Students Using AI for Math Homework Learn Less, Study Finds

Students are completing math homework faster since ChatGPT's arrival in late 2022, but a large-scale study suggests they're learning significantly less in the process.

Researchers at UC Irvine and McGraw Hill analyzed millions of student interactions with ALEKS, an online math platform used by more than four million students annually from fifth grade through college. The study compared student behavior and performance before and after ChatGPT's public release, focusing on two types of problems: word problems that can be easily copied into AI chatbots, and graphing problems that require manual recreation using platform tools.

The performance gap

After ChatGPT launched, students began spending dramatically less time on word problems while maintaining consistent effort on graphing problems. By late 2025, high school students spent 31 percent less time on word problems—dropping from about four minutes to under three. College students showed a 27 percent decline.

The learning impact became clear in college placement tests. Students who practiced with unsupervised access to ALEKS answered more word problems correctly during practice sessions but performed substantially worse on the same problem types during proctored exams. Historically, students answered about 80 percent of word problems correctly on supervised tests. After ChatGPT's introduction, that figure fell to roughly 60 percent—a 25 percent reduction in the odds of answering correctly.

Performance on graphing problems showed no decline, suggesting the issue isn't general skill deterioration from pandemic learning loss or digital distraction.

Why it matters

The research provides concrete evidence that AI tools may be creating what lead researcher Sina Rismanchian calls "cognitive surrender"—students outsourcing the mental work that builds understanding. This pattern likely extends beyond math to writing, science, and other subjects where AI can generate quick answers. The findings arrive as universities simultaneously warn students against AI misuse while providing free access to premium chatbots, creating mixed messages about appropriate use.

The broader pattern

The ALEKS study joins accumulating evidence of AI's double-edged impact on learning. A randomized experiment in Turkey found high school students who used AI for math study learned less than those who practiced without it. Anthropic has reported that many college students use its Claude chatbot primarily to obtain answers rather than develop understanding.

Carefully designed AI tutors have shown promise in controlled experiments by asking questions and withholding answers until students work through problems—approaches that should increase time spent, not decrease it.

Rismanchian experienced the phenomenon personally. As an international student, he initially used ChatGPT to polish his English writing. "I realized that I cannot write anymore," he said. "I was losing my writing abilities." He stopped using AI for writing, though he continues using it for coding.

The paper, titled "Faster Completion, Less Learning," was released in June 2026 as a working paper and has not yet been peer reviewed. While the researchers couldn't directly observe students using AI, the timing, problem-type specificity, and disappearance of effects under supervision point strongly to AI use as the explanation.

Rismanchian argues the solution isn't simply banning AI but helping students value learning enough to resist outsourcing it. A recent RAND survey suggests many students already recognize the threat, reporting concerns that AI is weakening their critical-thinking skills even as more admit using it for schoolwork.

These findings were first reported by Jill Barshay at The Hechinger Report.