Students who use AI for schoolwork generally scored lower in science than nonusers after socioeconomic adjustments, according to the OECD’s latest PISA report. The study draws on 2025 data from more than 760,000 15-year-olds across 91 countries.
The pattern depended on how and how often students used the tools. Asking AI to summarize assigned reading or draft written work was associated with worse performance, while preliminary research and general learning help showed smaller declines. Daily users and those using AI only once or twice a year performed worst; weekly or monthly use produced better results.
A more encouraging result appeared among students taught to assess the quality of AI-generated information. Frequent practice evaluating outputs was associated with regular AI users outperforming nonusers when the tool supported general learning. The data is observational, so it cannot prove that AI caused either the lower or higher scores. It does suggest schools should distinguish between tools that replace cognitive effort and teaching that uses AI while preserving critical evaluation.