AI Use Lowers Math Test Scores, Study Finds

A comprehensive study analyzing millions of student interactions with the online math platform ALEKS has revealed that the widespread adoption of generative AI tools like ChatGPT has led to students completing assignments faster while simultaneously experiencing a decline in their learning and academic skills. The research, conducted by Sina Rismanchian, a doctoral student at the University of California, Irvine, in collaboration with McGraw Hill, examined student behavior and performance on both low-stakes practice problems and college placement tests before and after the introduction of ChatGPT in late 2022.

To specifically measure the impact of AI, the researchers differentiated between word problems, which are easily outsourced to AI chatbots for direct answers, and graphing problems, which require more complex interaction and are harder to automate. The findings indicated a significant divergence in student engagement and performance starting in early 2023. Students began dedicating less time to word problems, while their time spent on graphing problems remained relatively stable. This trend, observed across high school and college students, widened over time.

By the end of the study period in late 2025, the average time spent by high school students on word problems had decreased by 31 percent, and college students by 27 percent. This translated to a reduction from approximately four minutes per word problem to under three minutes. While middle school students showed a smaller decline of 9 percent, fifth graders exhibited no significant change in time spent on these problems. The researchers attribute this shift to students leveraging AI to bypass the cognitive effort required for solving word problems, thereby eroding their mathematical understanding and problem-solving abilities.

The study's methodology focused on isolating the effects of AI by comparing performance on tasks with varying degrees of AI susceptibility. The consistent reduction in time spent on easily outsourced problems, coupled with stable engagement on more complex tasks, provides strong evidence for AI's detrimental impact on learning outcomes. This research highlights a growing concern among educators about the potential for AI to undermine fundamental academic skill development, particularly in subjects like mathematics.