Is the AI Actually Helping Your Child Learn? Three Signals

The homework gets done faster. The answers are right. Everyone seems happier — and you cannot tell whether anything is being learned.

That is a fair thing not to be able to tell. The visible outputs of AI-assisted work look identical whether the child understood it or not, which is exactly what makes it hard. Grades will not settle it either; they lag by months and measure the output, not the learner.

Three signals do settle it, and all three are things you can check at the kitchen table this week.

Last verified: September 2026.

Signal 1 · They can explain it a day later

Not immediately after — immediately after, anyone can repeat what they just read.

The check: the next day, ask about one thing from yesterday's work. "What was that thing about volcanoes?" Casually, not as a test.

What you are listening for is not accuracy. It is whether the explanation is theirs — different words, a different order, an example they made up. A child who understood something explains it differently every time. A child who processed it explains it the same way twice, or cannot start.

If it is missing: the work was done, not learned. The fix is not less AI; it is one question added to the routine — "explain that bit to me like I'm your little brother" — while the work is still in front of them.

Signal 2 · They argue with it

The strongest of the three, and the easiest to see.

A child who is learning pushes back: "that's not right", "that's not what my teacher said", "that doesn't make sense with the other thing." A child who is not learning accepts, copies, moves on.

The check: you do not need to engineer it. Sit nearby for ten minutes while they work and count the pushbacks. Zero in ten minutes, repeatedly, is the signal.

If it is missing: they may not know pushing back is allowed. Model it once, out loud, with your own question — "hang on, is that actually true? Let's check" — and it usually appears within a week. Children copy the posture they see far faster than the rule they are told.

Signal 3 · The questions get better

The clearest sign of learning, and the one nobody looks at.

Track the shape of what they ask, not how much. Early questions are requests for output: "write me a paragraph about the Romans." Later ones are requests for understanding: "why did the Romans build roads straight?"

The check: look at the last ten things they asked — most tools keep a history. You are looking for a drift from give me to why.

If it is missing: the tool has become a vending machine. The repair is a constraint, not a ban: for one week, they may only ask questions, not request output. It sounds restrictive and children usually find it more interesting.

Reading all three together

Two things worth saying. This is not a test your child can fail — every one of the three is a habit that is taught, not a trait. And you only need to do this every few months, not weekly. It is a check on a direction, not a performance review.

What does not tell you anything

Speed. Faster work is neither good nor bad news; it is the least informative thing you can measure.

How much they use it. Time on task has never predicted learning, with any tool, and it does not start now.

Whether the answers are right. They will be right. That is the whole problem with using rightness as your signal.

What the school says about their AI use. Useful, but it is about compliance with a rule, which is a different question from whether learning happened.

You now know whether it is helping. The next question is the one that actually changes the outcome: what do you do in the twenty minutes a week you have — which habit is worth building first, and in what order?