Students who rely on artificial intelligence to help them study generally score lower at school than peers who don't use AI, according to new data from the Organisation for Economic Co-operation and Development. The findings come from the OECD's Programme for International Student Assessment (PISA), which this year marks the first global education survey conducted since AI tools became widely adopted. The report reveals a nuanced picture: while most AI use correlates with weaker performance, certain approaches to these tools can offer students modest academic gains.
The PISA study drew on assessments of more than 760,000 fifteen-year-olds across 91 countries, testing their abilities in science, mathematics, and reading using data gathered in 2025. After controlling for socioeconomic factors, students who never touched AI tools generally outscored those who did in science. The performance gap varied significantly based on how students deployed the technology: those who turned to AI for tasks like summarizing assigned texts or drafting written work showed relatively weaker results, while students who used it for preliminary research or more general learning support experienced smaller performance drops. How often students used AI also mattered—both daily users and those who used it just once or twice yearly posted the poorest results, with monthly or weekly users performing better. International patterns differed sharply, with over 95 percent of Vietnamese students reporting AI tool use compared to only 60 percent in Japan. The data also showed AI adoption skewed toward students from more advantaged backgrounds, likely reflecting better access to hardware, reliable internet connections, and subscription-based AI services.
According to OECD director of learning and skills Andreas Schleicher, the mechanism behind these outcomes resembles the difference between watching sports and playing them. "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," Schleicher wrote. The report's authors suggest that "moderate and intentional use of AI for schoolwork and learning" could yield positive effects. One bright spot emerged for weekly AI users who deployed the tools broadly to support their learning—these students outperformed non-users, though the advantage remained small. That edge grew more pronounced among students whose teachers regularly asked them to evaluate the quality of AI-generated information, with all regular AI users for general learning help surpassing non-users when that critical assessment training was present.
The report identifies where technology helps versus hinders, drawing a line between tools that engage students' cognitive effort and those that bypass it. When AI enables or deepens the mental work of grappling with new material, students advance; when it short-circuits that productive struggle, it undercuts their development. The findings point to teaching students to critically assess AI outputs as a key factor that transforms these tools from performance drags into modest assets. The study also noted that AI users displayed higher curiosity levels that peaked among daily users, though for now that heightened curiosity isn't translating into improved grades. The takeaway is clear: AI's educational value hinges not on whether students use it, but on how they're taught to engage with it—schools that train learners to question and evaluate AI-generated content may unlock benefits that raw adoption alone won't deliver. The challenge for educators will be distinguishing between automation that substitutes for learning and support that genuinely scaffolds it, a balance that requires deliberate instructional design rather than leaving students to figure out effective AI use on their own.