OECD finds moderate AI use in school linked to better outcomes

A Goldilocks zone for AI in education

Students who use artificial intelligence tools moderately—roughly once or twice per week—perform about as well on standardized tests as peers who don't use AI at all, according to new findings from the OECD's Program for International Student Assessment.

The results, first reported by Fortune, suggest neither abstinence nor daily reliance produces optimal outcomes. Instead, purposeful, limited use appears to support learning without the downsides of overuse.

PISA surveyed students across OECD member countries and found that 86% had used AI chatbots for at least one schoolwork purpose. Among those who used AI to summarize assigned reading or conduct preliminary research, weekly users scored higher in science than both infrequent users and students who relied on AI almost every day.

After controlling for socioeconomic status, moderate users performed at roughly the same level as students who never used AI. But the 20% of students who reported using AI chatbots almost daily scored lower in science than non-users.

The risks of overreliance

The OECD cautioned that excessive AI use may create a feedback loop that undermines skill development. Students who routinely delegate reading, summarizing, or reasoning tasks to AI have fewer opportunities to practice those abilities. Weaker critical reading skills, in turn, make it harder to recognize when AI-generated answers are incomplete or incorrect.

The report noted a troubling decline in skills essential for navigating an AI-saturated information environment. Between 2018 and 2025, the share of "hasty readers"—students who rush through texts and give quick but incorrect answers—nearly doubled. Performance in evaluating information, making connections across sources, and thinking critically about content has also dropped.

Long-term trends predate AI

While AI adoption has accelerated, PISA data shows reading, mathematics, and science scores have been declining since 2003, indicating that AI is not the sole driver of lower performance. The organization emphasized that digital tools can support learning when used intentionally and in moderation, but unguided or excessive use contributes to distraction and shallower engagement.

"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," the OECD wrote, "but as a productive cognitive struggle of the mind with new material."

Why it matters

These findings arrive as schools worldwide grapple with how to integrate generative AI into curricula. The data suggests that blanket bans and unrestricted access both miss the mark. For education leaders and policymakers, the challenge is designing environments where AI serves as a scaffold for learning rather than a substitute for it—particularly as the technology becomes ubiquitous and students' ability to evaluate information critically continues to erode.

The OECD's analysis was published in its latest PISA report on September 11, 2026.