A May 13, 2026 Wall Street Journal feature by Lindsay Ellis reporting on Igor Chirikov's eScholarship paper Grade Inflation in AI-Exposed College Courses (UC Berkeley Center for Studies in Higher Education). The study analyzes more than 500,000 college grades from 2018 to 2025 at a large Texas public university and reports that AI-exposed classes awarded roughly 30% more A's than less-AI-exposed classes after ChatGPT's 2022 launch. The article pairs the finding with reporting on employer GPA use and on grading-reform proposals at Harvard and Yale.
URL: wsj.com Author: Lindsay Ellis (WSJ) Date: 2026-05-13 (May 14 print edition) Underlying paper: Igor Chirikov, Grade Inflation in AI-Exposed College Courses (UC Berkeley Center for Studies in Higher Education; eScholarship), released 2026-05-13. Class: foundational (long-form WSJ feature anchored on a load-bearing empirical study advancing a specific causal claim)
Summary of findings
Igor Chirikov (Berkeley CSHE) analyzed more than 500,000 college grades from 2018 to 2025 at a large Texas public university that publishes course syllabi and grade distributions. The paper reports that AI-exposed classes — humanities, engineering, and courses heavy on writing or coding — gave out about 30% more A's than less-AI-exposed classes after ChatGPT's 2022 launch, with corresponding declines in A-minus and B-plus grades. The pre-2022 trend showed little difference between the two groups. A's became even more likely in take-home and homework-heavy classes. Chirikov attributes the shift to student AI use rather than to gains in learning, an interpretation he asserts but does not directly test against a control.
The reported figures are summarized below.
| Finding | Number | Source |
|---|---|---|
| Sample size | 500,000+ grades | Chirikov paper |
| Time window | 2018-2025 | Chirikov paper |
| AI-exposed-classes A-grade increase post-2022 | ~30% | Chirikov paper |
| Pattern in take-home / homework-heavy classes | A's even more likely | Chirikov paper |
| Employer GPA-use 2023 | 37% | National Association of Colleges and Employers |
| Employer GPA-use latest | 42% (rebounded) | NACE |
| Handshake postings requiring GPA at all | Few | Handshake |
| Among those, requiring ≥3.5 GPA in 2025 | ~25% | Handshake |
| Same metric in 2020 | 9% | Handshake |
Employer and institutional response
The article reports that employer use of GPA in hiring has rebounded from 37% in 2023 to 42%, per the National Association of Colleges and Employers (NACE) annual survey, as the job market has tightened. Among the few Handshake job postings that require a GPA at all, about 25% demanded a GPA of 3.5 or higher in 2025, up from 9% in 2020 (per Handshake). Barclays and Morgan Stanley have set GPA minimums for some internship roles.
Two universities are reported as moving toward grading reform. A February 2026 Harvard College report stated that current grading policies do not allow employers to "compare students' performances," and faculty were voting in May 2026 on a proposal to cap the number of A's. An April 2026 Yale University report stated: "Grades exist to communicate what students have learned. At Yale, as at many peer institutions, they no longer do." At the course level, Wharton lecturer Chelsea Schein shifted grading weight from homework, which she described as easily completed in full by AI, toward in-class quizzes and midterms.
Chirikov is quoted saying, "AI may be helping people become more productive, to produce more, [but] I think it may harm their learning," and that "Learning requires productive struggle that is eroded by AI." Chelsea Schein, identified separately as a research VP at Veris Insights, is quoted that companies "are talking from both sides of their mouth," disliking AI in applications while wanting AI fluency in hires.
Key claims
| Claim | Confidence | Notes |
|---|---|---|
| 30% more A's in AI-exposed classes post-2022 | high | Chirikov peer-reviewed-equivalent (eScholarship); 500K-grade dataset is substantial. |
| Causal mechanism = student AI use, not learning gains | medium | Chirikov asserts; not directly tested via control. |
| GPA-use rising 37%→42% as job market tightens | high | NACE annual survey. |
| Harvard / Yale both moving to grade reform | high | Public reports cited. |
| AI is eroding "productive struggle" | medium | Pedagogy-research consistent; not specifically measured in Chirikov's data. |
Reception and interpretation
The study offers a large-N empirical measurement of grade inflation tied to AI use, with an accompanying shift in employer hiring practice. It provides acquisition-side evidence for the AI Deskilling thesis at the level of undergraduate learning rather than professional practice, a counterpart to the clinical maintenance-side evidence in Lancet Endoscopist Deskilling Study (2025). The reporting also documents a change in how GPA is used: from a 2020 framing in which GPA was treated as low-signal to a 2026 framing in which GPA is used as a filter while being treated as less reliable, with the Harvard and Yale grade-cap proposals presented as an institutional response. The pattern parallels the AI-driven manager purge reporting, which describes employers responding as credentials become less differentiating.
Several tensions remain unresolved. Universities are pushing back on student AI use while employers want graduates fluent in AI; the Harvard A-cap and Schein's in-class quizzes represent two different responses. A 30% increase in A's is not equivalent to a 30% increase in the quality those grades represent, and the data do not separate students' use of AI from grader leniency under AI. The sample is a single Texas public university, so generalizability across institution types — elite private, public flagship, community college — is not established.
Relationships
- supports: AI Deskilling (acquisition-side evidence), AI in Education, Education — AI Deployment
- contradicts: Naïve framing that AI tools simply boost learning; the data suggest scoring inflation outpaces skill gain
- depends-on: Jagged Frontier (different classes have different AI exposure), LLM Fallacy (production-quality output mistaken for learned skill)
- related: Igor Chirikov (entity created), Arvind Narayanan (AI-Snake-Oil framing on AI claims), 'I didn't want to be the guinea pig': inside tech's AI-fueled manager purge (Guardian, May 15 2026) (parallel "AI erodes signaling" thesis), Lancet Endoscopist Deskilling Study (2025) (clinical deskilling counterpart)
- regulated-by: Faculty governance (Harvard A-cap proposal); Handshake/NACE as informal hiring signals
Wiki Folding
Updates AI Deskilling (Chirikov 500K-grade dataset as flagship empirical anchor), AI in Education (grade-inflation section), Education — AI Deployment (Harvard/Yale institutional response), AI Divides (Literacy / Occupational / Ethico-Philosophical) (employer-credential-trust dynamic), and creates Igor Chirikov entity page.