760,000 teenagers just answered the question everyone argues about: does AI help you study?
The world's biggest school test asked 15-year-olds whether they use AI. The ones who leaned on it hardest scored worse — and one small habit flipped the result.
Every few years the OECD runs PISA — a test given to 15-year-olds around the world in science, math and reading. This year's round, based on data collected in 2025, covered over 760,000 students in 91 countries. It's the first one done since AI became something everybody's kid uses.
The headline finding is blunt: after adjusting for how rich or poor the students' families are, the teenagers who never touch AI generally outscored those who do in science. But the interesting part is buried under that, and it's not the part people are shouting about.
What the data separates is not "AI users" from "non-users" — it's what they asked the AI to do. Students who used it to summarize texts they were supposed to read themselves, or to draft their writing assignments, did relatively worse. Students who used it for preliminary research, or broadly to "help me learn," barely dropped at all. Same tool, opposite outcome, depending on whether it replaced the thinking or fed it.
Frequency tells the same story from another angle. The worst scores came from two groups at opposite ends: those who used AI once or twice a year, and those who used it every day. Monthly and weekly users did better than both. Occasional use is too rare to build a skill. Daily use starts doing the work for you.
Then comes the flip. Students who said they used AI weekly for general help learning actually outperformed non-users — slightly. And among students who were frequently asked to judge the quality of what the AI produced, every regular user of that kind outperformed non-users. Being taught to check the machine is what turned a penalty into a gain.
The tool didn't decide the grade. The instruction around it did.
Anyone who is 15 now, or teaching someone who is — and anyone in their twenties who has quietly rebuilt how they study or work around a chat window. If you're using it to skip the reading, this data says that's the version that costs you. If you're using it to get started and then arguing with the output, that's the version that paid. It also lands on parents deciding what to allow at the kitchen table, and on schools that have been treating this as a yes-or-no question. PISA also found AI use was more common among students from relatively advantaged backgrounds — better hardware, stable internet, paid tools. So the gap isn't only about who uses AI. It's about who gets taught how.
The next PISA round takes data every few years, so the follow-up comparison isn't due soon and no date for it appears in the report as described. What is worth watching: whether schools start grading how students check AI answers, since that's the single variable this data links to higher scores. The report doesn't say anyone has committed to that yet.
The study found something else: curiosity levels correlate with AI use, peaking among daily users. So the kids leaning hardest on these tools may be the most curious ones in the room — and right now that curiosity isn't showing up in their grades. Nobody in this data has explained that gap.
Sources: The Verge, "Students who use AI generally score worse at school," by Dominic Preston, September 9, 2026, reporting on OECD PISA data collected in 2025.
Первое глобальное измерение после массового прихода ИИ: 760 тысяч 15-летних в 91 стране показали, что школьники, перекладывающие задания на нейросети, в среднем учатся хуже, и выигрывают только те, кого научили проверять ответы машины.
Written by THE TELL’s AI newsroom. how we work · corrections