"Students Can't Reason": What AI Is Doing to Our Children's Minds

Let's start with a real study, told simply.

In Turkey, researchers worked with nearly 1,000 secondary school students learning maths. Some students got a ChatGPT helper while they practised. Others got nothing. Then everyone sat the same test, and nobody had AI.

While they practised, the ChatGPT group did 48% better. But on the test, with the AI taken away, they scored 17% lower than students who never had AI at all.

Here's what that looks like in marks, by our sum. Say a child who practises without AI scores 40. A child practising with ChatGPT scores about 59. Looks great, doesn't it? But on test day, when the first child scores 40 again, the ChatGPT child scores about 33.

The AI didn't teach them. It did the work for them. And on test day, it wasn't there.

That study came out in 2025 in PNAS, one of the world's top science journals. It's one of many now asking the same worried question: what is AI doing to our children's minds?

A teacher's warning

In January 2026, the Brookings Institution in America published a big report on children and AI. Its team spoke to more than 500 people in 50 countries and looked at over 400 studies. Fortune magazine quoted one of the teachers:

"Students can't reason. They can't think. They can't solve problems." A teacher in the Brookings study, via Fortune, January 2026

Strong words. And Fortune summed up the report's view like this: for now, the risks of AI for students are bigger than the benefits.

Is that teacher exaggerating? Maybe a little. But the studies below suggest she's onto something.

"Cognitive stunting": what it means

Rebecca Winthrop leads the education centre at Brookings. In May 2026, she gave the problem a name: "cognitive stunting".

"Cognitive" just means thinking. "Stunting" is the word doctors use when a child doesn't grow as tall as they should. So it means a child's thinking doesn't grow the way it should.

Her point is simple. When an adult leans on AI, they may get rusty at skills they already have. When a child leans on AI, they may never build those skills at all. She put it like this:

"Cognitive development, like learning to ride a bike, cannot be outsourced." Rebecca Winthrop, Brookings, May 2026

Think about it. Nobody can learn to ride a bicycle for your child. You can hold the seat. But your child has to wobble, fall and try again. Thinking works the same way.

What brain scans show, and what they don't

In late 2025, scientists at the Technion, a university in Israel, put children into a brain scanner while they chatted with ChatGPT for five minutes. They did the same with adults.

The parts of the brain that help us stay focused and in control were less connected in the children than in the adults. The lead scientist, Professor Tzipi Horowitz-Kraus, said young children shouldn't be left alone with AI tools. She was also reported as saying six is probably too young to start.

Let's be honest about this study, though. Only 15 children took part, all around six years old. And there was no group of children who didn't use AI to compare with. So it can't prove that ChatGPT caused anything. It's an early warning light, not proof.

The thinking muscle

In 2025, a researcher in Switzerland, Michael Gerlich, studied 666 people. Those who used AI tools the most scored lower on critical thinking. Critical thinking is the habit of asking, "Is this true? Does this make sense?"

Why? The heavy AI users tended to hand their thinking over to the tool. Scientists call this "cognitive offloading". You and I would call it letting someone else do your homework.

Two things to keep in mind. These were adults, not children. And the study shows a link, not a cause. But the younger people in it leaned on AI the most, and they scored the lowest.

It's happening in Nigeria too

This isn't only an American or European problem.

In February 2025, Nigerian lecturers spoke to Nairametrics about it. Professor Isaac Nwaogwugwu of UNILAG said about 40% of his 100-plus MBA students handed in the same AI-written answers. That's about 2 in every 5 students, with the same words. Dr Felix Echekoba of UNIZIK said students who copy from ChatGPT can't answer basic questions about what they submitted.

And Nigerians love AI. A Google and Ipsos survey reported in January 2026 found that 88% of Nigerian adults have used an AI chatbot. The world average was 62%. By our sum, a Nigerian adult is about 1.4 times as likely to have used one as the average adult worldwide. And 91% of Nigerian adults said AI helps them learn.

That's good news and a warning at the same time. Our children are growing up in one of the most AI-keen countries on earth. Those chatbots are on the same phones our children borrow.

One more sum. DataReportal's 2026 figures show that children aged 5 to 17 make up about a third (33.1%) of Nigeria's people. By our count, that's roughly 79 million children. If we don't guide them on AI, who will?

The hidden danger: feelings

The thinking problem is serious. But there's a second problem that worries doctors even more.

In June 2026, the medical journal JAMA Pediatrics published a US survey of 1,009 young people aged 12 to 21. About 1 in 5 (19%) had asked an AI chatbot for advice when they felt sad, angry, nervous or stressed. And 63% of them had told no one. Not a parent, not a friend, not a doctor.

By our sum, that's about 1 in every 8 young people in that age group quietly talking to a chatbot about their feelings, with no adult knowing.

That might be fine if chatbots were good at this. Are they?

In August 2025, the journal Scientific Reports tested 29 chatbots to see how they would answer someone thinking about suicide. Under the researchers' strict rules, not one of the 29 gave a good enough answer. Under looser rules, by our count, 15 were only "borderline" and 14 still failed.

In Britain, the charity Internet Matters surveyed 1,000 children aged 9 to 17 in 2025. Among children who were already struggling, such as those with special needs or health problems, 71% used chatbots. Half of these children said it felt like talking to a friend. And 23% said they used chatbots because they had no one else to talk to.

Let that sink in. The children who most need a real person are the ones most likely to turn to a machine.

Are parents talking about it?

Not enough. In the same Internet Matters survey, 78% of parents had talked to their child about AI. But only 34% had talked about how to check whether what AI says is true.

By our sum, for every 9 parents who talk about AI, only about 4 talk about checking it. That second talk is the one that matters most.

So, is AI bad for children?

Not exactly. Remember the Turkey study? There was a third group we haven't told you about.

They used an AI tutor built to give hints, not answers. In practice, they did 127% better. And on the test without AI, they did about as well as students who never had AI. No harm done.

In our 40-mark example, by our sum, that's about 91 in practice and about 40 on the test. Same as the child with no AI, but with a lot more practice along the way.

So the problem isn't AI itself. The problem is AI that does the thinking for the child. AI that makes a child think harder is a very different thing.

Our next article is all about that: the right way for children to use AI, age by age.

What you can do this week

1. Ask your child to show you which AI apps they use. Don't shout. Just be curious.
2. Ask one question after homework: "Explain this to me in your own words." If they can't, the AI probably did the work.
3. Keep feelings human. Tell your child: "When you feel sad or worried, come to me, a teacher, or an adult you trust. Not a chatbot."
4. Talk about checking. AI can be wrong and still sound very sure. Teach your child to ask, "How do I know this is true?"

If you'd like to see how your child is really doing, not just what they hand in, Scholarleveling gives parents a code to see their child's progress and teacher remarks every week.

Read next: the right way for children to use AI, or how teachers can spot AI-written assignments.