Is AI Making Your Child a Worse Thinker? What the Research Actually Says

Is AI Making Your Child a Worse Thinker?

What the Research Actually Says

Short answer: the research doesn’t show AI makes children worse thinkers. It shows that handing over the thinking does. Kids who use AI to skip the struggle lose the reasoning practice. Kids who use it and then explain the answer back in their own words don’t. That explain-it-back step is the whole difference.

The email arrived at 4:47pm on a Friday. “We’d like to discuss the authenticity of your son’s recent assignment.” Your stomach didn’t drop because of the cheating. It dropped because, sitting at the traffic lights, you realised you didn’t actually know what your 14-year-old had been doing on his laptop each night for the past term.

You’re not failing as a parent. You’re parenting through the fastest learning-tools shift in a generation, with very little instruction. Almost 80% of Australian kids aged 10–17 are now using AI, and most are using it with no guardrails at all.

The good news: the research points to a small set of habits that protect critical thinking, and they don’t require a tech degree, a household ban, or a fight at the dinner table.

What this article gives you

A plain-English read on what the 2025–2026 research actually shows about AI and children’s thinking. The single rule you can use tonight to tell whether your child is learning or just copying. A four-step framework for AI-assisted homework. And honest answers to the question your year-coordinator email left you with: what should I be doing differently?

What the research actually says (and what it doesn’t)

The headlines are loud. “AI is rewiring kids’ brains.” “ChatGPT is making children dumb.” The research is more careful, and more useful than the panic suggests.

The three studies matter most. In a 2025 MIT Media Lab EEG study, people who wrote essays with ChatGPT showed the lowest brain engagement of any group tested. Four months later, 83% of them couldn’t quote a single sentence from the essay they’d written. The lead researcher, Dr Nataliya Kosmyna, has been careful to say this was 54 adults aged 18–39, not children, and the paper is still pre-print. But she also went on the record saying she rushed it out because “developing brains are at the highest risk.”

Then there’s the strongest causal evidence we have: a 2025 randomised controlled trial of around 1,000 students published in PNAS. Year 9–11 students practised maths in three groups, no AI, unguarded ChatGPT, and a Socratic AI tutor with guardrails. With AI on, the unguarded ChatGPT group looked like superstars: 48% better than control. Then the researchers took the AI away for the test. The unguarded group performed 17% worse than the kids who’d never used AI at all. The Socratic-tutor group performed at control level, no harm done.

Translation: AI without guardrails made the kids look smarter, then made them learn less.

The Australian academic anchor is the Lodge & Loble report from UTS and the University of Queensland, published March 2026. They name this pattern the “Performance Paradox”  students using AI look better immediately, perform worse when AI is removed. They warn it could create a “Matthew Effect for an AI-mediated world,” where strong students benefit from AI and weaker students fall further behind.

So the honest answer isn’t “AI bad.” It’s that AI without guardrails is bad, AI with guardrails can be brilliant, and the dose, design and developmental timing make the difference.

Beneficial vs detrimental offloading: the distinction every Aussie parent needs

Here’s the idea that changes the conversation. It’s called cognitive offloading, letting a tool do mental work for you. We all do it. Writing things down, using a calculator, asking Google for a postcode. Offloading isn’t new and it isn’t always bad.

Lodge and Loble draw a line down the middle of it. Beneficial offloading frees your brain up for harder thinking, the calculator handles the arithmetic so you can focus on the maths problem. Detrimental offloading bypasses the learning entirely, the AI writes the essay, and the brain that should have been doing the writing never gets the workout.

Which type a tool delivers depends on when and how it’s used. “Just like calculators” is the most common parent reframe, and it’s worth taking seriously — but ACARA, the body that writes the Australian Curriculum, has put it in writing: “Generative AI systems are not search engines… their purpose and function is very different.”

A calculator gives you a number. ChatGPT gives you an answer plus a worldview, plus a framing, plus errors that sound true. That’s not a calculator. That’s a confident, articulate ghostwriter your child is borrowing a brain from.

The risk for kids 8–16 is specific. Adults using AI to draft a work email are offloading a task they already know how to do. An 11-year-old using AI to write a book report is offloading the building of the brain that does the task in the first place. The order matters more than the tool. You can’t atrophy a thinking muscle that was never built.

Tonight’s takeaway: the question isn’t “is my child using AI?” It’s “is the AI doing the thinking, or helping them do it?”

The “Explain it back” rule and the WISE framework

This is the rule worth printing and sticking on the fridge. When your child uses AI for homework, ask them to teach you what they learned in their own words. If they can’t, the AI did the thinking, they go back.

That’s it. No software, no monitoring, no tech literacy required. It’s the question every primary teacher in Australia has been asking for forty years,  “can you explain that to me?”  applied to a new tool. If your Year 9 son hands you an essay on the causes of WWI and can’t tell you the three biggest ones without reading from the screen, the laptop wrote it.

Wrap that single test inside a four-step homework habit. WiseFamilies calls it the WISE framework:

• W — Wonder. Before opening AI, what do you think? Five minutes of brain-only effort first. (This alone turns detrimental offloading into beneficial offloading.)
• I — Interpret. What did the AI actually say, and is it correct? Cross-check one fact. AI is wrong more often than its tone suggests.
• S — Safeguard. What didn’t you share with it? No personal details, no full essays, no school login.
• E — Engage. What would you change or add? The final work has to be yours, not the bot’s.

Your job isn’t to police every prompt. It’s to make sure the loop closes, that your child engages with the material after the AI, not just before.

Tonight, this week, this year, what to do

Tonight. Run the Explain it back rule on the next piece of homework that comes home. Don’t make it a confrontation. The script in the car works:

“Quick one. I’m not about to take ChatGPT off your laptop, and I’m not making this a thing. I just want to ask you something honestly. When you use AI for an essay, is it writing it, or is it helping you think? … Right. There’s actual research now — MIT, Sydney uni — that says when AI writes, your brain checks out and you can’t even remember what you wrote a week later. When you use it to push back on your own thinking, the opposite happens. So I don’t care that you use it. I care how. Try this for one week: never let it write a sentence, but ask it to find three holes in your argument. See what happens. That’s it.”

This week. Change one prompt. The single biggest swap for an HSC, VCE or QCE-bound teen is replacing “Write me an essay on…” with “Ask me three questions about my draft.“ Same tool, opposite effect.

This week, part two. Find out what your child’s school actually does. Three questions for the year coordinator: Which AI tools are permitted in homework, if any? Is the school using NSW EduChat or another tool? How is academic integrity assessed when AI is allowed? You’ll know more than 90% of parents in five minutes.

This year. Treat AI like the calculator your teen is allowed to use in some exams and not others. The in-class assessment is the real measure of what they know. Make sure homework practice prepares them for that, not for the AI-assisted version of it.

Your child’s thinking is built, not given

The future-of-work skill isn’t AI use. It’s AI judgement, knowing when to use it, when to push back on it, and when to close the laptop and think. The kids who win the next decade aren’t the fastest copy-pasters. They’re the ones who can spot a confidently wrong answer.

You don’t need to understand transformer architecture to teach this. You need three questions and a willingness to keep asking them: Can you explain that to me? Where did that come from? What did you actually do?

That is AI literacy. And you’ve already started.

Want the full system? Download the free AI-Ready Parent Starter Kit 

Or if you would like to get a shorter version of this in your inbox once every 2 weeks when the kids let me sit at the laptop for more than 5 minutes you can sign up below.