Should Children Use AI for Homework?

The Short Answer

Should children use AI for homework? Yes — to practise and to understand; no — to copy answers. And the deciding factor is usually not the child's willpower but the tool's design: put an answer machine in front of a tired twelve-year-old at nine in the evening, and it will get used as an answer machine.

That sounds like a dodge, so let us be concrete. The one question — “can my child use AI for homework?” — actually contains two completely different activities that happen to share a keyboard. One builds a learner. The other builds a dependency. The rest of this essay is about telling them apart, and about why the outcome is mostly decided before your child ever types a word.

Two Activities, One Keyboard

Activity one looks like this: the child attempts the problem, gets stuck, asks the AI where the reasoning broke, and finishes the problem themselves. The effort stays with the child; the AI supplies direction. This is what a good tutor does, and there is no serious argument against it.

Activity two looks like this: the child pastes the problem in, receives the solution, copies it out, and moves to the next one. The homework is finished and nothing else happened. The teacher receives a false signal that the class is keeping up. The exam, months later, corrects the record.

From across the room, the two activities look identical: a child at a screen, homework getting done. That is exactly why parents find this question so hard — you cannot tell them apart by watching.

What the Evidence Says

Cognitive science calls it the generation effect: an answer you produce yourself sticks; an answer handed to you evaporates. Decades of research on tutoring point the same way — the learning happens inside the effort, not at the moment a correct solution is displayed.

The newest evidence is blunter. Field studies of students given unrestricted answer-giving AI found homework scores rising and exam scores falling once the tool was removed. The machine had been doing the practising, so nobody got the practice. Note carefully what this does and does not show. It is not evidence that AI harms learning. It is evidence that answer-giving harms learning — now available at scale, on demand, and free of the social awkwardness of asking a classmate to let you copy.

Design Decides Which Activity Happens

A general-purpose chatbot is optimised to complete your task, so copying is its default setting. You can ask it nicely to tutor instead — and a determined child can ask it, just as nicely, to stop. A rule that lives in a prompt is a preference. A rule that lives in the product is a property.

Every OpenKids coach works under a teaching contract that no request can talk its way around. When a question looks like homework, the coach first asks what the child has already tried, works exactly one step together, then hands the next step back. A full worked solution exists only as an earned reveal, shown after a genuine attempt. Photograph a worksheet and the coach names what the child already got right, points at exactly one next step or one error, and asks a question about it.

None of this makes copying inconvenient. It makes copying unavailable — which is a different thing, and the thing that matters at nine in the evening.

A Test Any Parent Can Run

Before letting any AI near your child's homework — ours included — run one test. Type in a real homework problem and watch what comes back. If the reply is a finished solution, you are looking at an answer machine, whatever the marketing says. If the reply is a question — what have you tried? which step went wrong? — you are looking at something built to teach.

Then watch for the second signal, a week or two in: can your child do the next problem of that kind without the screen? That is the only measure we hold our coaches to, and it is a perfectly fair one to hold over any tool. A homework helper should be judged by what happens when it is not there.