Students who use AI generally score worse at school
| Source: The Verge AI
Tags: OECD, PISA, AI in education, EdTech, AI policy, student performance
An OECD PISA study of 760,000+ students across 91 countries finds that AI use generally correlates with lower science scores, but students who use AI weekly for general help AND are trained to critically evaluate AI outputs outperform non-users — pointing to intentional, supervised AI use as the key variable.
Details
The 2025 OECD PISA study — the first conducted after widespread AI adoption — draws on over 760,000 fifteen-year-olds across 91 countries. The headline finding is counterintuitive for AI optimists: students who use AI tools generally score worse in science than those who don't, after adjusting for socioeconomic status.\n\nThe performance gap is not uniform. Students using AI for delegated tasks (summarizing required reading, drafting assignments) scored worst. Students using it for preliminary research or general 'help me learn' saw a smaller drop. Frequency matters too: daily users and those who used AI only once or twice a year performed the worst, while weekly or monthly users fared better.\n\nThe most important exception: students who used AI weekly for general learning AND whose schools frequently asked them to evaluate the quality of AI-generated information outperformed non-users. OECD director Andreas Schleicher compared AI to watching sports — consumption without cognitive effort produces no growth.\n\nThe policy implication is clear: AI in education is not harmful by default, but deploying it without training students to critically assess its outputs may undercut the cognitive struggle that drives learning. For EdTech companies and school systems considering AI integration, this study provides the first large-scale evidence base for what structured AI use looks like in practice.