NBER: PhDs who draft with AI less likely to stay in academia

The headline finding, as Nature reports: among PhDs with US undergraduate degrees whose dissertation abstracts flagged as AI-written, graduates were 17% less likely to take an academic job and 16% more likely to go to industry than peers whose abstracts did not flag. Two of the researchers we follow picked up the piece on first circulation.

Economist Daniel Gross with co-authors Dror Shvadron and Hansen Zhang ran more than 226,000 US STEM dissertation abstracts from January 2019 to May 2026 through the detection service Pangram. Prevalence of AI-generated text climbed from 1.7% in 2023 to 18.8% in 2025 and 29.4% in the first five months of 2026, with mathematical and computer sciences highest at 36% and physical sciences lowest at 22%. The split by language background is wider: 41% of abstracts by graduates from non-English-speaking countries were flagged, against 23% from English-speaking countries.

Gross frames the concern bluntly. "If our students are only doing half the work and they're dishing the rest to AI, do they really have the depth of understanding" to be good scientists, he asks. Engineer Ang Liu of the University of New South Wales told Nature that in genuinely unexplored territory, "rigorous research training becomes even more important."

Gross says roughly 90% of flagged abstracts were entirely AI-generated rather than lightly edited. The paper classifies abstracts as AI-written or not and does not separate wholesale drafting from partial assistance.