Your Child's First AI Session: A Minute-by-Minute Script

Most first AI sessions with a child fail in the same forty seconds.

Not because the tool was wrong or the child was too young. Because of what the adult said first. The session opens, the parent turns to the child, and asks the most natural question in the world:

"So — what do you want to make?"

And the child says "I don't know."

That is not a lack of imagination. It is the correct response to an impossible question. You have just handed a seven-year-old an infinite canvas and asked them to pick a point on it. Adults freeze at this too; we just hide it better.

What follows is a tested script — the actual words, minute by minute, for three age bands. It assumes fifteen minutes, one adult, one child, and no prior setup.

Last reviewed: August 2026.

Before you start: the two things to get right

1. Use your own account, and be in the room

Not a suggestion about vigilance — a fact about the terms. As of August 2026, by their own published policies:

• Claude — 18+
• ChatGPT — 13+ (13–17 with parental permission)
• Gemini — 13+ for the consumer app; under-13 only inside a school-managed Workspace account with the district's switch on and Family Link consent

If your child is under thirteen, none of these is theirs to hold. So the session runs on your account with you present. This turns out to be the good version anyway: your presence is the difference between a workshop and a vending machine.

For the first session, none of those three is the right door. Start somewhere narrower — reasons below.

2. Decide what "success" means before you begin

Write it down if you have to, because in the moment you will drift toward the wrong one.

Success is not: an impressive output, a happy child, or fifteen minutes of quiet.
Success is: your child made at least one decision that changed the result, and can tell you what it was.

That is the whole bar. A session that produces an ugly drawing and one real decision beat a session that produced a gorgeous image the child merely requested.

The tool for session one

Not ChatGPT. Not an image generator. Quick, Draw! — quickdraw.withgoogle.com

The child draws; a neural network tries to guess what it is, out loud, while they draw.

Three reasons this is the right first door, and they are worth understanding because they are the criteria for every tool you pick after this:

1. The child acts first. The machine cannot do anything until the child has drawn. Authorship is structurally correct — you could not accidentally turn this into a wish machine if you tried.
2. It fails visibly and harmlessly. It will guess wrong, and the wrongness is funny rather than alarming. That single experience — it's confident and it's wrong — is the foundation everything else builds on.
3. Nothing to set up. No account, no card, no app. Two clicks from a cold browser.

There is a fourth reason that matters more than the other three: you can explain how it works in one sentence a six-year-old will actually understand. We will get to that sentence.

What most parents get wrong in the first three minutes

Even with the right tool, there are three moves that quietly cost you the session. I have made all of them.

Narrating the machine's success. "Wow, it got it! Clever, right?" — this frames the AI as the impressive one. The child becomes the audience. Ninety seconds in, you have set up the wrong relationship for the rest of the year.

Rescuing too fast. The child's drawing is unrecognisable, the AI is flailing, and you reach for the mouse. Don't. That flailing is the single most valuable thing that will happen in the fifteen minutes.

Explaining before they have seen anything. The urge to say "so, AI is a system that recognises patterns based on training data…" — the child has no hook to hang it on yet. Explanation lands after experience, never before. Two minutes of drawing buys you a sentence of explanation that actually sticks.

The script below is built to make all three mistakes hard to make.