Here's the thing: the largest OECD PISA study ever conducted on student AI use has landed, and the headline finding is not what the edtech industry ordered.
The Programme for International Student Assessment drew from over 760,000 students in 91 countries, testing 15-year-olds in science, math, and reading. It is the first PISA cycle conducted since AI use went mainstream. After adjusting for socioeconomic status, students who never use AI generally outperformed those who do in science.
Not all AI use is equally damaging. Students who used AI to summarize texts they were supposed to read, or to draft writing assignments, performed relatively worse. Those who used it for preliminary research or broadly to 'help me learn' saw a smaller performance drop. Frequency matters too: daily users and once-or-twice-a-year users both landed at the bottom, with monthly and weekly users faring better.
There is a narrow path to a positive outcome. Students who used AI weekly for general learning and were also frequently asked to critically assess the quality of AI-generated information outperformed non-users. The study authors describe this as evidence that 'moderate and intentional use of AI for schoolwork and learning' could have positive effects — but the operative word is intentional, not just frequent.
OECD director of learning and skills Andreas Schleicher put it plainly: 'In the same way that we do not become fit by watching sports but by doing sports, learning does not occur through the consumption of content, but as a productive cognitive struggle of the mind with new material.'
The distribution of AI use is telling. Students from relatively advantaged backgrounds were more likely to use AI tools — reflecting better access to hardware, stable internet, and paid subscriptions. Internationally, usage varied sharply: over 95 percent of Vietnamese students reported using AI tools, compared to just 60 percent in Japan. Daily AI users did score higher on one metric — curiosity — but that curiosity is not yet converting into better grades.
Policymakers are already moving. According to Semafor, 22 states and Washington DC have mandated bell-to-bell phone bans, and New York City and Los Angeles public schools are implementing restrictions on AI use affecting more than one million students combined.
Voltage's read: The demo is not the product, and 'AI in the classroom' has been one long demo. The data here does not indict the technology outright — it indicts passive consumption dressed up as learning. A student who outsources the cognitive struggle to a language model is not studying; they are watching someone else do push-ups. The finding that critical-assessment training flips the outcome is the most important number in the report, and it points toward a solution that requires actual teaching, not just a subscription. Schools that hand students an AI tool without the framework to interrogate it are not innovating — they are offloading accountability and calling it progress.
