Research on secondary students shows that faster assignment completion boosted grades but reduced non-AI-assisted exam performance over time, with subject disparities.

A new large scale study in China by the non-profit Centre for Economic Policy Research suggests that using AI to breeze through homework may come at a cost: lower non-AI-aided test scores.

Researchers tracking 26,811 Chinese secondary school students (grades 7 through 12) over 30 months had found that generative AI cut homework time by about 30% and boosted homework scores by 18%, but it was also associated with a 20% drop in monthly exam performance, which did not allow AI use.

The findings point to a familiar educational tradeoff: AI can make assignments faster and easier, but that efficiency may weaken the effort students need to build lasting understanding. The study’s authors say the biggest losses appeared among students who completed homework unusually quickly while still earning strong homework grades, suggesting that AI may have replaced the mental work that normally helps learning stick.

Over time, the damage in the tested cohorts seemed to extend beyond routine quizzes. The study reported that penalties on higher-stakes entrance exams reached 18% to 24% after two years, with the sharpest declines showing up in social sciences and notable effects in Science, Technology, Engineering and Mathematics (STEM) and language subjects as well.

At the same time, the research does not argue that AI is inherently harmful in classrooms. Other studies have found that AI tutoring used during instruction can improve learning efficiency and engagement, underscoring that the impact depends heavily on how the tools are deployed.

The practical challenge now is policy. Schools and teachers in various countries will need to decide whether AI should be treated as a shortcut for homework, a study aid, or a supervised classroom tool that supports learning without replacing it. A growing number of US college professors have shifted to oral exams and in-class assignments to limit AI-enabled shortcuts, according to the Associated Press. Whether similar interventions can scale across K-12 systems — particularly in China, where AI-powered educational technology is a market reportedly valued at over US$43bn — remains an open question.