that order, every time, without you having to run the lesson. That's the entire pedagogical argument for creation over conversation.

Questions parents ask

I don't understand AI myself. Can I still teach this?: Yes, and your not-knowing is useful. "I don't know, let's find out" models exactly the behaviour you want. The three core ideas — it predicts, it inherits our mistakes, people made choices about it — are all you need to hold.
What age should I start?: Around five for the conversational version, with no solo use. Around seven for hands-on with a guided tool. The concepts scale down further than most parents expect — children are used to things that guess.
Should schools be doing this instead?: Many are starting, unevenly. But the habits form at home, in the moments when your child actually reaches for the tool. That's not a gap in schooling so much as a difference in what home is good at.
How do I teach this without making them scared of AI?: Frame it as capability, not danger. "You can catch it being wrong" is a more powerful sentence to a nine-year-old than "be careful," and it produces a child who engages critically instead of one who avoids the subject.
How is teaching a 7-year-old different from a 12-year-old?: Same ideas, different delivery. Under about ten, teach AI away from the screen through pattern games and shared use on your account. From ten up, move to real tools and real projects, and add the harder topics: bias, privacy, and why a confident answer can still be invented. The table above splits it row by row.

Four mistakes that are easy to make

Teaching it as a warning — A child taught that AI is dangerous learns to avoid the subject with you, not to avoid the tool. They'll still use it — they'll just stop mentioning it. Capability framing keeps you in the conversation, which is the only position from which you can help.
Explaining the technology instead of the behaviour — Neural networks are interesting and almost irrelevant at this age. What changes outcomes is the habit of checking and explaining. Save the architecture for a child who asks — and some will.
Doing it once, as a talk — This isn't a sit-down conversation, it's twenty small moments over a couple of years. The autocomplete demo at a bus stop teaches more than a prepared lecture, because it happens when the child is curious rather than when you're ready.
Waiting until you understand it properly — The most common reason parents don't start. You will never feel qualified, and you don't need to be — the three core ideas fit on a postcard, and "let's find out together" is a legitimate teaching method.

What good looks like at 12

It's worth knowing what you're aiming at, because the target isn't a child who can define machine learning. A twelve-year-old who has been taught well will reach for AI on purpose rather than reflexively — using it for the parts where it helps and doing the rest themselves, because they've noticed the difference.

They'll be casually sceptical: they check things, they notice when an answer is too smooth, and they're not impressed by fluency alone. They can say what they want precisely, and they get annoyed rather than confused when the result is wrong. They can explain what they made. And they think of it as a tool they operate, not an authority they consult or a companion who understands them.

None of that requires technical education. It requires a few correct ideas delivered early and a lot of small, repeated practice — which is well within what an ordinary household can do.

Let the tool teach the habits

Luchi missions build the four habits by design — plan, direct, check, explain — while kids make games and stories they're proud of. Free to start.

Start free