Do AI Screeners Prefer AI-Written Resumes?
AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights
Jiannan Xu, Gujie Li, Jane Yi Jiang · arXiv:2509.00462 (PDF) · submitted August 30, 2025
Summary by Andrew Johnson, OneTwo Resume editorial team · updated · We are not affiliated with the authors. arXiv papers are preprints and may not be peer-reviewed.
Hiring has quietly become an AI-versus-AI interaction: applicants use language models to polish resumes while employers use language models to screen them. This paper asks the uncomfortable question that setup creates — do LLM screeners systematically favor content that resembles their own output?
Computer-science research had already documented "self-preference bias" in lab settings: models rating their own generations above human text of comparable quality. The authors take that finding into the hiring context specifically, empirically evaluating what happens when LLM-refined resumes meet LLM-based screening.
What the paper reports
- The study frames and empirically tests self-preferencing in hiring: LLM screeners evaluating resumes that were themselves refined by LLMs.
- It builds on established evidence that models tend to favor their own generated content over human-written content.
- The authors position this as the first real-world-context empirical evaluation of the effect in hiring rather than an abstract benchmark.
What this means for your resume
Our editorial interpretation — the paper does not give job-seeker advice.
- If employers in your pipeline use AI screening, an AI-polished resume is not "cheating" — it may be table stakes. The research suggests machine-style phrasing can be scored more favorably by machine screeners.
- The facts must stay yours: this effect concerns style and phrasing, not fabricated content. A polished lie still dies in the interview.
- Full disclosure: we sell an AI resume tool, so we have an interest in this finding being true — which is exactly why we link the paper and encourage you to read it rather than take our word.
Read it with these caveats
arXiv preprint (v4, 2025); peer-review status not verified by us. Effect sizes and which models were tested matter enormously here — read the paper before treating this as settled. Self-preference measured in one context does not guarantee your target employer’s stack behaves the same way.
Primary source: AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights — always read the paper before citing it. Spotted an error in our summary? Tell us and we'll fix it with a visible correction.