Students who use AI generally score worse at school
Students who use AI for studying generally perform worse academically than non-users, according to OECD PISA 2025 data from 760,000+ students across 91 countries The type and frequency of AI use significantly moderates outcomes: summarizing and drafting tasks correlate with worse performance, while preliminary research and general learning assistance show smaller or even positive effects Critical AI literacy training—specifically assessing the quality of AI-generated information—can flip the out
Analysis
TL;DR
- Students who use AI for studying generally perform worse academically than non-users, according to OECD PISA 2025 data from 760,000+ students across 91 countries
- The type and frequency of AI use significantly moderates outcomes: summarizing and drafting tasks correlate with worse performance, while preliminary research and general learning assistance show smaller or even positive effects
- Critical AI literacy training—specifically assessing the quality of AI-generated information—can flip the outcome, enabling regular AI users to outperform non-users
- Usage frequency follows a U-shaped curve: daily and rare (once/twice yearly) users perform worst, while weekly or monthly users see better outcomes
- AI access remains unequal, with higher usage among advantaged students and dramatic international variation (95% in Vietnam vs. 60% in Japan)
Why It Matters
This is the first large-scale, globally representative study to examine academic outcomes in the era of mainstream AI, providing educators and policymakers with empirical evidence rather than speculation about how AI integration affects learning. The findings challenge both uncritical adoption and outright prohibition of AI in education, pointing toward a nuanced middle ground where intentional, moderate use combined with critical evaluation skills can yield benefits.
Technical Details
- Data source: OECD PISA 2025, testing 15-year-old students in science, math, and reading across 91 countries with over 760,000 participants—the first administration since AI became mainstream
- Adjustments made for socioeconomic status to isolate AI use effects from confounding demographic factors
- AI use categorized by task type (summarizing texts, drafting assignments, preliminary research, general learning assistance), frequency (daily, weekly, monthly, rare, never), and presence of critical assessment training
- Key metric: performance differentials between AI users and non-users across subject areas, with interaction effects examined for students regularly asked to evaluate AI output quality
- Cross-national comparison of AI adoption rates revealing significant disparity, with Vietnam at 95% and Japan at 60%
Industry Insight
- Educational technology developers should prioritize building tools that support active learning and cognitive engagement rather than passive content consumption, aligning with the OECD's finding that AI should enhance—not replace—the "productive struggle" of learning
- Schools and curriculum designers should embed AI literacy and critical evaluation training into existing programs, as this single intervention was shown to reverse negative performance trends associated with AI use
- The access gap in AI tool usage between advantaged and disadvantaged students, and across countries, signals an urgent need for equitable deployment strategies to prevent AI from widening existing educational inequalities
Disclaimer: The above content is generated by AI and is for reference only.