Science1 publisher3 min readPublished
Four in five studies in a 173-paper review of AI in education recruited university students
Hilde Storrosaeter and colleagues read the generative AI education literature from 2022 to spring 2025, and most of what they found was students in higher education describing their own thinking in questionnaires, while 39% of Norwegian 11- and 12-year-olds already use AI.
The Scientist · Science desk
What happened
- Hilde Storrosaeter of NTNU's Department of Teacher Education analysed 173 international research articles on generative AI use in schoolwork and education, with colleagues from Czech universities.
- In 80% of those studies, the participants were university or college students, which is the first of three knowledge gaps the reviewers name.
- The Norwegian Media Authority's report on 9- to 18-year-olds finds that 39% of Norwegian 11- and 12-year-olds use AI.
Compiled by The ScientistSomething wrong?How this is made
Why it matters
- constraint A head teacher asking what a chatbot does to a 10-year-old's independent problem-solving cannot get the answer from this base, because the measured population is roughly a decade older and further along in the skill being tested.
- decision A district putting chatbots in front of Year 5 now has two options: wait for age-matched evidence that does not yet exist, or measure its own pupils' work before and after and accept a single-school sample.
- exposure As children's questions move out of search boxes and public youth channels and into chatbots, the traffic adults used to count disappears from view, and so does the record of what children were asking.
- precedent The review is framed as input to pedagogy and policy, so any authority that acts on it in the next year will be extrapolating from higher-education participants down to primary classrooms.
Four in five of the 173 articles drew their participants from universities and colleges [5]. In whole numbers, that leaves about 35 papers for everyone else [1]. Those 35 carry whatever is known about school-age pupils, and the reviewers add that the research is unevenly distributed geographically [7].
Most of the articles rest on self-reporting, which the reviewers rate as less reliable than objective study of change over time [6]. A pupil who says a chatbot helped them think is reporting a belief about their own process. Two pupils reporting identical help can finish the year with different unaided skill, and a questionnaire will not separate them. "It is striking how little solid knowledge we have about how AI use affects basic skills in younger pupils," Storrosaeter said [11].
The review's headline result is about conditions of use. "The most important finding is that AI in and of itself does not make us any better or worse at thinking," Storrosaeter said. "We found a clear dual effect. The same technology can do both" [8]. The supportive pattern involves using the tool to challenge, ask questions, compare different viewpoints and build one's own line of reasoning [9]. The other pattern is the shortcut: "when AI becomes a shortcut to a ready-made answer, the chance of overdependence, more superficial thinking and weakened independent problem-solving increases," she said [10]. The description is qualitative, and no effect size for either direction appears in the account of the review [4].
"We can't simply assume that what applies to a 22-year-old university student also applies to a 10- or 14-year-old," Storrosaeter said [12]. Her concern is about sequence: "If AI takes over the thinking process before pupils have had time to develop their own skills for critical thinking and problem-solving, the consequences could be completely different for them than for adults" [13].
The adoption numbers are further along. The Norwegian Media Authority's survey of 9- to 18-year-olds put Google search use at 72% in 2024 and 47% in 2026 [14], a fall of 25 percentage points, or about a third of the 2024 level [2]. Visits to ung.no, Norway's information channel for 13- to 20-year-olds, went from 24 million in 2024 to 18 million in 2025 [16], a quarter of its traffic [3]. The report attributes that decline partly to young people switching to AI tools and partly to Google's AI summaries [16].
The review covers 2022 through spring 2025 [3], so a school that handed out chatbot accounts last autumn sits outside its window on both the age group and the dates. What would settle the question for a class of 10-year-olds is objective measurement of the same pupils over time. The review reports that design as the one its evidence base mostly lacks [6].
What to watch
- Whether the full paper breaks the non-university fifth down by school stage, which would show how many studies touch primary pupils at all.
- The Norwegian Media Authority's next survey, and whether the 39% figure for 11- and 12-year-olds keeps climbing while Google search use falls below 47%.
- Any before-and-after measurement of school-age writing and reasoning under chatbot access published after spring 2025, where this review stops.