The Pulse
AI-Written Dissertation Abstracts Track With Lower Academic Employment
A working paper finds that AI-generated text became more common in U.S. STEM dissertation abstracts and was associated with graduates’ employment paths. The authors caution that the findings show an association, not that AI use caused gradu
AI.info Team · Written with AI ·
AI-generated text appeared in 29.4% of U.S. STEM dissertation abstracts filed during the first five months of 2026, up from 1.7% in 2023. A new National Bureau of Economic Research working paper also finds that graduates whose dissertations contain AI-written text are less likely to enter academic research careers and more likely to work in industry or non-tenure-track jobs.
The results come from a study by Daniel P. Gross, Dror Shvadron and Hansen Zhang, which examines a near-census of U.S. science, technology, engineering and mathematics PhD dissertations from 2019 through 2026. The paper, issued in October, is a working paper and has not been peer-reviewed. Its findings describe a link between AI-detected writing and career outcomes; they do not show that using AI caused graduates to leave academia.
Abstracts reveal a steep rise in flagged text
The researchers assessed dissertation abstracts with Pangram, an AI-text detection service. The share flagged rose to 18.8% in 2025, then reached 29.4% for submissions from January through May 2026. The analysis classified abstracts as AI-written or not, without distinguishing between text generated in full and text partly produced with AI.
Use varied by field and graduate background. Mathematics and computer-science abstracts had the highest reported rate, at 36%. About 41% of dissertations by graduates educated in non-English-speaking countries contained AI-generated abstract text, compared with 23% for graduates educated in English-speaking countries.
The employment gap is an association, not a verdict
To study career paths, the authors linked graduates’ education records with employment profiles on LinkedIn. Among graduates whose undergraduate degrees were from the United States, those with AI-generated abstracts were 17% less likely to be employed in academia and 16% more likely to be employed in industry than peers whose abstracts were not flagged. Nature reported that the relationship with academic employment was less pronounced among graduates who completed undergraduate studies outside the United States.
“Among graduates in the same program and cohort, those whose dissertations contain AI writing are less likely to enter academic research careers and more likely to enter non-tenure-track roles or industry, with this pattern strongest among US-origin students.”
Daniel P. Gross, Dror Shvadron and Hansen Zhang, authors of the NBER working paper
The authors also write:
“At the same time as AI may increase research productivity, it thus appears to also be displacing a potentially important portion of the work through which scientific expertise is developed.”
Daniel P. Gross, Dror Shvadron and Hansen Zhang, authors of the NBER working paper
Comparing graduates from the same program and cohort helps account for differences between institutions and graduation years, but the pattern remains observational. It cannot determine whether AI use affects later career choices, whether students already oriented toward industry use AI differently, or whether other factors shape both decisions.
Writing support and research training
The findings raise a specific question for doctoral programs: how much does dissertation writing contribute to the training they aim to provide? Nature quoted Ang Liu, an engineer at the University of New South Wales, saying that AI can help when students work in areas with substantial existing knowledge, while uncertainty makes rigorous research training more important.
The paper also discloses that its authors used Claude Fable 5.1 to draft an initial version of the discussion of the empirical findings and part of the data appendix, then substantially revised the AI-generated text by hand. That disclosure does not resolve what the employment association means, but it places the researchers’ own use of AI alongside the study’s examination of doctoral writing.