When the homework is done but nothing was learned
Finished, correct, and completely hollow. How to spot work a child has passed through rather than done — and what to say about it.
There is a particular feeling that comes from looking at a finished worksheet and knowing something is off. Every answer is right. The handwriting is theirs. And when you ask a single follow-up question about it, there is nothing behind it at all.
This is older than AI. Copying off a friend, working backwards from the answer key and following a worked example without understanding it are all the same move. But the tools have made it frictionless, and frictionless matters, because most of the reason children used to struggle through things was that not struggling was inconvenient.
The tell is not the work, it is the second question
You will not reliably catch this by inspecting the work. Correct is correct. What you can do is ask one question that the answer does not contain.
Not 'did you do this yourself', which invites a defensive yes and teaches them that the goal is to not get caught. Something more like: which one of these was the hardest? Or: show me the one you nearly got wrong. Or, best of all, pick a single item and ask them to talk you through it as if you did not know how.
A child who did the work has something to say, even if it is halting. A child who passed it through has nothing, because there is nothing to have. The gap is unmistakable and it is not a trap — it is just the difference between having been somewhere and having seen a photograph of it.
Do not make the first conversation about cheating
The instinct is to treat this as a dishonesty problem, and sometimes it is. More often, especially with younger children, it is an avoidance problem wearing dishonesty's clothes. The work felt impossible, or boring, or they had already decided they were bad at it, and here was a door.
If the first conversation is about character, you find out nothing about why. You also guarantee the next one happens further away from you. It is more useful, and usually more accurate, to open with the difficulty rather than the shortcut: that looked like it was going to be miserable, wasn't it.
You can be clear that the work has to be theirs. You can be clear about that and still not make it the opening line.
The difference a hint makes is measurable
This is not a matter of taste. When researchers ran a large trial in maths classes, students given a plain AI assistant did worse on a later unassisted exam than students who had never used AI. Students given the same model — configured to offer hints and ask questions rather than produce answers — did no worse than the control group at all.
The children in the second group were not using less AI. They were using it differently. What the tool was willing to hand over turned out to be the whole story.
This is directly usable at home, because you control the instruction. 'Ask it to give you a hint, not the answer' is a sentence a nine-year-old can follow. So is 'ask it to tell you what you got wrong, and then fix it yourself.' So is 'ask it to make you three more like the one you missed.'
Give the struggle somewhere to go
The reason children reach for the shortcut is usually that being stuck feels like failing rather than like the ordinary middle of doing something. That is worth naming out loud, repeatedly, because nobody is born knowing it.
It also helps to make being stuck actionable. A child who is stuck and has no move will find a door. A child who has three moves — read the question again out loud, do the easiest part first, come and ask me for a hint — has somewhere to put the feeling.
And the offer has to be real. If asking you for help reliably produces a sigh, the chatbot will win, and it will deserve to.
Let some of it be bad
The hardest part of this for a lot of parents is tolerating work that is worse than the work AI would have produced. A wobbly paragraph that is theirs is worth more than a clean one that is not, and it is worth more specifically because it shows you where they actually are.
Teachers say a version of this too. What tends to disappear first when a class starts leaning on AI is not correctness. It is variety — the ideas all start arriving in the same shape.
That shape is the thing to be suspicious of. Not the mistakes.
The goal is not a child who never uses the tool. That is not the world they are going to live in, and pretending otherwise mostly just moves the use out of sight.
The goal is a child who can tell the difference between being helped and being replaced, and who would rather be helped. That is learnable, and the place it gets learned is a kitchen table with somebody asking them to explain one of the answers.