The teach-back test: if your kid can’t explain it, the AI learned it — not them

Try this tonight, with a topic your child “learned” from an AI sometime this week. Close the app, the tab, the laptop. Then say:

“So — explain it to me.”

If what comes back is a confident overview with reasons in some kind of order, genuine learning happened somewhere in that session. If what comes back is fragments, a grin, and “I don’t know, it just said it,” you’ve learned something important too — and it cost you sixty seconds.

This isn’t an accusation of laziness. AI answers feel like understanding. They arrive organised, complete, and in exactly the order a good explanation should be. Reading one produces the warm sense of “oh, I get it now” — the same sense you get reading a solution to a puzzle you couldn’t solve, right up until someone hands you a slightly different puzzle. Psychologists call it the illusion of explanatory depth; parents can just call it the copy-paste problem. The material passed through your child’s eyes without ever being rebuilt by their brain.

The teach-back test is how you tell the difference — and, repeated gently, it’s also how you fix it.

The ritual (3 minutes, four moves)

1. Close the app. This is the non-negotiable step. Explaining alongside the transcript tests reading aloud; explaining without it tests knowing. (Bonus: kids cheat less when the rule applies to the device, not to them. “We close it every time” is a house rule, not a suspicion.)

2. “Explain it to me.” Not “repeat what it said” — explain. Bonus points for the harder version: “explain it like I’m ten.” Simplification forces real understanding; nobody can dumb down an idea they only half-hold.

3. Ask the only follow-up that matters: “Where did you get stuck?” Not “did you understand it?” (always yes), but where the explanation wobbles. The wobble point is gold: it’s the exact seam between what the AI delivered and what your child owns.

4. Reopen the app — together — and ask it about exactly that spot. “We got confused about why the moon doesn’t fall down.” Now the AI session that follows is targeted, driven by a real gap your child noticed themselves. Then close it again and let them patch the explanation. Two minutes, and the AI has been demoted from author to tutor — which is the only job it should have.

What each stage tells you

You don’t need a dashboard to grade this. Borrow the scale the newest kids’ AI apps are using — emerging, developing, excelling — and translate it into parent language: needs a prompt → does it with a nudge → does it unprompted.

• Recall: Can they start at all? Emerging = can’t begin without the transcript. Excelling = launches in, chooses their own starting point.
• Sequencing: Does the explanation have a spine — first this, so then that? Nudge-level kids get there with “and then what happened?” Asked twice a week, that question quietly stops being necessary.
• Gap-spotting: The real prize. A child who says “the bit I don’t get is why…” has moved from consuming explanations to monitoring their own understanding. That’s the skill every teacher wishes for, and it only appears when someone regularly asks “where did you get stuck?”

None of this needs recording, scoring, or reporting. You’re not building a gradebook; you’re taking a temperature.

Make it fit your child’s age

Ages 5–8: Keep it playful and short. “Teach Teddy how rain works.” A stuffed animal is a forgiving student, and a child performing for a toy forgets to perform for you. Sixty seconds is plenty; the wobbles are adorable and informative in equal measure.

Ages 9–12: Give it an audience upgrade. Teach a younger sibling (real or borrowed), or record a 60-second voice note “for the family chat.” Kids this age rise to being the expert; half the learning happens when they hear their own explanation played back and wince.

Teens: Skip the cute framing — they’ll smell it instantly. Instead, make the teach-back a critique: “Explain the topic, then tell me where the AI was weakest. What would you check?” This respects their competence, and it layers verification on top of recall — the two skills from our earlier piece on teaching kids to question AI working as one habit.

When teach-back fails (and what to do instead)

The word-perfect parrot. Some kids can recite an AI answer almost verbatim — and still not understand it. The tell: they can’t survive a changed example. Ask, “Okay, what if it rains harder?” or “Why isn’t it the opposite?” If the explanation collapses, it was memorised, not learned. That’s not failure by the way — that’s the test working.

It turns into an exam. If your child starts dreading “explain it to me,” you’ve accidentally built a quiz with a parent as examiner. Shrink it: one topic, twice a week, keep your face curious rather than evaluative, and be visibly delighted by the stuck points — those are the interesting bits, not the shameful ones.

Some nights, let it go. Not every AI interaction needs auditing. Looking up a spell score, chatting about dinosaurs, settling a family argument about the tallest building — that’s life, not coursework. Teach-back is for the sessions that claim to produce learning: homework help, exam prep, the “I studied with AI” afternoons. Judge those, let the rest be fun.

One session tells you little. A month tells you a lot.

Run the ritual once and you get a snapshot — useful, but noisy. Run it twice a week for a month and you get something better: a trend. The explanations get longer, the stuck points get more specific, the nudges get fewer. You’ll see emerging become developing become doing-it-without-being-asked — in your kitchen, not in a vendor’s gradebook.

That’s the subject of the final piece in this series: what “excelling” should actually look like at home, and how to spot real progress over a term without outsourcing the report card to the app that sold you the subscription.

Missed the start of the series? Post 1 is a five-check rundown for judging any kids’ AI tool before it settles into your home.