August 12, 2026

AI chatbots linked to less study time and lower math scores in analysis of 3.2 million ALEKS interactions

A working paper analyzing 3.2 million ALEKS math interactions found less study time and lower proctored test accuracy after ChatGPT’s release.

A student solving a math word problem on a laptop next to a notebook - AI chatbots and math learning loss

Photo by Matheus Bertelli on Pexels

A working paper analyzing 3.2 million student interactions with the ALEKS math platform found that after ChatGPT’s late-2022 release, students spent measurably less time on word problems and answered fewer of them correctly on proctored placement tests, evidence the authors link to AI chatbots and math learning loss. The paper, “Faster Completion, Less Learning: Generative AI Reduced Study Time on Math Problems and the Knowledge They Build,” was posted to arXiv in June 2026 by Sina Rismanchian, a doctoral student at the University of California, Irvine, together with four researchers from McGraw Hill: Hasan Uzun, Jeffrey Matayoshi, Eric Cosyn and Eyad Kurd-Misto. It has not yet been peer reviewed.

ALEKS is an adaptive math learning and assessment platform used by more than four million students a year, from fifth grade through college, according to the paper and reporting by The Hechinger Report. Because the platform includes both everyday practice problems and proctored college placement tests, the researchers could compare student behavior and accuracy before and after ChatGPT’s arrival across roughly ten years of data.

How the study isolated AI’s effect on math learning

The researchers compared two kinds of problems that differ in how easily a student can hand them to a chatbot: word problems, which can be copied and pasted into an AI tool for an instant answer, and graphing problems, which require uploading a screenshot and recreating the graph inside ALEKS. Beginning in early 2023, time spent on word problems fell steadily while time spent on graphing problems held roughly steady, a gap that widened every quarter through the end of the study period near the end of 2025.

Decline in time spent on AI-susceptible word problems, by grade band, after ChatGPT’s release Decline in time spent on word problems since ChatGPT 0% 10% 20% 30% High schoolers 31% College students 27% Middle schoolers 9% Grade 5 students ~0%
Average time spent on word problems fell furthest among high school and college students, who could most easily copy problems into a chatbot; grade 5 students showed no detectable change. Source: Rismanchian et al., “Faster Completion, Less Learning,” arXiv, June 2026, reported by The Hechinger Report, July 6, 2026.

By the end of the study period, average time on a word problem had fallen 31 percent among high school students and 27 percent among college students, from about four minutes to less than three, according to The Hechinger Report’s account of the paper. Middle school students showed a 9 percent decline, and fifth graders showed essentially no change. During unsupervised practice sessions, students answered more word problems correctly, but performance told a different story under supervision.

Proctored test accuracy fell after AI’s arrival

Many colleges let incoming students retake a math placement test after further practice in ALEKS, a route that historically improved their odds of qualifying for a higher-level course. Historically, students answered about 80 percent of word problems correctly on supervised placement tests; after ChatGPT’s introduction, that fell to about 60 percent, a roughly 25 percent reduction in the odds of answering a word problem correctly, The Hechinger Report reported based on the paper’s findings. Performance on graphing problems, the type harder to outsource to a chatbot, did not decline over the same period.

Detail Value
Interactions analyzed 3.2 million ALEKS student engagements
Platform reach 4 million+ students a year, grade 5 through college
Time on word problems, high schoolers -31% (about 4 minutes to under 3)
Time on word problems, college students -27%
Time on word problems, middle schoolers -9%
Time on word problems, grade 5 students No detectable change
Proctored accuracy on word problems, before About 80% correct
Proctored accuracy on word problems, after About 60% correct
Graphing-problem performance No decline detected
Peer review status Working paper, not yet peer reviewed

Source: Rismanchian et al., “Faster Completion, Less Learning,” arXiv, June 2026; The Hechinger Report, July 6, 2026.

“What makes me nervous is that it’s not only about the word problems,” Rismanchian told The Hechinger Report. “This cognitive surrender might be going on in writing, science, everything.” The researchers said the pattern of decline appearing only in AI-susceptible problems, disappearing under supervision, and growing steadily for nearly three years made an AI-driven explanation more plausible than alternatives such as pandemic-era learning loss, which would also be expected to depress graphing performance.

The finding adds to a wider body of evidence on AI chatbots and math learning loss. A separate randomized experiment in Turkey found high school students who used AI to study math learned less than students who practiced without it, and Anthropic has separately reported that many college students use AI to obtain answers and offload cognitive work, according to The Hechinger Report. Other research has found the reverse effect: carefully designed AI tutors that ask guiding questions and withhold answers have improved outcomes in controlled studies, a distinction relevant to the ongoing debate over whether AI can raise grades while hurting learning.

Background on the ALEKS research

ALEKS, owned by McGraw Hill, is an adaptive assessment and learning system used across primary, secondary and higher education math instruction. Rismanchian’s earlier research, released in March 2026, had already documented patterns of AI usage in undergraduate short-response essays at a large California research university, according to The Hechinger Report. The new paper extends that work to quantitative problem-solving using a decade-long dataset spanning the period before and after ChatGPT’s public launch in late 2022, giving researchers a rare before-and-after comparison at scale. Because the paper is a working paper and has not yet completed peer review, its authors and outside researchers caution that it does not settle how much of the change reflects AI use specifically, though the pattern is consistent with growing concern among educators and students about AI’s effect on independent problem-solving skills.


Sources: arXiv preprint (Rismanchian et al.); The Hechinger Report; Forbes

Featured image: photo by Matheus Bertelli on Pexels (free Pexels license).


Become a Sponsor

Our website is the heart of the mission of WINSS – it’s where we share updates, publish research, highlight community impact, and connect with supporters around the world. To keep this essential platform running, updated, and accessible, we rely on the generosity of you, who believe in our work.

We offer the option to sponsor monthly, or just once choosing the amount of your choice. If you run a company, please contact us via info@winssolutions.org.

Select a Donation Option (USD)

Enter Donation Amount (USD)

What do you feel about this?