NBER: AI-drafted PhDs 17% less likely to stay in academia

Nearly 30% of US STEM PhD dissertation abstracts filed in the first five months of 2026 contained AI-generated text, and the graduates who filed them were markedly less likely to take an academic job. That is the headline finding of NBER Working Paper 35859, by economist Daniel Gross with Dror Shvadron and Hansen Zhang, reported by Nature.

The authors ran the AI-detection service Pangram over more than 226,000 STEM dissertations completed between January 2019 and May 2026 at US universities. AI-generated text in abstracts rose from 1.7% in 2023 to 18.8% in 2025 and 29.4% so far this year. Math and computer sciences led at 36%. Dissertations by international students were flagged at 41%, versus 23% for those from English-speaking countries. About 90% of flagged abstracts were classified by Pangram as entirely AI-generated.

Among US undergraduate degree holders in the sample, PhDs with AI-generated abstracts were "17% less likely to be employed in academia" and "16% more likely to be employed in industry" than peers without flagged AI use.

Gross framed the worry as a question about training rather than output. "Do they really have the depth of understanding" to be good scientists, he asked, if students offload a share of the writing. Ang Liu, an engineer at the University of New South Wales in Sydney who was not an author, told Nature that AI drafting helps most where prior knowledge is thick, and that at the frontier "rigorous research training becomes even more important." The paper is a working paper, not yet peer-reviewed.