The short answer
For most children, the useful thing is not a formal AI course at all. It is understanding three ideas: that these tools are machines rather than people, that they can be completely wrong while sounding completely certain, and that what you type into them does not stay private. All three can be taught at home in ordinary conversation, and they matter more than any curriculum. Free courses do exist and some are genuinely good, but a child who has those three ideas is better prepared than a child who has finished a course without them.
What is actually worth a child understanding?
The right answer depends far more on your particular child than on their birthday, so treat these as loose bands rather than stages to tick off.
Young children. Keep it to two sentences they can repeat back. It is a machine, like a very chatty calculator. It is not alive, it does not know you, it has no feelings, and it sometimes makes things up without realising. That is genuinely all a small child needs, and it heads off the more unsettling belief that there is a friend inside the screen.
Roughly primary age. Add the idea that it guesses. It has read an enormous amount of text and it predicts what words usually come next, which is why it is fluent about things it has no real knowledge of. Children of this age find that funny rather than frightening, which is exactly the reaction you want. Add one rule about privacy: we do not tell it who we are.
Roughly secondary age. Now the detail is worth having. Roughly how these systems are built and why that makes them confident and unreliable at the same time. How to check a claim rather than accept it. What happens to a conversation after you close the tab. And the distinction that will matter for the rest of their education: using a tool to understand something is not the same as using it to avoid understanding it.
In plain English
- Chatbot:
- A program you type to and it types back. Fluent, but it has no idea what it is saying.
- Hallucination:
- When the tool invents something and states it confidently. Fake facts, fake quotes, fake sources.
- Training data:
- The huge pile of text a system learned patterns from. Not a library it can look things up in.
- Prompt:
- What you type in. Clearer prompts get better answers, which is most of what people mean by prompt skills.
What are the rules about age and accounts?
Every major AI service sets a minimum age in its terms, and many require a parent or guardian to be involved below a certain age. Those thresholds are not the same from one service to the next, they are not the same in every country, and they change as rules and products change.
So do not take a number from a blog post, including this one. Open the service your child wants to use, find its terms or its help pages, and read what it currently says about age and parental consent. It takes about five minutes. If you cannot find a clear answer, treat that as an answer in itself.
Two practical notes. Setting up an account in your own name and handing the device over is not the same as the service permitting your child to use it, and it is worth being honest with yourself about which one you are doing. And your child's school may have its own rules about which tools pupils may use for schoolwork. Where the school has a policy, the school's policy wins.
What should I check before letting a child use a tool?
Four questions, and you can answer all of them from the service's own pages.
What does it say about age? As above, from the source, today.
Are conversations used to improve the service? Many are, by default. Some let you turn that off in settings. This is the difference between a chat that stays between you and one that may end up in a pile of text used for future training or reviewed by a human.
Can you see and delete the history? A tool that lets you review what your child asked and remove it is a tool you can actually supervise. One that keeps no visible record, or offers no way to delete, gives you nothing to work with.
Supervised or alone? The same tool can be fine at the kitchen table and unsuitable in a bedroom at midnight. Decide which one you are agreeing to, and say so out loud.
Nothing that identifies your child goes in: full name, school, class or teacher, home address, phone number, photographs of themselves, or anything at all about their friends and siblings. Other people's information is not theirs to share. A line children remember: if it names you, shows you, or tells someone where to find you, it does not go in the box.
How do I show a child that AI can be confidently wrong?
Telling them does not work. It is an abstract warning about an object that sounds authoritative, and the object wins. Showing them takes ten minutes and they never quite forget it.
Pick the thing your child knows better than you do. A game, a sport, a series, a hobby, an obsession. Ask the tool to explain it in detail, then hand the answer to your child and ask them to mark it.
Explain [the thing my child is an expert in] to someone who has
never heard of it.
Include specifics: names, numbers, rules, dates, who did what.
About 200 words. Write confidently.
What usually happens is that the answer is broadly right and specifically wrong, and your child spots it immediately. That character is not from that series. Those two rules do not go together. Nobody says that. The tone never wavers while it is getting things wrong, and that is the whole lesson: it does not sound less certain when it has no idea. If your child asks why, the honest answer is that it is not lying, it is predicting plausible words, and it has no way of knowing when it is wrong.
Then turn it round. Ask it about something none of you know, and notice that the answer sounds exactly the same. That is the point where the idea properly lands.
Where is the line between using AI to learn and using it to avoid learning?
Practically, the test is simple. Did the child do the thinking, or did the machine? A tool that explains a step your child is stuck on, quizzes them before a test, or asks them awkward questions about their own argument is doing something a good tutor would do. A tool that produces the finished piece has removed the part where the learning happened and left a plausible object behind.
โ Weak prompt
Prompt
Write me a paragraph about why the Romans built roads.
Output
A tidy, well-organised paragraph, ready to copy out.
The homework is finished and nothing was learned. Repeated over a couple of years, the ability to build a paragraph never develops, and that shows up at exactly the wrong moment.
โ Good prompt
Prompt
I am writing a paragraph about why the Romans built roads. Here is what I think so far. Ask me three questions that would make my answer better. Do not write it for me.
Output
Three questions about who benefited, what it cost, and what the evidence is.
The child still writes every word, but has to think harder first. This produces better work than the shortcut does, and it is defensible if a teacher asks.
School policies on this vary enormously and are still settling. Ask what your child's school expects rather than assuming, and if the school says no AI for a particular piece of work, that is the rule.
Does LearnAI work for children?
Honestly: not for younger ones. LearnAI is written for adults. The courses assume an adult reader with an adult's context, jobs, hiring, managing people, running a small business, and the examples reflect that. A ten year old would find them dull rather than harmful, but dull is enough of a reason not to bother.
A motivated teenager is a different case. The beginner material works fine for them, and two things genuinely help: the site has no accounts and collects no personal data, so there is nothing for a young person to sign up to and nothing of theirs stored. If you have a teenager who wants to understand this properly, we wrote them their own guide: AI for High School Students is addressed to them rather than to you, which tends to go down better. Getting Started with ChatGPT and AI Safety and Verification are the two courses worth their time, and the full course list shows what else is there.
For a younger child you want something built for children, and that is not us. We would rather tell you what to look for than sell you something ill-fitting: material aimed at your child's actual age, no account required or a clear parental route if one is, plain answers about what happens to conversations, and activities that involve doing something rather than watching a video. Judge any product against that list, including ours.
If you are a teacher rather than a parent, the classroom version of all this is in our free AI for Educators course. And if you want to understand these tools yourself first, which is the best possible preparation for explaining them to a child, start here.