OECD PISA: General AI use linked to worse scores

The OECD's latest PISA results, based on data collected in 2025, represent the first major look at student performance since AI usage became widespread. While overall AI use appears negatively correlated with science scores, the details are more nuanced: performance varied based on the specific task, frequency of use, and whether students were taught to assess AI-generated information. Andreas Schleicher, the OECD's director for learning and skills, suggests learning involves productive cognitive struggle, which technology can either enhance or undermine.

The study also highlights a digital divide, with AI use more common among advantaged students, likely tied to hardware, internet, and paid tools. The international gap is wide: over 95% of Vietnamese students reported AI use, compared to just 60% in Japan.

The path forward may involve fostering the kind of moderate, intentional use that the study hints could be beneficial. The data suggests educational outcomes may depend on embedding critical assessment skills into instruction, so that students engage with AI rather than letting it replace intellectual effort.