Faster homework and weaker exam results, study tracks academic achievement gap with generative AI

The return of millions of students to classrooms coincides with a radical shift in learning methods, as generative AI models now complete school assignments in a matter of seconds. This reality raised a fundamental question for teachers and researchers: what is the actual usefulness of homework when a machine solves it, and how much knowledge remains in the student’s mind after the screen is turned off? Answering this question led researchers to track students over thirty months to uncover what happens to cognitive abilities when these technologies are used without discipline.

A research paper by David Stromberg of Stockholm University, in collaboration with Victor Lee and Yan Hui Wu of the University of Hong Kong, documented the results of one of the largest field studies in this area. The study followed about 27,000 students aged 12 to 18 in China across nine subjects, linking homework data with closed monthly tests up to the decisive entrance examinations. The results showed that homework scores rose by 18 percent after using generative AI, while the time taken to solve each assignment fell by roughly 30 percent, from an average of 64 minutes to 45 minutes.However, those students themselves, when taking closed monthly tests without AI assistance, saw their scores drop by 20 percent over six months.The decline also extended to their performance on long term decisive entrance exams.

These findings align with what the Organisation for Economic Co-operation and Development concluded in its early-year report titled Digital Education Futures 2026. The OECD affirmed that assigning school tasks to models not built specifically for teaching or used within a guided educational architecture improves momentary formal performance without delivering real learning gains. In the same vein, research co-authored by Dragan Gasicvic warned of the erosion of cognitive mental effort, describing the phenomenon as “post-cognitive laziness.”

The nature of use outside classrooms reveals a deeper gap in student behavior. An Ipsos survey commissioned by Vodafone involving 7,000 students in seven European countries found that 56 percent use AI to obtain information, 45 percent to clarify concepts, while 31 percent rely on it to provide complete task solutions. By contrast, only 29 percent used interactive and adaptive content for guiding independent learning, and the proportion of those who create personalized plans and track their progress did not exceed 20 percent. In parallel, a survey of British undergraduate students conducted by the Hip and ReflexTest institutes of 1,054 participants showed that 49 percent believed the technology improved their university experience and saved time, while its psychological impact on feelings of loneliness varied: 59 percent felt it made no difference, 21 percent felt less isolated, and 20 percent felt more isolated.

This reality forces education ministries and academic institutions in the Gulf, Egypt and the Levant to quickly recalibrate assessment tools and digital curricula. As the UAE, Saudi Arabia and Qatar accelerate the integration of AI tools into their study plans, the research shows that measuring student competence through traditional homework has become misleading, and reliance on unregulated assessments inflates superficial pass rates while weakening students’ actual ability in standardized and final exams. The change needed now does not mean banning the technology, but redirecting technology education purchases toward adaptive platforms that engage the student and develop their thinking, together with revising test designs to rely on direct critical thinking rather than recalling text that a machine can generate in seconds.