How Kids Use AI as a Tool and Still Make It Themselves: 6 Rules

Can kids use AI as a tool and still make it themselves? Yes, but “use it as a tool” is too vague to work as a rule. A hammer is a tool. So is a contractor. The question is which one your kid hired.

The research gives surprisingly specific guidance. In a field experiment with nearly a thousand high school math students, an AI that gave answers raised practice scores and then left students 17% worse once it was taken away; an AI designed to give hints mostly avoided that damage. In a coding study, three of six ways of using AI preserved learning and three didn’t. The difference in both cases was who did the thinking. Here are six family rules built from those findings.

Key Takeaways

• In a PNAS field study (Bastani et al., 2025), students with an answer-giving GPT-4 tutor scored 17% lower once access was removed than students who never had it; a hint-giving tutor largely mitigated the harm.
• In a randomized coding study (Shen & Tamkin, 2026 preprint), people who asked AI conceptual questions or asked for explanations kept their learning; people who delegated code writing didn’t.
• AI ideas made adults’ stories more creative individually but more similar collectively (Doshi & Hauser, 2024), so the first idea should come from your kid.
• Youth programs and contests draw the same line: AI can help with brainstorming and research, but the final product and the decisions belong to the kid.
• The six rules: first version is yours; hints, not answers; explain it or lose it; you decide; say how; some projects stay AI-free.

What “using AI as a tool” actually means

Using AI as a tool means the child stays responsible for the thinking that the project exists to build, while the AI handles tasks that support that thinking: explaining, quizzing, checking, organizing or illustrating. When the AI does the thinking the project was meant to practice, it isn’t a tool anymore. It’s a substitute.

That distinction is not just philosophy. It shows up in data.

Math tutoring. Hamsa Bastani, Osbert Bastani and colleagues gave nearly a thousand high school math students access to one of two GPT-4 tutors (Proceedings of the National Academy of Sciences, 2025). “GPT Base” acted like standard ChatGPT. “GPT Tutor” was prompted to safeguard learning with teacher-designed hints. During practice, grades rose 48% with GPT Base and 127% with GPT Tutor. But when access was taken away, GPT Base students “actually perform worse than those who never had access (17% reduction in grades).” The authors say students used it “as a ‘crutch’.” The negative effects “are largely mitigated by the safeguards in GPT Tutor.”

Learning to code. Judy Hanwen Shen and Alex Tamkin, at Anthropic, ran randomized experiments on 52 developers learning a new programming library (arXiv preprint, 2026, not peer-reviewed). The AI group scored 17% lower on a follow-up quiz without being significantly faster. Inside the AI group, they found six patterns. Three scored poorly: full “AI Delegation,” “Progressive AI Reliance” (starting with questions, ending with delegation) and “Iterative AI Debugging” (letting AI fix errors without understanding them). Three scored well: “Conceptual Inquiry” (asking only conceptual questions), “Hybrid Code-Explanation” (asking for code plus explanations) and “Generation-Then-Comprehension” (generating code, then asking follow-ups to understand it). Each pattern included only two to seven people, so treat them as signals.

Creative work. Anil Doshi and Oliver Hauser (Science Advances, 2024) found AI story ideas made adults’ stories rated more creative but more similar to one another.

Every rule below comes from one of those findings.

The 6 rules

Rule 1: You make the first version. The first draft, sketch, plan or attempt comes from your kid before any AI opens. This protects originality (the Doshi and Hauser sameness effect comes from starting with AI’s ideas) and it means your kid has something to compare AI feedback against. The Story Sprout program, a youth-led writing project described by Digital Promise in 2026, works this way: kids write the story; AI only illustrates.

Rule 2: Hints and explanations, not answers. Ask the AI to explain a concept, give a hint, or quiz you. Don’t ask it to solve the problem you’re supposed to learn to solve. This is the clearest lesson from the Bastani study: same model, different instructions, very different learning. Your kid can set this up with a line like “Don’t give me the answer. Give me one hint at a time.”

Rule 3: If AI made it, you must be able to explain it. Anything AI produced that stays in the project, your kid can explain line by line or sentence by sentence. That’s the Generation-Then-Comprehension pattern, and it’s also what science fair rules require: the Regeneron ISEF table allows AI-written code only “with explicit citation stating which portions of the code were AI generated and with a log of the prompts.”

Rule 4: AI suggests; you decide. The AI can offer options. Your kid picks, rejects and changes. The Scholastic Art & Writing Awards allow AI for brainstorming, outlining and research but reject entries “for which the final product is itself AI-generated,” including work where the creator can’t “demonstrate and articulate their creative decision-making process.” Decisions are where authorship lives.

Rule 5: Say how you used it. One line at the end of the work: what you did, what the AI did, what you changed. California’s model AI policy expects students to disclose AI use when they submit work. Saying how keeps the tool visible. Our disclosure guide for citing AI in homework has formats by age.

Rule 6: Some projects stay AI-free. Keep certain things fully human: hands-on builds, first attempts at a new skill and anything your kid is making for someone they love. A two-year study of Making in a public elementary school (Schlegel et al., Computers & Education, 2019) found gains in Making self-efficacy, science self-efficacy and science identity. The confidence comes from doing it with your own hands.

How to Teach Your Kid the Tool Rules

Ages 5–8: “helper or doer?”

Young kids rarely use AI on their own, so teach the idea with any helper. Before a puzzle, a drawing or a Lego build, ask: “Do you want a helper or a doer?” A helper gives a hint (“Try the corner pieces”). A doer finishes it. Let them feel the difference: a puzzle finished by a doer doesn’t feel like theirs. When they do meet a voice assistant or AI picture tool with you, use the same words. “Let’s ask it for a hint, not the answer.”

Ages 9–12: write the tutor rules

Have your kid write a short instruction to paste at the start of any homework chat: “I’m in 6th grade. Don’t give me answers. Ask me questions and give one hint at a time. If I ask for the answer, remind me of this rule.” Then test it together on a math problem and see whether it obeys. Kids learn two things: AI behavior depends on instructions, and the tool can be shaped to serve their learning. Our article on when AI homework tools help and when they hurt has more examples.

Ages 13+: the six-pattern audit

Show your teen the six patterns from the Shen and Tamkin study. After their next AI-assisted project, have them look back at their chat history and label each exchange: delegation, progressive reliance, AI debugging, conceptual question, code-plus-explanation, or generate-then-understand. Which patterns dominated? Teens are often surprised how quickly “just one question” slid into delegation. This builds metacognition, thinking about how you think, which no chatbot can do for them.

The question to ask: “If the AI disappeared tomorrow, which part of this project could you still do yourself?”

When kids use AI as a tool: which rules matter most

Project type	Rule that matters most	AI roles that keep it the kid’s work	AI roles that take over
Math practice	2: hints, not answers	Hints, checking a finished answer, practice questions	Solving the problem
Creative story or art	1 and 4: first version, you decide	Feedback on a draft; illustrating the kid’s descriptions	Plot, characters, final images as “art”
Coding project	3: explain it or lose it	Conceptual questions; code plus explanation	Writing and debugging the code
Science fair	3 and 5	Help choosing a statistical test, logged	Research plan, abstract, conclusions (banned at ISEF)
Essay	1 and 5	Counterarguments after drafting; grammar notes	Thesis and body paragraphs
Hands-on build	6: AI-free (or reference only)	Explaining how a part works	Designing it for them
Gift or card for family	6: AI-free	None needed	Writing it

Making the rules stick at home

Post them where the computer is

Write the six rules on an index card next to the family computer. Short versions work: “Mine first. Hints only. Explain it. I decide. Say how. Some stuff is just mine.” Kids follow rules they can see.

Model the rules out loud

When you use AI, narrate. “I’m writing my own draft of this email first, then I’ll ask AI to check the tone.” Kids copy what adults do far more than what adults say.

Check process, not just product

Once a week, ask to see one AI conversation from a homework session. Not to catch anything, but to see which pattern it followed. Praise good patterns specifically: “You asked it to explain, then did it yourself. That’s exactly right.”

Adjust by age, not by trust

Younger kids need more structure because the research suggests novices need more guidance. As your kid shows they can follow the rules, loosen supervision, but keep the rules themselves. Even professional engineers follow Rule 3.

What not to do

Don’t ban AI outright and then hope. Common Sense Media found that 59% of teens who hit a school AI block switched to a personal device. Rules your kid understands and agrees with travel with them. Bans mostly move use out of sight.

What to Watch For Over the Next 3 Months

• Week 4: Your kid can recite the rules in their own words and has used the “hints, not answers” instruction at least once.
• Month 2 red flags: Homework scores going up while quiz or test scores stay flat or fall. That’s the Bastani pattern: AI helping practice, hurting learning. Check whether chats are drifting from hints to answers.
• Month 3 self-check: Can your kid do a recent assignment’s core task without AI? Do their AI chats mostly look like conceptual questions and explanations? Is at least one ongoing project fully their own?

Frequently Asked Questions

What does it mean for kids to use AI as a tool?

It means AI supports the thinking a project is meant to build instead of replacing it. AI explaining a concept, giving a hint, quizzing your kid or giving feedback on a draft are tool uses. AI solving the problem, writing the essay or designing the project are substitutions.

Is it okay for kids to use ChatGPT for homework?

It can be, depending on how. A PNAS study found an answer-giving AI tutor left students 17% worse once removed, while a hint-giving tutor mostly avoided that. Set the AI to give hints only, check your school’s rules for each assignment, and have your kid disclose how they used it.

How do I set up ChatGPT or another chatbot to give hints instead of answers?

Have your kid start each session with a clear instruction: their grade, “don’t give me the answer,” “give one hint at a time” and “ask me questions.” Test it on one problem together. Chatbots don’t always follow instructions perfectly, so your kid should still recognize when it hands over an answer.

Should some projects be completely AI-free?

Yes. Hands-on builds, first attempts at a new skill and personal gifts are good candidates. Research on Making in school found gains in kids’ self-efficacy and science identity, which come from doing the work themselves. AI-free projects also give kids a baseline to judge what AI adds.

At what age can kids start using AI tools themselves?

Many AI services set minimum ages of 13 or older in their terms, and UNESCO suggested 13 as a minimum for classroom use in 2023. Younger kids can learn the ideas, helper versus doer, with any kind of help, and use AI only alongside a parent.

---

About the author

Ricky Flores is the founder of HiWave Makers and an electrical engineer with 15+ years of experience building consumer technology at Apple, Samsung, and Texas Instruments. He writes about how kids learn to build, think, and create in a tech-saturated world. Read more at hiwavemakers.com.

---

Sources

1. Bastani, H., Bastani, O., Sungu, A., Ge, H., Kabakcı, Ö., & Mariman, R. (2025). “Generative AI without guardrails can harm learning: Evidence from high school mathematics.” Proceedings of the National Academy of Sciences, 122(26). https://doi.org/10.1073/pnas.2422633122
2. Shen, J. H., & Tamkin, A. (2026). “How AI Impacts Skill Formation.” arXiv preprint, not peer-reviewed. https://arxiv.org/abs/2601.20245
3. Doshi, A. R., & Hauser, O. P. (2024). “Generative AI enhances individual creativity but reduces the collective diversity of novel content.” Science Advances, 10(28). https://doi.org/10.1126/sciadv.adn5290
4. Schlegel, R. J., Chu, S. L., Chen, K., Deuermeyer, E., Christy, A. G., & Quek, F. (2019). “Making in the classroom: Longitudinal evidence of increases in self-efficacy and STEM possible selves over time.” Computers & Education, 142. https://doi.org/10.1016/j.compedu.2019.103637
5. Alice (with Jack and Cindy). (2026, September 10). “What Children’s Creative Writing Taught Us About Responsible AI Use.” Digital Promise. https://digitalpromise.org/2026/09/10/what-childrens-creative-writing-taught-us-about-responsible-ai-use/
6. Scholastic Art & Writing Awards. (2026). “Copyright & Plagiarism.” https://www.artandwriting.org/awards/how-to-enter/copyright-plagiarism/
7. Society for Science. (2025, October). “Use of generative AI to support a research project.” Regeneron ISEF. https://sspcdn.blob.core.windows.net/files/Documents/SEP/ISEF/2026/Rules/Generative-AI-Use-Table.pdf
8. California Department of Education. (2026). “Model Policy: Artificial Intelligence in Education.” https://www.cde.ca.gov/ci/pl/aipolicy.asp
9. Common Sense Media. (2026, August 18). “Teens in the AI Era: Schoolwork and Skills That Matter.” https://www.commonsensemedia.org/research/teens-in-the-ai-era-schoolwork-and-skills-that-matter