Students are trading mastery for minutes. Using generative AI to breeze through math homework is leading to a significant drop in actual learning and performance on proctored exams.
Math homework is getting done faster, but students are forgetting the material. Completion time for high school and college students dropped by about a third after the release of ChatGPT, yet their ability to solve similar problems on supervised, AI-free tests plummeted by 25%.
Speed used to be a proxy for competence, but in the AI era, it is a red flag for "cognitive surrender." If your teen is suddenly finishing math sets in record time, they aren't necessarily getting smarter—they are likely outsourcing the mental heavy lifting required to build long-term memory.
This creates a "learning debt" that stays hidden during unproctored homework but comes due during high-stakes, in-person exams like the SAT or final finals.
Large Language Models are exceptionally good at solving standard, text-based word problems. Researchers tracked 3.2 million learning interactions over a 10-year period to see how the sudden availability of tools like ChatGPT changed student behavior. They wanted to know if AI was a "tutor" that sped up learning or a "shortcut" that bypassed it.
AI is turning homework into a clerical task rather than a cognitive one. The data shows a clear divide in how different age groups use these tools:
- The High School/College Gap: Study time on word problems dropped by about 30% for older students since late 2022.
- The Retention Penalty: In settings where AI assistance was impossible, the odds of these same students answering questions correctly fell by a quarter.
- The AI Blindspot: The time-saving effect vanished entirely on interactive, graph-based problems that AI cannot easily "read" or solve.
- The Age Buffer: Fifth graders showed no change in behavior, suggesting younger children aren't yet using AI to offload their thinking.
We are entering a "trust but verify" era of education. Because students are performing better on unproctored digital assessments, they are likely using AI to solve test questions as well as homework. This creates a false sense of security for parents and teachers. The "efficiency" of AI is a trap: the less time a student spends struggling with a concept, the less likely they are to actually own that knowledge.
The paper is a preprint and has not yet undergone formal peer review by other scientists. The findings are also limited to students using the ALEKS digital math platform; while the sample size is massive, it may not perfectly represent students who do math exclusively with pencil and paper or in different software environments.
- If your high schooler is finishing math homework significantly faster than last year, ask them to explain the "why" behind three random answers to ensure they aren't just copying AI outputs.
- If you want to test true mastery, give your child a problem involving a graph or a visual diagram; AI currently struggles with these more than text-based word problems.
- If your child is in 5th grade or younger, focus on building solid mental math habits now before the temptation to use AI tools becomes accessible in middle school.
- If your student is performing well on "take-home" digital quizzes but struggling on in-class tests, move homework sessions to a common area without phone or browser access to rebuild their cognitive stamina.
Finishing the work is not the same as doing the work. Generative AI has made "doing homework" faster, but it has made "learning the material" harder by removing the productive struggle necessary for memory. Focus on the process and the explanation, not just the completed screen.
Sina Rismanchian, Hasan Uzun, Jeffrey Matayoshi et al. (2026). Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build. arXiv (preprint). — http://arxiv.org/abs/2605.21629v1


