The OECD’s Programme for International Student Assessment (PISA) based on 2025 data presents a nuanced picture of artificial intelligence (AI) use in schools. The assessment involved more than 760,000 15‑year‑old students from 91 countries and measured knowledge in science, mathematics and reading.
Authors of the study — and summaries such as the one published by The Verge — report that students who said they used AI tools for learning generally scored lower in science after adjusting for socio‑economic status. That finding, however, does not imply that AI use is inherently harmful in all contexts.
Degree and purpose matter
The PISA results show that both how and how often AI is used influence outcomes. Students who used AI for specific task execution — for example to summarize texts or to generate an outline — performed slightly worse. By contrast, students who used AI mainly for preliminary research or as a general study aid showed smaller declines in performance.
Frequency of use also matters and does not have a linear relationship with outcomes: the weakest results were observed among students who used AI very infrequently (about 1–2 times per year) and among those who used it daily. The best average results were associated with monthly or regularly monthly AI use. Notably, students who reported weekly use of AI as a general study aid performed better than those who did not use AI at all, though the difference was modest.
Critical stance and education are key
The PISA analysis and related commentary emphasize that moderate and targeted AI use can have positive effects on school tasks and learning, particularly when students are taught to adopt a critical stance toward AI tools and to understand their limitations. In other words, the tool itself is not decisive; how it is integrated into learning processes and which skills are fostered alongside its use matter more.
Practical implications
Based on the PISA findings, schools and policy makers should consider how they integrate AI: moderate, purposeful application combined with instruction in critical digital and information literacy may be beneficial, whereas excessive or unreflective, purely occasional use is less likely to help. The results also suggest that the spread of AI tools may require new pedagogical strategies and assessment approaches.
Context and sources
PISA 2025 was the first major assessment conducted after AI tools became widely available; the findings have been summarized in outlets including The Verge and the Hungarian news site HVG. The international data analysis indicates that AI’s effect on student achievement is complex and highly context‑dependent.
