A large scale study of secondary students in China has found that AI adoption results in a short term improvement in scores and decline in homework duration, but lowers monthly exam scores by 20% within six months.
The study by Stockholm University’s David Stromberg, with Victor Lei and Yanhui Wu from the University of Hong Kong, found that AI was not the problem – but unfettered access to and use of AI were a particular issue.
Drawing on 30 months of data, covering the performance of more than 26,000 Chinese students in grades 7-12, the researchers write that while AI could enable workers to be complete knowledge-intensive tasks faster, the use of AI, “may also crowd out the effort through which people acquire skills.”
“The possibility of an AI learning penalty is especially important for young students,
who perform a wide range of cognitive tasks in school. For students, completing these
tasks efficiently is not the goal; learning from them is,” the authors write.
“Hence, the rapid diffusion of generative AI tools among students in recent years has created widespread concerns about their learning. Yet systematic evidence on the learning losses due to generative AI use in natural school settings is still lacking.”
The researchers analysed student performance aligned with uptake of AI, which rose from around zero in 2022 to around 80% by June 2025.
Generative AI boosted homework performance by 18% after six months of use and completion time per assignment shrank from 64 to 45 minutes. Short and long term learning outcomes were diminished as a result of a reliance on AI, driven primarily by the 81% of students who would seek or be open to outsourcing their homework. This showed up in closed-book exams.
“By contrast, AI users who spend as much time on homework as non-users achieve similar exam scores, even though their higher homework scores indicate that they use AI. These students are not differentially selected on prior achievement, suggesting that generative AI does not reduce their learning efficiency,” the authors write.
The impact of AI use varies across disciplines, with social science subjects most significantly affected.
The findings highlighted the importance of understanding and addressing demand-side issues with AI, the authors said. AI tutors which would enhance learning were readily available, but students instead chose to outsource work to other free tools. Finding ways to change the ways students chose to use AI would be critical to improve learning outcomes.