From consumer to creator: safe first projects for a child in their first month with AI

The first month shapes everything that follows. A week-by-week plan of small, safe, genuinely creative projects for a child meeting AI tools for the first time, with the ground rules that make the whole thing workable for parents.

There is a moment, in most households now, when a child gets access to an AI tool for the first time. Sometimes it is planned. More often it arrives sideways - through a friend's phone, a school account, a feature that appeared inside an app they already had. Either way, the first month matters more than any month after it, because that is when the child decides what this thing is. A vending machine for answers? A toy that talks? Or a material you can make things with?

Habits set early are stubborn. A child who spends their first month asking the machine to do things for them will keep doing that, and it will cost them at school and beyond. A child who spends the first month making things with it - and catching it being wrong - starts down an entirely different road. The difference is not the tool. It is the projects. So here is a month of them: small, cheap, tested by ordinary family life, and designed so that an adult can supervise without needing any technical background.

Before week one: the ground rules

Ten minutes of setup buys you the whole month. Four rules, agreed out loud, before the first session:

1. The account belongs to the adult. Whatever tool you use, a parent or guardian holds the account and settings. If a platform offers a proper parent-managed account with a sealed space for children, prefer it over a general-purpose tool that merely tolerates young users. If the tool has no child provision at all, the sessions happen on your account, together.
2. Personal details stay out. No full name, no school name, no address, no photos, nothing about friends. Frame it the way you frame road safety: calm, absolute, not up for debate. This rule matters less on a well-designed children's platform and absolutely on everything else, so teach it as universal.
3. Sessions happen in shared space. Kitchen table, not bedroom. Not because you expect disaster, but because the best moments of the first month - the machine says something absurd, the story takes a turn - are worth being around for. The supervision is a side effect. The companionship is the point.
4. Every session ends with something to show. This is the creator rule, and it does more work than the other rules combined. If there is nothing to show, the session was consumption, and one of those a week is plenty.

Week one: interview the machine

Resist the urge to start with something useful. Start with the question every child is already carrying: what is this thing?

The project: the child interviews the AI, and writes down (on actual paper) the three most interesting answers. Good interview questions, which children largely generate on their own once given permission: Are you alive? Do you have feelings? Do you remember me from yesterday? What are you bad at? Can you lie? What happens if I ask you the same thing twice?

That last one is a gem - ask it, then ask it again, and watch the answer change. For a child, discovering that the machine does not give the same answer twice does more to establish "this is a system that generates text" than any adult explanation. A well-built children's tutor will answer these questions honestly (it is software, it does not have feelings); if the tool your child is using gets evasive or whimsical about what it is, that is worth knowing in week one rather than month six.

To show at the end: the paper interview, and one sentence from the child - "the most surprising thing it said was..."

Week two: the child as author, the machine as editor

Now make something. The classic first build is a story, because every child has one and the roles are easy to police: the child holds the pen, the machine is staff.

The child writes an opening - four or five sentences, however rough. Then the machine gets used three ways, in order: as a describer ("give me five words for how a cave smells"), as a challenger ("ask me three questions about my hero that I have not thought about"), and as an editor ("point at my weakest sentence and say why, but do not rewrite it"). That last instruction is the important one, and the child should type it themselves. Teaching a child to explicitly forbid the machine from doing their work is possibly the single most valuable prompt they will ever learn.

If stories do not land, the same structure works for a comic script, a song, instructions for an invented game, or a menu for a horrible restaurant. Authorship is the constant; the genre is negotiable.

To show: the finished piece, read aloud, with one honest sentence about which ideas were theirs and which the machine offered.

Week three: catch it out

This is the week the child learns the machine can be wrong - not by being told, but by catching it personally.

The project: pick a subject the child knows better than almost anyone. Their football club. A dinosaur. The plot of a book they have read nine times. Ask the machine to write a paragraph about that subject, and have the child mark it like a teacher: tick what is right, circle what is wrong, hunt down the truth for anything uncertain using a book or an adult or a proper source. Errors found are points scored. Most children find something, and the ones who find a confident, fluent, entirely wrong sentence have received an inoculation that will protect them for years.

End the week with the follow-up question at dinner: if it can be wrong about your dinosaur, what else might it be wrong about? Let them get there themselves.

To show: the marked-up paragraph, red pen and all. Stick it on the fridge. It might be the most important artefact of the month.

Week four: make a thing that leaves the screen

The final project has one requirement: the output must exist away from the device. A printed quiz for Sunday lunch, with the child as quizmaster. A poster for their bedroom, planned with the machine, made by hand. A treasure hunt with riddle clues they drafted and the machine helped sharpen. Instructions for a card game the family will actually be made to play.

The reason for the requirement is honest and slightly sneaky: things that leave the screen get audiences, and audiences complete the creative loop. A quiz that makes Grandad laugh teaches a child something no app can - that making things is a way of reaching people. That lesson, more than any technical skill, is what turns a consumer into a creator.

To show: the thing itself, in use, with witnesses.

After the month

Four weeks in, you will know things about your child's relationship with these tools that no settings screen could tell you - and your child will have a working model of AI that is roughly right: a capable, tireless, sometimes-wrong collaborator that holds no pen it is not handed. Keep the rules. Keep the show-me habit. Retire the training wheels at your own pace.

What to try this week

Start week one tonight. Twenty minutes, kitchen table, paper and pen ready. Agree the four ground rules out loud, then let your child interview the machine - and try, hard as it is, to let them ask their own questions. Your job this week is not to teach. It is to sit close, be curious alongside them, and enjoy being present at the moment your child asks a machine whether it is alive.