When ChatGPT arrived in late 2022, teachers quickly asked whether students would use artificial intelligence to cheat, learn, or simply do their homework more efficiently. Evidence is beginning to point to a troubling answer: Many students appear to complete their homework faster while learning less from it.
This finding comes from one of the largest studies of how generative AI changes students’ behavior and academic skills. Sina Rismanchian, a doctoral student at the University of California, Irvine, teamed up with researchers at McGraw Hill to analyze millions of student interactions with ALEKSan online mathematics platform used by more than four million students per year, from fifth grade to university. Because ALEKS includes both low-stakes practice problems and college placement tests, the researchers were able to compare student behavior and performance before and after the arrival of ChatGPT.
To isolate the effects of the AI, the researchers compared two types of math problems that differ in how easily students can give them to the AI: word problems and graph problems.
Word problems can be copied and pasted directly into AI chatbots for instant responses. Graphical representation problems are much more tedious. A student will need to upload a screenshot while recreating the graph in ALEKS using its tools.
After ChatGPT was launched, students’ behavior and performance on the two problem types began to diverge.
Starting in early 2023, students began spending less time on word problems while continuing to spend about the same amount of time on graph problems. The gap widens every quarter. By the end of the study period, near the end of 2025, the average time spent on word problems had decreased by 31% among high school students and 27% among college students, from about four minutes per word problem to less than three. (Middle school students showed only a slight 9 percent decline, and fifth graders showed virtually none.)
Researchers believe these averages are dragged down by some students who spend only a few seconds on word problems because they use AI to answer them.
The same pattern showed up on college placement tests. When exams were taken without supervision, students spent significantly less time on word problems after ChatGPT was released. On proctored exams, time spent on word problems returned to historical norms.
But time is only half the story. The most disturbing discovery concerns what happened to learning.
Many colleges allow new students to retake placement tests after practicing more math at ALEKS, giving them the opportunity to qualify for a higher-level course. Before ChatGPT, this practice was generally paid for. After ChatGPT, students answered more word problems correctly during unsupervised practice sessions, but performed significantly worse on those same types of problems when they later took a proctored placement test.
Historically, students answered approximately 80% of these word problems correctly on supervised placement tests. After the introduction of ChatGPT, this figure dropped to around 60%, a reduction of around 25% in the chance of answering a word problem correctly.
On the other hand, performance on graphics problems did not decrease.
After the release of ChatGPT, students performed worse on (AI-aware) word problems on proctored exams, but answered more word problems correctly in unproctored contexts.
If students’ math skills had generally deteriorated due to pandemic-related learning loss, lower high school preparation, or digital distraction, graphics performance should have deteriorated as well. This is not the case.
The study cannot definitively prove that students were using AI. The researchers couldn’t see what else was happening on the students’ screens besides ALEKS. But it is difficult to find another explanation. The changes only appeared in problems that were easy to outsource to AI, disappeared under surveillance, and grew steadily for almost three years.
“What makes me nervous is that it’s not just problem words,” Rismanchian told me. “This cognitive abandonment could happen in writing, science, everything.”
The newspaper, “Faster completion, less learning“, was published in June 2026 as a working paper and has not yet been peer-reviewed. Like any study, it does not answer the question of how much students use AI in their schoolwork, whether it harms learning and to what extent. But it joins a growing body of evidence that generative AI causes students to ignore the brain work that leads to learning, and that this “cognitive dropout” is becoming commonplace.
A randomized experiment in Turkey found that high school students who used AI to help them study math ultimately learned less than those who practiced without it. Anthropic, the creator of Claude, separately reported that many students appear to be using AI to get answers and offload cognitive work. Rismanchian’s previous research, published in March 2026, documented troubling patterns of AI use in short-response essays among undergraduates at a large California research university.
This is not to say that AI always undermines learning. Carefully designed AI tutors have improved student performance in controlled experiments by asking questions, personalizing instruction, and withholding answers until students think through a problem. But using AI in this way should increase the amount of time students spend on a problem, Rismanchian said. The ALEKS data shows the opposite.
Rismanchian doesn’t believe the answer is simply banning AI. Instead, he argues, students need to value learning enough to resist the temptation to outsource it.
A recent RAND survey suggests that many already recognize the threat to their brains. Students say they are worried that AI will be weaken their critical thinking skills while more of them admit to using it for their school work.
Students are not entirely responsible. Even though many professors have warned students not to use AI to complete their assignments, universities themselves have embraced the technology, often offering students free access to premium chatbots.
“I think we need to make students understand that you need to value your learning,” Rismanchian said. “If ChatGPT does it for you, then you haven’t learned it.”
Rismanchian understands the temptation.
An international student, Rismanchian started using ChatGPT to help him brush up on English for his homework. The ideas were always his. But after several months, he said, he noticed something troubling.
“I realized I couldn’t write anymore,” he says. “I was losing my writing skills.”
So he stopped using AI to write.
He still uses it for coding.
Contact staffwriter Jill Barshay at 212-678-3595, jillbarshay.35 on Signal or barshay@hechingerreport.org.
This story about AI use eroding math skills was produced byThe Hechinger reportan independent, nonprofit news organization that covers education. Register for Proof pointsand othersHechinger Newsletters.
The article Faster Fixes, Lower Test Scores: How AI Is Eroding Math Skills appeared first in the Hechinger Report.
