Artificial Intelligence and Grade Inflation

Synthesis: Artificial intelligence and grade inflation — a quasi-experimental study by Igor Chirikov (2026) analyzing 500,000+ grades across a balanced panel of 319 courses (84 departments) at a large research university, 2018–2025. Using a difference-in-differences design, it finds that after ChatGPT's release, courses with more AI-exposed tasks (writing, coding) saw the share of A grades rise by 13 percentage points (~30% relative to baseline) and GPA by 0.12 points, with grade-distribution compression. A triple-differences analysis shows the effect is concentrated in homework-heavy courses — evidence that AI-driven task displacement, not broad learning gains, is the primary mechanism.

Overview

Grades summarize student performance and signal skill to students, graduate programs, and employers. Generative AI threatens this certification function by performing graded tasks before instructors observe and assess them. Even with unchanged grading standards, grades can inflate when AI improves submitted work without a corresponding rise in underlying skill. This paper identifies a novel, technology-driven mechanism of grade inflation — distinct from the instructor-incentive, signaling, and measurement-error channels documented in prior literature — operating upstream of grading, on the production of graded work.

Method

• Data: Over 500,000 student-course enrollments across a balanced panel of 319 courses spanning 84 departments at a large selective research university in Texas, 2018–2025.
• AI exposure: Share of writing and coding tasks among all required tasks in each course's Fall 2022 syllabus (published before ChatGPT's release), validated in the companion syllabi paper as predicting AI-policy adoption.
• Design: Difference-in-differences comparing grade distributions before/after ChatGPT's release across courses with varying AI exposure; triple-differences (DDD) exploiting variation in homework weight (median = 30%) to distinguish task displacement from learning gains or student sorting.
• Robustness: Alternative exposure definitions, enrollment thresholds, pre-periods, and a placebo using oral-presentation tasks (where AI capabilities are weaker).

Key findings

• Grades rose substantially in AI-exposed courses. The share of A grades increased by 13 percentage points (~30% relative to the 2022 baseline of 0.44); mean GPA rose by 0.12 points; within-course GPA standard deviation fell by 0.09 points (grade-distribution compression). Effects diminish down the distribution (9 pp for A- or better, 5 pp for B+ or better, insignificant below) — AI primarily converts A- and B+ grades into A grades.
• The mechanism is task displacement, not learning gains or sorting. The triple-differences estimate shows above-median-homework courses gained an additional 16 pp in the share of A grades relative to below-median courses with the same AI exposure. If gains reflected genuine learning or sorting, they would appear regardless of assessment format; their concentration in unsupervised homework is consistent with AI substituting for student effort where instructors cannot observe production.
• Placebo and robustness confirm specificity. Oral-presentation task share (weak AI capabilities) shows no effect on grades. Results are robust to exposure definitions, enrollment thresholds (15–40), pre-periods, and clustering.

What this means for practice

• Instructors. Audit which graded tasks AI can complete unaided, and treat measured rises in student performance as a claim about how submitted work was produced rather than evidence of learning: the A-grade share rose 13 percentage points and mean GPA 0.12 points in AI-exposed courses.
• Instructors. Redesign assessments so AI use is structurally constrained or purposefully incorporated — process documentation, justification, follow-up interaction — rather than relocating all assessment to supervised in-person settings, which would narrow the capabilities being measured.
• Administrators. Read grade distributions by course task profile and homework weight instead of campus-wide GPA: because effects concentrate in unsupervised homework-heavy courses, the distortion sharply reduces transcript comparability across courses for employers and graduate admissions.
• Researchers. Measure AI exposure at the task level. Aggregate measures of AI use are unlikely to yield consistent effects, because the outcome depends on whether engagement with AI constitutes displacement or augmentation.

Limitations

• The analysis covers a single institution — a balanced panel of 319 courses across 84 departments at one large selective research university in Texas — which the author states limits external generalizability.
• The panel is restricted to courses large enough to appear in public grade distributions, which may underrepresent small seminars and upper-division courses, although varying the enrollment threshold between 15 and 40 students left results unchanged.
• AI exposure is inferred from course task composition in Fall 2022 syllabi; the study does not directly measure student AI use at the course level.
• The design identifies an average effect only: heterogeneity in how individual students engage with AI within a course remains unobserved, so the distribution of displacement versus augmentation across students is not identified.

Citation

Chirikov, I. (2026). Artificial Intelligence and Grade Inflation. CSHE Higher Education Working Paper Series, 26(3).