How do I get my child to make things with AI instead of just asking it for answers?

Nothing about the interface suggests a second attempt. A box, a response, a cursor waiting for the next unrelated question. Everything a child has learned about search says the answer arrives and the task is over.

That shape is fine for a question with an answer and useless for making anything, because the first version of anything is bad. Not bad because the tool is weak, but because the child has not yet seen a version to react against. The second prompt is usually the first one that contains real information, since it is the one written by somebody who now knows what they did not want.

So the habit to break is not "using AI". It is stopping at the first result.

The reliable trick is a task with a condition the child can check, where a single generic result visibly fails.

A cover for a book they have actually read, that somebody who read it would recognise with the title removed. A short piece of music for the moment their pet does the specific irritating thing it does. A portrait of a character from their own story, that their friend can match to the description without being told which one it is. An advert for a made-up cereal that has to be funny to their brother, who is a hard audience and right there.

Every one of those has a judge that is not the child’s patience: another person, a book, a real animal. That is what forces the second attempt and the eighth. "Make a picture of a dragon" has no judge, so the first dragon wins by default.

A parent-set standard produces compliance, which looks like making and is not. The child iterates until you are satisfied and learns your taste instead of finding their own.

The move is to get them to say what they are going for before they start, out loud and specific enough to be wrong. "Sad but not pathetic." "Like the cover of the book, but not a copy." Once that exists, they can tell whether a result missed, and you are relegated to an audience, which is the correct role.

It is also the moment the tool stops being the interesting part. A child comparing a result against something they already pictured is doing the same thing an illustrator does, with a faster hand.

They get annoyed, genuinely. It is a mild, productive irritation that the thing in their head is not on the screen yet. Nobody is irritated by a result they had no expectations about.

They start narrowing rather than restarting. Early attempts are usually a completely new prompt each time; later ones change one variable and keep the rest, which means they have worked out what each part of the description is doing.

And they can tell you about a version that did not survive. A child who remembers the one from four attempts ago that was nearly right has been paying attention the whole way through, which is the thing you were actually hoping for.