Moderate AI use in schools shows better academic performance, OECD finds
Students using AI once or twice a week perform similarly to non-users, study reveals.
1The Goldilocks effect of AI in classrooms
OECD's PISA study reveals a 'Goldilocks zone' for AI use in education. Students who use AI moderately (once a month to twice weekly) show better science scores than both heavy and rare users. This suggests an optimal frequency exists where AI enhances rather than hinders learning. The effect is most pronounced when AI is used for summarizing readings or preliminary research.
2Performance comparison by usage frequency
Weekly AI users perform similarly to non-users after accounting for socioeconomic factors. However, daily users and rare users both show lower scores. About 86% of OECD students have used AI for schoolwork, with 14% never using it. Interestingly, non-users generally outperform users, highlighting AI's double-edged nature in education.
3The risks of over-reliance on AI
One in five students use AI almost daily, correlating with lower science scores. Heavy reliance creates a feedback loop: weaker readers struggle to identify AI errors, while routine AI use reduces skill-building opportunities. This mirrors concerns about declining critical thinking skills, particularly in evaluating information and synthesizing multiple sources.
4Declining fundamental skills
PISA reports troubling drops in reading, math, and science scores since 2003. The share of 'hasty readers' - students who skim texts and give quick, incorrect answers - nearly doubled from 2018 to 2025. These are precisely the skills most needed in an AI-dominated information landscape, where discerning signal from noise is crucial.
5The cognitive struggle principle
OECD emphasizes that learning occurs through cognitive struggle, not passive content consumption. Like physical fitness requiring exercise, mental fitness demands active engagement. AI that performs reading or reasoning for students deprives them of this essential mental workout, potentially stunting long-term cognitive development.
6Subject-specific impacts
Science performance shows the clearest correlation with moderate AI use. The advantage disappears in other subjects when usage becomes daily, suggesting different disciplines may require tailored AI integration strategies. This challenges the one-size-fits-all approach to educational technology adoption.
7Global adoption patterns
Across 38 OECD countries, AI usage varies significantly by region and socioeconomic status. The study accounts for these factors when comparing performance, revealing that moderate use benefits persist across demographics. This suggests universal principles for responsible AI integration in education systems worldwide.
8Preparing for an AI future
The findings advocate for teaching students to use AI as a tool rather than a crutch. With 86% already using AI for schoolwork, the focus should shift to developing critical evaluation skills. Educators must balance AI's efficiency with preserving fundamental cognitive processes that underpin lifelong learning and adaptability.
Original article: Fortune — Moderate AI use in schools shows better academic performance, OECD finds