Study Finds Generative AI Improves Homework but Hampers Exam Results Among Students
A Chinese study reveals that while generative AI raises homework scores by 18%, it leads to significant drops in exam performance among students.
By John Mtongoi Newsroom
August 21, 2026
2 min read
A recent study conducted in a Chinese county over two and a half years has shed light on the double-edged impact of generative Artificial Intelligence (AI) on student learning outcomes.
The research tracked students from grades seven to twelve and found that those who independently used generative AI tools:
- Spent 30% less time completing homework.
- Achieved an 18% increase in homework scores.
- Experienced a 20% decline in monthly examination scores after five months of AI use.
- Recorded drops of 18% and 24% in two separate high-stakes entrance exams.
These findings suggest that while AI may help students complete assignments more efficiently, it might hinder the deep learning and skill acquisition necessary for success in formal examinations.
The study raises concerns about the growing accessibility of AI technologies through smartphones and search engines, creating an uncontrolled environment for AI use in education. The World Bank has noted the potential risks when AI substitutes actual learning rather than supporting it.
Highlighting the scale of the issue, the study reported a 1.4 standard deviation drop in exam performance, a significant effect size compared to other education research. In contrast, structured AI-assisted tutoring programs, such as one in Nigeria, showed positive learning gains.
Experts emphasize that AI is not inherently detrimental to education but caution that unsupervised use may lead to unintended learning losses. The findings serve as an early warning for education systems worldwide as AI becomes increasingly integrated into students’ daily lives.
