Students who use AI generally score worse at school

OECD PISA data collected in 2025 indicate that students using AI for study generally perform worse than peers who do not use it. The effect is nuanced: certain AI applications provide a slight benefit, particularly when students are taught to critically evaluate the tools' output. The findings could increase scrutiny of how AI is deployed in education, but are unlikely to materially move public markets.

Analysis

This is not a near-term monetization signal for broad AI infrastructure; it is a demand-quality signal for the education application layer. If unsupervised use becomes associated with weaker outcomes, schools and regulators are more likely to procure controlled, auditable tutoring products rather than permit open-ended chatbot access. That favors vendors able to sell workflow, assessment integrity, teacher controls, and outcome measurement—not merely token volume.

The likely second-order effect is higher compliance friction for consumer AI in classrooms and a shift in bargaining power toward incumbents embedded in school procurement. Duolingo (DUOL), Chegg (CHGG), Pearson (PSO), Instructure (INSTI), and PowerSchool (PWSC) face differing exposures: AI-native study assistance without demonstrated learning efficacy risks lower engagement quality and institutional restrictions, while platforms that can document pedagogical gains can convert restrictions into premium product tiers. The key distinction is whether AI cannibalizes paid content or raises completion, retention, and school contract value.

Over the next 1-3 months, the news flow is unlikely to move mega-cap AI names materially. Over 6-18 months, education departments may require age gating, human oversight, audit trails, and efficacy evidence; these requirements raise product costs but create a moat for enterprise education software. The contrarian view is that the negative association may largely reflect adverse selection—struggling students use AI more—so it does not establish that AI causes worse performance. Controlled-study evidence, rather than survey correlations, is the catalyst that matters for valuation.

Key Decisions for Investors

• No directional trade in NVDA, MSFT, GOOGL, or META on this item alone; education is immaterial to near-term revenue and the causal evidence is insufficient to change AI-capex assumptions.
• Place a 6-12 month watch on PSO and INSTI for district contracts that bundle AI governance, assessment, and teacher workflow tools. Upgrade only if bookings or net revenue retention show measurable AI-led acceleration; absent disclosed contract metrics, this is not yet a recommendation.
• Maintain skepticism on CHGG as a high-beta education-AI exposure: any evidence that schools restrict generic AI study tools would worsen its traffic and subscription-recovery path. A short is actionable only after confirming renewed subscriber declines or reduced guidance; a sustained stabilization in paid subscribers falsifies the thesis.
• For DUOL, monitor whether AI features lift paid conversion and retention without undermining learning outcomes. Consider long exposure only following evidence of improved paid conversion or ARPU; regulatory or school-level restrictions on youth AI features are the principal downside catalyst.