AI lifts homework scores 18% but cuts exam results 20% in China study

September 2, 2026 at 11:01 AM2 min

This digest was compiled by AI from multiple sources — links to the originals are below.

A study of 27,000 Chinese students found AI use raised homework scores by 18% while cutting time spent on assignments by 30%. Monthly exam scores fell 20% within six months, and high-stakes entrance exam performance also declined. The OECD's Digital Education Outlook 2026 reached a similar conclusion, warning of 'metacognitive laziness' when AI is used without structure.

Key Facts

The study tracked 27,000 students aged 12 to 18 in China over 30 months across nine subjects.
After adopting generative AI, homework scores rose 18% while time per assignment fell from 64 to 45 minutes.
Monthly closed-book exam scores fell 20% within six months of AI adoption.
The OECD's Digital Education Outlook 2026 found that unstructured AI use improves task performance without real learning gains.
An Ipsos survey of 7,000 European students aged 12-17 found 31% used AI to provide complete solutions to tasks.

The China Study

David Stromberg of Stockholm University, with Victor Lei and Yanhui Wu of the University of Hong Kong, ran one of the largest studies of its kind. The working paper combined data on homework, monthly closed-book exams, and high-stakes entrance exams across nine subjects. After adopting generative AI, students' homework scores rose by 18 percent, while time spent per assignment fell by about 30 percent, from 64 minutes to 45 minutes on average. When the same students sat monthly exams without AI, their scores fell by 20 percent within six months. Performance in high-stakes entrance exams also declined, with the full effect emerging over a longer period.

OECD Findings

The Organisation for Economic Co-operation and Development (OECD) reached a similar conclusion in its Digital Education Outlook 2026, published in January. The report found that where generative AI is neither built to teach nor used under any structure, outsourcing tasks to the model improves performance with no real learning gains. Research co-authored by Dragan Gasevic describes this potential erosion of productive cognitive effort as 'metacognitive laziness'.

Student Usage Patterns

A survey by Ipsos for the Vodafone Foundation asked 7,000 students aged 12 to 17 across seven European countries how they use AI when studying outside school and without teacher instruction. The most common use was getting information, reported by 56 percent of students. Another 45 percent used AI to have terms and concepts explained to them. But 31 percent said they used it to provide complete solutions to tasks. Fewer students used AI in ways that more directly support independent learning.