Students are using AI in school and test scores are down, but the OECD finds that those who use it once or twice a week perform similarly to nonusers

AI in school may be doing more than one thing at once. The OECD’s latest PISA results point to lower student skills and weaker test scores overall, but they also hint at a narrow sweet spot: moderate use can line up with better performance than either very light use or heavy, near-daily use.

The clearest pattern showed up in science. Students who used AI about once a month to once or twice a week posted higher average science scores on some common assignments than students who rarely used it and those who used it almost every day. The advantage was most visible when AI helped with summarizing reading or doing early research on a new topic.

Weekly users who relied on AI for those tasks tended to score about the same as nonusers once socioeconomic status was taken into account. But the students at the extremes did worse. Those who used AI almost every day, and those who barely touched it, were both more likely to post lower scores. The OECD says that doesn’t prove AI itself is hurting performance. It may instead reflect who is using it, and how.

Still, the broader picture is not rosy. PISA has already shown declines in reading, mathematics and science since 2003. In the latest round, 86% of students across OECD countries said they had used AI chatbots for at least one of the schoolwork tasks studied, while about one in five said they used them almost every day. The report’s warning is blunt: overuse can mean more distraction, less depth, and fewer chances to build the very skills students need to judge whether an AI answer is complete, wrong, or just lazily confident.