AI hiring judges may favor the AI that wrote an application
The writing model can change the verdict even when the applicant does not. Researchers tested AI-written cover letters against AI judges while holding candidate qualifications steady. Some judges favored certain writers, including their own model. The study tests model behavior; it does not measure actual hiring decisions.
What the source reports
Pagan and colleagues ran a preprint audit of AI systems that judge job applications. They crossed 11 AI models writing cover letters with nine models judging them. The test used ten job descriptions and 50 candidate profiles per job. The abstract reports that some judges favored letters from certain writing models. Some also favored their own model when qualifications were held steady. In the worst pairing they report, the estimated mismatch would filter 1.8 of 25 qualified candidates. That estimate comes from a controlled test, not from actual employers rejecting applicants. A hiring team would need to test its own writer-and-judge pair before relying on such a tool.
Original source
- Title
- AI evaluators favor job applications written by certain AI models, independent of candidate quality
- Author
- Nicolò Pagan, Michael Andreas Ebling, Stefania Ionescu, Anikó Hannák
- Publication
- arXiv cs.CY
- Date
- Monday, November 2, 2026