How to teach kids to prompt AI (without letting it do the thinking)

The only distinction that matters

"Write me a paragraph about the water cycle" and "here's my paragraph about the water cycle — what did I leave out?" use the same tool for opposite purposes. The first outsources the work. The second uses the machine as a demanding reader.

Kids default to the first because it's shorter and it works. Nobody teaches them the second, because most adults are still using the first themselves.

Move one: show your work first

The rule is that your child's own attempt goes into the box before anything comes out of it. Even a bad attempt. "Here's my thesis, here's my three reasons — which reason is weakest?" produces feedback on their thinking rather than a substitute for it.

This one rule does most of the work. It also makes the disclosure conversation trivial: a chat log that starts with your child's own paragraph is self-evidently their work.

Imagine an eligible student is allowed feedback on a paragraph arguing for a community garden. Before requesting help, they write their claim and choose two reasons. A useful request asks which reason needs clearer evidence, rather than asking for a better paragraph. When feedback arrives, compare it with the original goal. The student chooses what to revise and supplies the actual evidence; a polished alternative isn't a substitute for those choices.

Suppose the suggestion recommends adding a statistic that the student has not researched. That isn't evidence yet. Ask where they would find a reliable source and whether the assignment allows that kind of research. If they cannot verify the statistic, leave it out. A good-looking paragraph can still contain an unsupported claim, so the prompt should not become a route around checking the material.

For a math example where explanations are permitted, imagine the student writes their own setup for a ratio problem and marks the step they don't understand. The request can be about why the units should match, using a separate practice example. Close the tool before the next independent attempt. Keep both attempts so you can ask what changed. The useful output is an explanation the student can apply, not another completed line to copy.

Move two: ask for questions, not answers

Teach the phrase "ask me three questions about this" as a reflex. It flips the direction of the exchange — the AI becomes the one demanding evidence, and your child does the retrieval.

• **Ages 6–9:** "Ask me three questions about my drawing." "Pretend you're five — did my explanation make sense?"
• **Ages 10–13:** "Here's my answer. What's one thing I got wrong and how would I check?" "Quiz me on this chapter, one question at a time."
• **Ages 14–18:** "Steelman the opposite of my argument." "Find the weakest evidence in this paragraph and tell me what would strengthen it."

Suppose your child wants to practice a topic they already enjoy, such as how a bicycle's gears work. An adult can handle an eligible account to request a few general practice questions, check them, and ask them offline. If a question is beyond the child's knowledge, set it aside or consult a reliable explanation. You don't have to accept every question simply because it was generated for practice.

Imagine the first question is too broad: 'Explain everything about gears.' Ask for a smaller question about what happens when changing gear on a hill, then discuss it with the child. The prompting lesson is to describe the kind of practice needed. Asking for one manageable question helps you see the missing idea more clearly than asking for a long list whose difficulty you haven't reviewed.

For a school topic, suppose the teacher permits quiz practice but not generated notes. Use the child's own notes to identify the topic and keep the boundary explicit. After the offline questions, have the child correct one gap from the class material. A practice session is useful when it leads to something they can explain afterward. It doesn't need to continue until every generated question has been answered.

Move three: make it argue back

The most useful habit an AI-literate teenager has is treating output as a draft by someone who is confidently wrong about one thing in five. Teach "what would someone who disagrees say?" and "what's your source for that?" as automatic follow-ups.

This is also the fastest way to demonstrate hallucination in the wild. Ask for citations on an obscure claim, then check one together. Nothing teaches skepticism like watching a made-up page number.

Practicing when there's no homework

Prompting is a habit, and habits don't form during high-stakes assignments. Run it on low-stakes things: planning a birthday party, arguing about the best pizza topping, writing a fake news headline and then fact-checking it. Our post on how to teach kids about AI has more of these, and your child is already using AI covers the literacy foundation underneath them.

Imagine a family activity planning an imaginary pet enclosure. Start with a made-up animal and a few constraints, rather than using AI advice to care for a real pet. Ask which details a suggestion overlooked and what questions you would need answered. This is practice in describing a problem and examining a response, not a substitute for expert animal-care guidance.

Suppose everyone proposes a different request for the same pretend project. Compare the outputs for relevance, not just length or enthusiasm. One may follow the size limit but ignore the available materials; another may be creative but impossible to build. Ask which request communicated the constraints best, then test a clearer version. You can finish after one comparison. The point is learning to steer and check a tool, not spending the whole afternoon perfecting a prompt.

How to tell whether it's working

You'll know the habit has stuck when your child complains that the AI got something wrong. That's the whole outcome: not a kid who avoids the tool, and not a kid who trusts it — a kid who argues with it. If you want a read on whether a specific piece of homework shows their thinking or the model's, paste it into the Homework Check on our homepage.