Does AI make kids dumber? What the research actually shows

Not on the evidence so far, but it can stop them learning. When students used unrestricted GPT-4 to practise, they scored 17% lower on a later exam taken without AI (Bastani et al., PNAS 2025), while AI tutors designed to make students think have matched or beaten other teaching in trials at Harvard, in UK schools and in Nigeria.

Try Ask Poppy free

Ask Poppy is voice-first homework help for kids around 7 to 10, and teens and adults too: it asks what you have tried and gives one hint at a time, so you learn how to find the answer. Free with 30 messages a day, then $3.99/week or $9.99/month (US App Store) · iPhone, iPad, Mac

Written by the Ask Poppy team, who make one of the tools mentioned here. The advice works with any AI, or none. We haven't run a controlled study of our own app.

Key papers: Bastani et al., PNAS 2025, doi:10.1073/pnas.2422633122 · Kosmyna et al., arXiv:2506.08872 · Stadler et al., doi:10.1016/j.chb.2024.108386 · Kestin et al., doi:10.1038/s41598-025-97652-6.

What the research shows

	Who was studied	What happened	Read
Bastani et al., PNAS 2025	About 1,000 high-school students, Turkey	Unrestricted GPT-4 practice: 17% lower on the later no-AI exam. A hint-only version avoided the drop but didn't beat no AI.	Summary · Paper
Kosmyna et al., arXiv 2025 (MIT)	54 adults writing essays	ChatGPT group: weakest EEG connectivity, least ownership, poor recall of own essays. Small, not peer-reviewed at release.	Summary · Paper
Stadler et al., Computers in Human Behavior 2024	91 university students	ChatGPT lowered mental effort compared with Google search, but reasoning in their answers was lower quality.	Summary · Paper
Fernandes et al., Computers in Human Behavior 2026	Adults solving LSAT logic problems	AI raised scores, but people overestimated how well they'd done: smarter, but none the wiser.	Summary · Paper
Gerlich, Societies 2025	Survey of 666 UK adults	Heavier AI use went with lower critical-thinking scores, through cognitive offloading; youngest group relied on AI most. A link, not a cause.	Paper
Kestin et al., Scientific Reports 2025	194 Harvard physics students, randomised	A scaffolded AI tutor beat an active-learning class on a short-term test.	Summary · Paper
LearnLM Team and Eedi, 2025	165 UK pupils aged 13 to 15, randomised	Tutor-supervised AI tutoring did at least as well as human tutors, and better on new problems.	Summary · Paper
De Simone et al., World Bank 2025	Secondary students in Nigeria, 12 sessions	After-school AI tutoring with teacher guidance: +0.3 standard deviations on the final assessment.	Summary · Paper

How to read this as a parent

It's about use, not exposure. The harms show up when AI does the work the child was meant to practise. The gains show up when AI makes the child do more of the thinking.
Short-term and mostly older students. Most trials last weeks, not years, and few involve primary-age children. Nobody yet knows the long-run effect of growing up with AI.
Overconfidence is the quiet risk. Several studies find people feel they understand more than they do after using AI. Asking your child to explain an answer back is a cheap check.
"Brain rot" headlines overreach. No study here shows lasting damage to a child's brain. They show skipped learning, which is real and fixable.

What to do at home

Try first: your child has a go, even a wrong one, before asking AI.
Ask for hints, not answers. Our hint ladder gives a script.
Explain it back: if they can't explain it without the screen, they don't have it yet.
No AI for the actual graded work unless the teacher allows it.
Agree the rules together, for example with house rules for AI.

The longer version of the "cognitive debt" idea is in Cognitive debt: what AI can do to kids' thinking.

Where Ask Poppy fits

We make Ask Poppy, so weigh this accordingly. It's a voice-first thinking companion app for iPhone, iPad and Mac. A child holds to talk and Poppy speaks back; it asks what they already know and offers one hint at a time rather than leading with the answer. Parents can add up to three child profiles and open each child's conversations (children are told a parent can see them). It can look things up on the web and check national curriculum goals (fullest for Norway, Sweden and Finland; core subjects for England, France, Germany and Spain). Free with 30 messages a day, then $3.99/week or $9.99/month (US App Store). Like any AI, it can make mistakes.

Sources (checked 28 Oct 2026): Bastani et al., PNAS, 2025 · Kosmyna et al., arXiv preprint, 2025 · Stadler et al., Computers in Human Behavior, 2024 · Fernandes et al., Computers in Human Behavior, 2026 · Gerlich, Societies, 2025 · Kestin et al., Scientific Reports, 2025 · LearnLM Team and Eedi, arXiv preprint, 2025 · De Simone et al., World Bank, 2025 · OpenAI Help: age requirements · Ask Poppy on the App Store. Ask Poppy facts are from the current app (version 1.8.3) and its App Store listing. US prices; they vary by country, retailer and sale. Spot something out of date? Email [email protected].

Want an AI that makes your child do the thinking? Ask Poppy is voice-first homework help, built first for kids around 7 to 10 and their parents. Your child talks the question through, and Poppy asks what they've tried and gives one hint at a time instead of handing over the answer, so they learn how to find it. Parents set up each child's profile and can see their conversations, and memory can be turned off per child. Teens and adults use it too. A one-button, screen-free Poppy device is in development.

Free with 30 messages a day, then $3.99/week or $9.99/month (US App Store prices) on iPhone, iPad and Mac. (Disclosure: this site is made by the Ask Poppy team.)

Try Ask Poppy free See the screen-free device

Questions

Does using ChatGPT make you dumber?

Not permanently, as far as anyone has shown. But using it to do the thinking for you reduces the learning that would have happened. In a PNAS study of about 1,000 high-school students, those who practised with unrestricted GPT-4 scored 17% lower on a later exam without AI than students who never had it.

Is AI brain rot real?

"Brain rot" isn't a scientific term. What researchers do measure is cognitive offloading: handing a mental task to a tool. With AI, offloading can mean less effort, weaker recall of your own work and overconfidence. The effects depend on how the AI is used, and the long-term picture isn't known yet.

Is ChatGPT bad for kids' brains?

There's no evidence it damages a child's brain. The risk is skipped practice: if ChatGPT writes the essay or solves the problem, your child doesn't build the skill. ChatGPT also requires users to be 13 or older, with parental permission under 18.

Does AI make students lazy?

It can make it easy to skip effort. Students using ChatGPT for research reported less mental effort than those using Google, but produced lower-quality reasoning. Tutoring set up to make students work, with hints rather than answers, has done better in trials.

What does the MIT "Your Brain on ChatGPT" study show?

In 54 adults writing essays, the ChatGPT group showed the weakest brain connectivity on EEG and the least sense of ownership, and many couldn't quote what they'd just written. It's small, about adults, and was a preprint when published, so treat it as an early warning rather than proof of harm.

Can AI help kids learn instead?

Yes, when it's built to make them think. A Harvard physics trial, a UK secondary-school maths trial and a World Bank study in Nigeria all found gains from AI tutoring that guided students step by step, under some supervision.

By Simon Dreyer, founder of Ask Poppy · Last updated 28 October 2026 · How we write these pages