Did ChatGPT inflate college grades? What more than 138,000 students' records show
Evidence level: Level 3: Controlled comparison. The study compared courses before and after ChatGPT's release using baseline records, a quasi-experimental design (not a randomized trial). As a non-randomized study it cannot go above Level 3, and it is also a preprint, not yet reviewed by other experts. Other labels: higher-ed students. How we grade
Source: Generative AI Availability, Grades, and Student Satisfaction at a Large University, arXiv preprint, July 2026. We read the abstract, not the full paper.
The question
A common worry: students use generative AI instead of thinking, and earn high grades without learning. If that is true, grades should rise most in courses that rely on take-home work like problem sets and essays, rather than in-class exams.
What the study did
The researchers used syllabus and administrative data from a large U.S. university covering 2016 to 2025: 138,386 students and 72,730 course offerings. They used a human-validated AI pipeline to sort each course's assessment types (such as take-home or in-class) from its syllabus and estimate how exposed it was to AI. Then they compared courses before and after ChatGPT's release (a difference-in-differences design), with different ways of accounting for COVID-19.
What it found
The authors report that:
- effects on grades and on student satisfaction (self-reported understanding and interest) were small, and changed depending on how COVID was modeled
- there were no detectable differences by students' prior academic preparation
- in exploratory analyses, the link between pre-college test scores and grades had not detectably weakened in more exposed courses, and instructors had been making courses less exposed since before ChatGPT
What this means for you
- At this university, the feared grade inflation did not clearly show up. The authors suggest one reason may be that instructors adapted their teaching.
- Grades are not the same as learning. This study looks at grades and satisfaction, not at what students could do without AI.
- One university's records may not match your school. Check how your own assessments rely on take-home work.
What we have not checked
We have not read the full paper, so we cannot report the exact sizes of the effects.
---
Want studies like this graded in plain language every Tuesday? Subscribe free to The Evidence Signal.
Reference
Dumlao, J. M. Z., Wang, M., Xie, Z., Hu, J., Bar, I., Chaney, G., III, Gold, H., & Teplitskiy, M. (2026). Generative AI availability, grades, and student satisfaction at a large university [Preprint]. arXiv. https://doi.org/10.48550/arXiv.2607.21534