Students Who Use AI to Study Tend to Score Lower, Global Report Finds
A 760,000-student OECD study found that AI use generally hurt academic performance, but the picture changes when students are taught to question what the AI tells them.
Key points
- A 2025 OECD study of more than 760,000 students across 91 countries found that, after adjusting for family income, students who never use AI outperformed those who do in science.
- Students who used AI to write up readings or draft essays scored worse than those who used it for general learning or early-stage research.
- Daily users and near-never users (once or twice a year) both scored lower than students who used AI monthly or weekly.
- Students taught to critically evaluate AI-generated answers showed the strongest results, with all regular AI users in that group outperforming non-users.
- AI use was more common among students from wealthier backgrounds, ranging from over 95 percent in Vietnam to roughly 60 percent in Japan.
A major international education report has landed a finding that will unsettle anyone who assumed AI tutors were a straightforward win for learners.
The PISA study (Programme for International Student Assessment), a recurring global test of 15-year-olds in science and maths, published its first results since AI went mainstream. The data, collected in 2025 from more than 760,000 students across 91 countries, shows that students who never use AI generally scored higher in science than those who do.
Our earlier look at how one school moved from banning AI to teaching with it now reads like a preview of what this data confirms at scale.
Does how students use AI change anything?
The type of use matters as much as whether students use AI at all.
Students who asked AI tools to write up texts they were supposed to read, or to draft written work for them, saw the steepest performance drops. Those who used AI for early-stage research or general "help me learn" purposes fared better, though they still tended to trail non-users.
Frequency also played a role. Daily users scored poorly, and so did students who touched AI only once or twice a year. Weekly or monthly use produced the best outcomes among AI users.
One finding stands out. Students who used AI weekly for general learning actually outperformed non-users slightly. That gap widened further among students whose schools regularly asked them to judge the quality of AI-generated information. In that group, all regular AI users for general learning beat non-users. The study's authors read that as evidence that "moderate and intentional use of AI for schoolwork and learning" can have real benefits.
Andreas Schleicher, the OECD's director of learning and skills, put it plainly: "In the same way that we do not become fit by watching sports but by doing sports, learning does not occur through the consumption of content, but as a productive cognitive struggle of the mind with new material."
What does this mean for ordinary students and parents?
The practical takeaway is not "ban AI" but "use it differently."
When a student gets AI to write an essay or read a text on their behalf, the mental work that builds understanding never happens. When a student uses AI to explore a topic and then has to judge whether the answer is actually correct, something closer to real learning takes place. That distinction is the whole ballgame.
The report also flags an equity concern. AI use was more common among students from better-off households, likely because of easier access to hardware and paid AI tools. Students in Vietnam reported the highest AI use, at over 95 percent, while Japan sat at around 60 percent.
One small bright spot: curiosity levels were highest among daily AI users. That trait just isn't yet showing up in test scores.
The findings were first reported by The Verge AI.
Common questions
Should parents stop their children using AI for homework?
Not necessarily. The research suggests the problem is passive use, getting AI to do the work instead of the student. Using AI as a starting point and then checking whether its answers hold up appears to be genuinely helpful.
Why did daily AI users score worse than weekly users?
The study doesn't give a definitive answer, but the pattern fits the idea of over-reliance: students who reach for AI every day may be skipping the mental effort that builds lasting knowledge.
Does this apply to older students and adults too?
The PISA data covers only 15-year-olds, so direct comparisons are limited. The underlying principle, that passive consumption of AI answers is less useful than active engagement with material, is likely to apply more broadly.
