Students using AI for homework tend to score lower, OECD study finds—unless they're trained to evaluate it

A sweeping international study tracking 15-year-olds across 91 countries has found that students who reach for AI tools to help with their coursework generally earn lower grades than their peers who don't—but the picture shifts when critical thinking enters the picture.

The OECD's Programme for International Student Assessment (PISA), drawing on responses from over 760,000 students surveyed in 2025, marks the first major academic assessment to capture AI's influence after its tools reached mass adoption. The findings paint a complicated picture: certain patterns of AI use correlate with worse outcomes in science, math, and reading, while selective, purposeful use paired with analytical skills can actually yield modest gains.

Where AI use backfires#

Students who deploy AI for specific shortcuts—summarizing assigned texts or auto-generating essays—perform noticeably worse than non-users after accounting for family wealth and other socioeconomic factors. The damage isn't uniform, though. Those who tap AI for broader exploratory learning or general study support see smaller performance declines. Equally striking is the role of frequency: both daily users and those who use AI just once or twice yearly fare worse than students who use it monthly or weekly.

The training advantage#

A critical threshold emerges when students receive explicit instruction to evaluate AI output critically. Among learners taught to assess the quality of AI-generated content, regular use shifts from a liability to an asset—weekly users outperform those who shun AI tools entirely. This finding suggests that intention and metacognitive scaffolding matter far more than AI access alone.

According to OECD director of learning and skills Andreas Schleicher, the distinction mirrors learning itself: "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."

Equity gaps widen#

AI adoption is uneven globally and by socioeconomic status. Vietnamese students report the highest penetration at over 95 percent, while Japanese student usage sits at 60 percent. Students from advantaged households are more likely to have access to devices, reliable internet, and paid AI services. Paradoxically, daily users show higher curiosity levels than non-users—an intellectual appetite that hasn't yet translated into stronger grades for most.