A new study estimates that AI-generated writing appeared in 29% of U.S. STEM PhD dissertations filed in 2026.
The study also links flagged writing with lower odds of academic research jobs, but does not show that AI caused that career outcome. Researchers from Duke University and the University of Toronto authored the working paper. They analyzed more than 226,000 dissertation abstracts submitted between 2019 and May 2026.
A commercial AI detector flagged 29.4% of abstracts from the first five months of 2026, compared to 1.7% in 2023 and 18.8% in 2025. The researchers also examined a representative sample of full dissertations to see if abstracts accurately reflected AI writing across full texts. Because AI detector results do not definitively prove authorship, the figures should be viewed as estimates of flagged text.
The paper notes that AI flags appeared more often among graduates from non-English-speaking backgrounds and lower-ranked programs. Among graduates from the same program and graduating class, flagged writing correlated with a lower chance of getting an academic research job. Those graduates were more likely to take jobs in industry or positions not on the tenure track, a pattern that was strongest for graduates from the United States.
The authors emphasize that these results show a correlation rather than proof that AI use caused graduates' career outcomes. They also point out that the working paper has not yet gone through peer review.