OECD Data Links Student AI Use to Lower Scores, While Critical Training Offers a Better Path

Students who use AI chatbots for schoolwork generally score lower than students who do not, according to findings from the OECD’s Programme for International Student Assessment. The result is one of the broadest international signals yet that simply adding a chatbot to studying does not guarantee better academic performance.

The pattern does not apply equally to every form of use. Students trained to assess AI responses critically received a small boost in test performance, while unexamined reliance on generated answers was associated with weaker results. The distinction puts the focus on how students use AI, rather than treating access to the tools as an educational intervention by itself.

AI can reduce the effort that produces durable learning when students use it to complete assignments instead of working through difficult ideas themselves. Generating an answer, summarizing a reading, or rewriting prose can finish a task quickly while leaving students with less practice in recall, reasoning, writing, and error correction.

That concern fits wider evidence about classroom use. A Center for Democracy and Technology report found that 85% of teachers and 86% of students used AI during the 2024-25 school year, while also linking rising use to weaker teacher-student relationships and other risks. A review of higher-education research identified digital fatigue, loneliness, technostress, and reduced academic engagement among the concerns surrounding expanded AI use.

Students gain more from AI when they treat its output as material to check, challenge, and revise. Training in evaluating accuracy appears to be the key factor separating productive use from answer outsourcing, according to the OECD findings. That approach turns a chatbot into a source of feedback or explanation while keeping the intellectual work with the student.

Educational guidance points in the same direction. The American Psychological Association recommends interactive and constructive uses that preserve active learning, including brainstorming, explanation, and assessment preparation rather than passive copying. A study from researchers at Kent State University examines AI’s effects on study habits, time management, feedback, and learning, adding to a growing research focus on the conditions under which these tools help.

AI use has already reached a large share of students, though adoption and attitudes vary. An ACT report summarized by the Lumina Foundation found that 46% of high school students surveyed had used AI tools, and 46% of those users had used them for school assignments. Students who avoid the tools often cite a lack of interest and distrust of the information they provide.

College use is higher in some surveys. The Digital Education Council found that 86% of bachelor’s, master’s, and doctoral students used AI for coursework in 2024, commonly for brainstorming, explanations, reading summaries, and assessment preparation. Those figures show why schools face a practical policy problem: restricting access alone won't address a tool already embedded in academic routines, while unrestricted use can blur the line between assistance and substitution.

The OECD findings carry weight because they draw on students across countries, but performance comparisons do not by themselves prove that AI use caused lower scores. Students who turn to chatbots more often might already be struggling, have different study habits, or attend schools with different levels of support.

A 2026 Stanford review found that only a small subset of the existing AI-in-education literature, 20 papers, provides strong causal evidence. It found no high-quality U.S. K-12 causal studies focused on students and only limited evidence involving teachers. The result is a clear warning about careless use, paired with a research base that still needs controlled studies measuring specific tools, assignments, subjects, and student groups.

Schools are likely to put more emphasis on AI literacy, in-class writing, oral explanation, and assignments that reveal a student’s reasoning. Teachers also face a verification burden as generated text becomes harder to distinguish from student work. In Canada, Manitoba’s Division Scolaire Franco-Manitobaine has responded by limiting homework in some grades and making more assignments optional, a policy shaped partly by the difficulty of identifying AI-written submissions.

The strongest educational response combines boundaries with guided practice. Students need clear rules about when AI is allowed, instruction on checking factual claims and citations, and assessments that reward independent thinking. The OECD result gives schools a measurable reason to build those habits now, while the limited causal evidence argues against blanket claims that every form of classroom AI is harmful.