AI-Assisted Dissertation Writing Linked to Lower Academic Employment Among US PhD Graduates

Summary

A non-peer-reviewed working paper examined more than 226,000 STEM dissertations completed at US universities between January 2019 and May 2026. Using the Pangram AI-detection service, the researchers classified abstracts as AI-written or not, without distinguishing fully generated text from partial assistance. AI-generated text first appeared in the dataset in 2023, rising from 1.7% of abstracts that year to 18.8% in 2025 and 29.4% in the first five months of 2026. The rate reached 36% in mathematics and computer science; it was 41% among graduates educated in non-English-speaking countries, compared with 23% among those educated in English-speaking countries. The researchers also compared education and employment profiles on LinkedIn. Among graduates who completed undergraduate study in the United States, those with AI-generated abstracts were 17% less likely to work in academia and 16% more likely to work in industry than those without detected AI text. The association was weaker for graduates who had completed undergraduate study outside the United States. The authors say the findings raise questions about whether students develop enough depth of understanding when AI performs part of the writing, while another researcher notes that AI may be more useful when established knowledge exists and more problematic in uncertain research areas.