26,811 pupils, 30 months: homework marks rose 18 percent and exam results fell 20
A working paper circulated by the Centre for Economic Policy Research as DP21577 followed 26,811 secondary pupils aged 12 to 18 in a county in central China for 30 months, comparing those who took up generative AI tools with classmates who did not. Homework scores among AI users rose about 18 percent while the time spent on each assignment fell from 64 minutes to 45. Exam scores went the other way, down about 20 percent; on the high-stakes entrance exams the gap was 24 percent on the zhongkao and 18 percent on the gaokao. The authors, David Stromberg of Stockholm University with Victor Lei and Yanhui Wu of the University of Hong Kong, report that roughly 80 percent of AI users showed the signature of outsourced homework: finishing unusually fast and still scoring unusually well. Losses were largest in social science subjects, then STEM, then languages, and were worst among younger pupils, high achievers and boys. The paper has not been peer reviewed. The Economist published an analysis of it on 18 August that pushed the findings into wide circulation.
Related stories
For two years schools tried to keep chatbots out. Now they are handing them to pupils and asking them to catch the mistakes.2026-08-21Two school years, 18 schools, one AI tutor: the gain was about what practising without it produces
2026-08-20A university system pointed a model at 14,000 syllabuses and asked it to find the forbidden topics
2026-08-19ChatGPT now guesses whether it is talking to a child — and if it decides yes, the product changes around them
2026-08-18Mexico's largest university is making tens of thousands sit its entrance exam again — in a room, on paper, with nothing in their pockets
2026-08-17