The 2025 OECD Programme for International Student Assessment (PISA) report analyzed data from over 760 000 students aged 15 across 91 countries. The study indicates that students who never use AI generally outperform those who do in science, even after accounting for socioeconomic factors.
Performance outcomes vary by task and frequency. Students using AI to summarize texts or draft assignments showed lower results, while those using it for preliminary research saw smaller drops. Daily users and those using AI once or twice a year performed the worst, whereas weekly users showed better results.
Intentional use can yield positive outcomes. Students who use AI weekly for general learning purposes outperformed non-users. This advantage increases when students receive training on how to critically evaluate AI-generated information, suggesting that moderate, guided use supports academic progress.
The report highlights disparities in access, noting that AI usage is more common among students from advantaged backgrounds, likely due to better hardware and internet access. International adoption rates vary significantly, ranging from over 95 percent in Vietnam to 60 percent in Japan.
While daily AI users report the highest levels of curiosity, this does not consistently translate into higher grades. OECD director Andreas Schleicher emphasized that technology aids development only when it enhances the cognitive struggle of learning rather than bypassing it.
