AIToday

AI homework help cuts exam scores by 20%, damage takes two years to surface

THE DECODER23h ago5 min read
AI homework help cuts exam scores by 20%, damage takes two years to surface

Key takeaway

A large-scale Chinese study found that students using AI tools for homework showed strong short-term grade gains but suffered 20 percent exam score drops within six months, with damage to high-stakes entrance exams taking about two years to fully emerge. The harm concentrates among students who finish work unusually fast, suggesting they rely on AI to outsource rather than learn; students who spend normal time on homework while using AI see no exam penalty. This long-term cost is largely invisible in the short term because teachers notice only single-subject declines and the aggregate effect takes years to accumulate.

Summaries like this, in your inbox every morning.

Sign up free →

3 Key Points

  • What happened

    A study of 26,000 students in China tracked how AI adoption affected learning outcomes. Six months after students began using AI tools like DeepSeek and ChatGLM, homework scores rose 18 percent and completion time fell from 64 to 45 minutes, but closed-book exam scores dropped 20 percent. The impact on high-stakes entrance exams took about two years to reach full effect, ranging from an 18 to 24 percent decline.

  • Why it matters

    Most students using AI appear to be outsourcing homework rather than learning; about 81 percent finished assignments in under 50 minutes while earning high grades but bombing exams. Only students who spent similar time on homework as non-AI peers saw no exam damage. Schools notice little because teachers see only one subject, and the aggregate county-wide effect didn't reach minus 10 percent until June 2025. This suggests short-term studies miss the real long-term cost to learning.

  • What to watch

    The learning penalty shrank from about 25 percent in early 2023 to 16 percent by June 2025, pointing to some adaptation by students and teachers, though losses persist. Social science subjects took the hardest hit at 27 percent decline, versus STEM at 22 percent. The study recommends shifting grading toward proctored in-class exams and tracking completion time rather than homework grades, since high homework scores now predict worse exam results among AI users.

FAQ

Which AI tools were students using?
The most popular tools were Doubao, DeepSeek, ChatGLM, Ernie Bot, and Qwen. Self-reported AI usage jumped with the releases of DeepSeek V2.5 in September 2024 and DeepSeek R1 in January 2025.
Which student groups were hit hardest?
Younger students in lower secondary school lost more than older ones (24 versus 17 percent), boys were hit harder than girls (21.6 versus 18.4 percent), and top performers suffered the most, with the top third seeing a minus 24 percent effect compared to minus 16 percent in the bottom third. A dose-response pattern showed students using AI for up to one hour per week lost about 5 percent, while those using it five hours or more lost 30 percent.
What happens if students spend normal time on homework with AI?
AI users who spent a similar amount of time on homework as their non-AI classmates scored just as well on exams while earning better homework grades and showed no sign of learning loss.

Discussion

No discussion yet for this article

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

1 minute a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →