AI-driven homework gains mask 20% drop in student exam scores
A large-scale study reveals a decoupling of task completion from actual learning as students outsource cognitive effort to generative AI.
Generative AI is creating a dangerous divergence between student performance on assignments and their actual mastery of subject matter. While the technology accelerates task completion, it appears to be eroding the cognitive foundations required for independent success.
A study of 27,000 students aged 12-18 in China found that adopting generative AI increased homework scores by 18% and significantly reduced the time spent on assignments, which dropped from 64 to 45 minutes. However, these short-term gains vanished in high-stakes environments; closed-book exam scores for the same group of students fell by 20% within six months of AI adoption. This phenomenon suggests that the ease of obtaining an answer replaces the mental struggle necessary for learning.
The Illusion of Competence
This trend is part of a broader global shift in how students interact with educational tools. An Ipsos survey of 7,000 European students between the ages of 12 and 17 found that 31% use AI to provide complete solutions to their tasks, while only 20% utilize the technology to create personalized learning plans. This suggests that a plurality of students are using AI as a replacement for thought rather than a supplement to it.
These findings are echoed in the OECD Digital Education Outlook 2026, which reports that outsourcing tasks to generative AI without a structured teaching approach results in improved performance on the specific task but yields no actual learning gains. The result is an "illusion of competence," where students believe they understand the material because they can produce a correct output, despite lacking the underlying knowledge.
Systemic Risks to Education
The risk extends beyond individual grades to the very purpose of academic instruction. As AI is integrated into global curricula, educators are increasingly questioning the utility of traditional homework. When the process of producing a correct answer is decoupled from cognitive mastery, the traditional feedback loop of assignment and assessment breaks down.
While some students report a better overall experience—49% of UK undergraduates believe AI has improved their student experience, according to the OECD report—the academic cost is becoming clear. The erosion of productive cognitive effort threatens long-term critical thinking skills and leaves students vulnerable in environments where AI assistance is unavailable.
The Path Forward
As schools grapple with these results, the focus is shifting toward how AI can be integrated without bypassing the learning process. The current data suggests that without a rigorous teaching structure to guide AI usage, the technology may act as a crutch that weakens the student's intellectual autonomy. Educators must now determine if homework can be redesigned to prioritize the process of inquiry over the final answer, or if the current model of take-home assignments has become obsolete in the age of generative AI.