AI Speeds Up Homework but Test Scores Drop: A Mixed Picture in Classrooms
Mohamed Hussein
A large-scale study following thousands of students found that generative AI boosted homework scores by 18% while cutting time spent on assignments, yet exam scores fell by 20% within six months when AI was unavailable. The technology's impact hinges on how students choose to use it.
Hundreds of millions of primary and secondary school students worldwide have just entered a new academic year amid artificial intelligence's deepening penetration into classrooms, raising the question: what meaning does homework still hold when a chatbot can complete it in seconds?
Some countries have been quick to integrate AI into their curricula. China mandates a minimum of 8 hours of AI instruction per year; the United Arab Emirates, Saudi Arabia, and Qatar have also begun incorporating AI into their educational frameworks.
Faster Homework, Lower Exam Scores
A research team comprising David Stromberg (Stockholm University) alongside Victor Lei and Yanhui Wu (University of Hong Kong) conducted one of the largest studies on this topic, tracking approximately 27,000 students aged 12 to 18 in China over 30 months. The study combined data on homework, monthly closed-book exams, and high-stakes admissions tests across 9 subjects.
The results reveal a stark gap between completing assignments and genuinely learning from them. After using generative AI, students' homework scores rose by 18%, while the time spent on each assignment fell by about 30%, from an average of 64 minutes to 45 minutes. However, when these same students sat for monthly exams without AI assistance, their scores dropped by 20% within just 6 months. Results on high-stakes admissions tests also declined, with the full impact emerging over a longer period.
The Organisation for Economic Co-operation and Development (OECD) reached a similar conclusion in its 2026 Digital Education Outlook report, synthesizing studies published in January. The OECD noted that where generative AI is not designed for teaching and is not used within any framework, delegating assignments to AI models improves performance but yields almost no real learning benefits.
Research co-authored by Dragan Gasevic terms this decline in effective cognitive effort "metacognitive laziness."
How Students Use AI Matters Most
How students use AI tools turns out to matter more than whether they use them at all. An Ipsos survey commissioned by the Vodafone Foundation of 7,000 students aged 12 to 17 across 7 European countries found that when studying independently outside school without teacher guidance, 56% used AI to search for information and 45% asked it to explain terms and concepts. But as many as 31% admitted to using AI to solve entire assignments.
Only 29% used interactive or adaptive content to guide their learning, and just 20% used AI to create personalized study plans and track progress — approaches considered more directly supportive of self-directed learning.
AI in Student Life
For university students, AI is not merely a homework tool but is increasingly becoming part of a broader learning experience, from receiving instant help to finding companionship.
The 2026 Student and Generative AI survey conducted by HEPI and Kortext of 1,054 full-time undergraduates in the UK found that 49% believe AI has improved their student experience by saving time, enhancing understanding of material, and providing immediate personalized support. The impact extends beyond academics: for the first time, the survey asked about loneliness. A majority of students (59%) said AI made no difference; among the rest, the split was nearly even — 21% felt less lonely, while 20% felt lonelier.
