A new OECD report using data from the 2025 PISA study has found that students who use AI to help with schoolwork generally perform worse in science, maths and reading than those who don't, even after adjusting for socioeconomic background. However, the relationship is nuanced: certain patterns of AI use, particularly moderate, intentional use paired with critical evaluation of AI outputs, were linked to better rather than worse performance, suggesting the technology's impact depends heavily on how it is used rather than whether it is used at all.

The study, drawing on responses from over 760,000 15-year-olds across 91 countries, found that using AI to summarise texts or draft assignments correlated with worse outcomes, while using it for research or general "help me learn" purposes had a smaller negative effect. Frequency mattered too, with very occasional and daily users faring worst, while weekly or monthly users did better; those who used AI weekly for general learning support and were also trained to assess AI output quality actually outperformed non-users. The report also found AI use was more common among advantaged students, likely due to better access to devices and paid tools, and varied hugely by country, from over 95% of students in Vietnam to just 60% in Japan.