AI shortcuts linked to lower science scores in PISA 2025 results
OECD data reveals that students using AI for specific schoolwork tasks score the equivalent of one year of schooling behind their peers.
The integration of artificial intelligence into classrooms is creating a stark divide in student achievement based on how the technology is used. New data from the OECD's PISA 2025 results show that while AI can support learning, using it as a shortcut for core academic tasks is linked to significant performance drops.
According to the OECD, students who employ AI for specific schoolwork—including drafting research, summarizing texts, or writing—score approximately 20 points lower in science than those who do not use these tools. This gap is equivalent to being one full year of schooling behind. The report notes that approximately 46% of students across OECD countries now use AI chatbots weekly or more to assist their learning. However, the impact on grades varies by intent: students who use AI for general purposes show performance outcomes similar to non-users.
The Literacy Gap
These findings are part of a massive study involving over 760,000 students across 91 countries. The PISA 2025 cycle observed a general decline in average performance across mathematics, science, and reading. Reading saw the most severe impact, with average scores across OECD countries falling by 28 points between 2015 and 2025. Additionally, the OECD found that instances of "hasty reading"—where students read quickly but inaccurately—nearly doubled to 9% between 2018 and 2025.
Implications for Learning
The data suggests a critical distinction between AI as a replacement for effort and AI as a tool for enhancement. The OECD found that students who use AI for general purposes and have access to formal AI literacy skills training at school tend to achieve slightly higher science scores. This indicates that the negative correlation between AI and performance is not inherent to the technology, but rather to how it is applied without guidance.
The Path Forward
This trend highlights a growing "AI divide," as opportunities for literacy training are currently more prevalent among socio-economically advantaged students. The results suggest that using AI to bypass the cognitive struggle of summarizing or researching may hinder deep learning and critical thinking skills. For education systems, the findings signal an urgent need to integrate AI literacy directly into curricula. Rather than simply banning or permitting the tools, the data suggests that institutional support is the primary factor in turning AI from an academic liability into a performance booster.