AI-assisted students generally lag behind peers, unless they critically evaluate the tools

Data from the OECD's 2025 Programme for International Student Assessment, which tested 15-year-olds in reading, math and science across 91 countries, indicates that students who rely on AI tools generally achieve lower scores than those who do not, even after accounting for socioeconomic background. The negative effect is strongest for AI use in specific tasks such as text summarisation or essay drafting, whereas broader uses like preliminary research or general learning assistance produce a smaller performance drop.

Usage frequency also influences outcomes: students who use AI only once or twice a year or every day score the lowest, while those who engage weekly see modest gains, particularly when they receive training to evaluate the quality of AI output. The report suggests that “moderate and intentional use of AI for schoolwork and learning” could be beneficial. Andreas Schleicher, OECD director of learning and skills, likened active learning to physical exercise, warning that technology that shortcuts cognitive effort can hinder development. The study also noted higher AI adoption among advantaged students and large cross-national differences, with over 95 % of Vietnamese pupils reporting use versus about 60 % in Japan.