What the research really says

Every study we cite, in plain language: what the researchers did, what they found, how strong the evidence is, and what it means for your family.

How to read this page

Three levels of evidence

We rate every study so you know how much weight to give it. We say plainly when the evidence is weak.

STRONG

Strong evidence

Controlled experiments, long-running cohort studies or reviews of many studies. Act on these with confidence.

EARLY

Early evidence

Small or not-yet-peer-reviewed studies. Worth knowing; not yet settled.

SIGNAL

A signal

Surveys, correlations and company data. They show what's happening, not what causes it.

−17%exam grades once the AI was removed

STRONG Turkey · high school maths

AI raised practice scores, then lowered exam scores

What they did: A field experiment with nearly 1,000 high-school maths students. Some practised with an ordinary ChatGPT-style tool, some with a version that gave hints rather than answers, and some with no AI.
What they found: Students with the ordinary tool scored 48% higher during practice, then 17% lower on the exam taken without it. The hint-giving version largely avoided the harm.
Honest reading: A large, well-designed experiment. It shows how AI is used matters more than whether it is used.

What to do at home: Ask your child to use AI for hints, not answers, and to explain the answer back to you.

Bastani et al., PNAS, 2025 · Original source

54%of US teens use chatbots for schoolwork

SIGNAL United States · ages 13–17

More than half of US teens use chatbots for schoolwork

What they did: A national survey of US teenagers aged 13 to 17 about how they use and view AI.
What they found: 54% use AI chatbots for schoolwork, and about 1 in 10 use them for all or most of it.
Honest reading: A survey: it tells us how common AI homework is, not whether it helps or harms.

What to do at home: Assume AI is part of homework. Talk about how it is used, rather than whether.

Pew Research Center, February 2026 · Original source

78%of children aged 10–17 have used one

SIGNAL Australia · ages 10–17

Nearly 8 in 10 Australian children have used an AI assistant

What they did: A survey of 1,950 children aged 10 to 17 about AI assistants and companions.
What they found: 78% had used an AI assistant, and 1 in 5 users did so daily or more.
Honest reading: Strong on how common use is; it does not measure effects on learning.

What to do at home: Set family habits early: daily use starts young, and early habits last.

eSafety Commissioner, 2026 · Original source

1 in 4children's Google searches are about AI

SIGNAL Worldwide

A quarter of children's Google searches are now about AI

What they did: Anonymised data from a child-safety app for January and February 2026.
What they found: AI tools made up a quarter of children's Google searches.
Honest reading: Company data from one app's users, not a representative sample. A useful signal that curiosity about AI is global.

What to do at home: Be curious with your child: ask what they look up about AI and why.

Kaspersky Safe Kids (company data), May 2026 · Original source

54participants, small preprint

EARLY United States · adults

The “ChatGPT and your brain” study, read carefully

What they did: 54 participants wrote essays with ChatGPT, with a search engine, or with nothing, while their brain activity was measured.
What they found: The ChatGPT group showed the weakest brain connectivity and struggled to quote their own essays.
Honest reading: A small study, not yet peer reviewed, with adults. An early warning light, not a verdict.

What to do at home: Let the first draft be your child's own. AI can help them improve it afterwards.

Kosmyna et al., MIT Media Lab, 2025 · Original source

666people surveyed

SIGNAL United Kingdom · ages 17+

Heavier AI reliance goes with weaker critical thinking

What they did: A survey and interviews with 666 people about their AI use and their critical thinking.
What they found: Those who leaned on AI more scored lower on critical thinking; the youngest participants were the most reliant.
Honest reading: A link, not proof that AI causes weaker thinking. Still, a pattern worth watching.

What to do at home: Ask “How do you know that's right?” whenever AI gives an answer.

Gerlich, Societies, 2025 · Original source

72%of US teens have tried an AI companion

SIGNAL United States · ages 13–17

72% of teens have tried an AI companion

What they did: A national survey of US teenagers about AI companions: chatbots designed for friendship or emotional support.
What they found: 72% had tried one, and about a third had discussed serious matters with an AI companion instead of a person. Common Sense Media advises no AI companions for under-18s.
Honest reading: A survey of what teens report. It shows how common companions are, not their long-term effects.

What to do at home: Gently ask who your child talks to when something is wrong, not just which apps they use.

Common Sense Media, 2025 · Original source

48%of parents worry most about dependence

SIGNAL United States · 2,053 parents

Parents' biggest worry isn't cheating. It's dependence.

What they did: A survey of 2,053 parents about their concerns with children's AI use.
What they found: 48% named over-dependence on AI as their top concern; 16% named cheating.
Honest reading: A survey of opinion. It tells us you're not alone.

What to do at home: Agree a few family rules together. The Family Playbook has a ready-made agreement.

NSCAF parent survey, July 2026 · Original source

1 in 3UK children enjoy reading in free time

SIGNAL United Kingdom · 114,970 responses, ages 5–18

Children's enjoyment of reading is at a 20-year low

What they did: The UK's largest annual survey of children's reading.
What they found: About 1 in 3 children aged 8–18 enjoy reading in their free time, the lowest in 20 years and down more than a third since 2005. Fewer than 1 in 5 read for fun every day.
Honest reading: A survey, and the decline began years before AI chatbots arrived in 2022. AI did not start this, but it may speed it up.

What to do at home: Protect a little reading for joy every day, with no test and no summary afterwards.

National Literacy Trust, 2025 · Original source

40%of US fourth graders read below Basic

SIGNAL United States · grade 4

4 in 10 US fourth graders read below the basic level

What they did: The US national reading assessment for fourth graders.
What they found: 40% scored below the NAEP Basic level in reading.
Honest reading: A national test result: it tells us where children are, not why.

What to do at home: Read together, at every age. Even older children benefit from being read to.

NAEP, The Nation's Report Card, 2024 · Original source

45.5%of UK 16–18s use AI to summarise texts

SIGNAL United Kingdom · 60,000+ young people, 2,908 teachers

Nearly half of older teens use AI to summarise what they read

What they did: A survey of young people and teachers about generative AI and literacy.
What they found: 45.5% of 16–18-year-olds had used AI to summarise texts. One in four young people who use AI for homework say they just copy what it gives them. Nearly half of teachers think AI might reduce how much pupils value reading skills.
Honest reading: Surveys, not experiments: no proof yet that AI causes worse reading, but summaries are replacing some reading.

What to do at home: Read it first, summarise it after. The book before the summary.

National Literacy Trust, August 2025 · Original source

54studies, 171,055 readers

STRONG Review of studies worldwide

Paper beats screens for deep reading

What they did: Meta-analyses combining dozens of studies that compared reading on paper with reading on screens.
What they found: Understanding was slightly better on paper, especially for factual texts and when time was short. Screen readers were also more overconfident about how much they had understood.
Honest reading: Strong, consistent evidence, though the differences are modest.

What to do at home: Use paper for the deep stuff: textbooks, revision and long reads.

Delgado et al., Educational Research Review, 2018; Clinton, 2019 · Original source

10,243young adolescents followed

STRONG United Kingdom & United States

Reading for pleasure builds the reader, and the mind

What they did: Two large studies following children over years: about 6,000 members of the 1970 British Cohort Study, and 10,243 young adolescents in the US ABCD study.
What they found: Children who read for pleasure at 10 made more progress by 16 in vocabulary, spelling and maths. Early reading for pleasure was linked to better thinking skills, better mental wellbeing and longer sleep.
Honest reading: Strong long-term studies, but observational: they show links, not proof of cause.

What to do at home: A summary gives the plot. The book builds the reader. Keep whole books in your child's life.

Sullivan & Brown, UCL, 2013; Sun et al., Psychological Medicine, 2023 · Original source

19randomised trials

STRONG Review of trials worldwide

Sharing picture books with young children grows their language

What they did: A meta-analysis of 19 randomised trials that trained parents to share picture books with their young children.
What they found: Shared reading training improved young children's language.
Honest reading: Strong evidence from randomised trials.

What to do at home: Read together and ask, “What would you have done?”

Dowdall et al., Child Development, 2020 · Original source

Every Friday

The Parents' Briefing

One new piece of research each week, rated and explained, by email and in the app. Past issues are in the Briefing archive.

Get it free Get notifications

Next page · PlaybookTake it home: the free Family Playbook