What happens when the AI reading your resume was trained on content made by other AI tools? According to new research published on arXiv, the answer is that you lose - unless your resume was also written with AI.
The paper, titled "AI Self-preferencing in Algorithmic Hiring," provides empirical evidence that AI-powered applicant screening systems score resumes written or heavily edited by AI tools significantly higher than human-written ones. The researchers call this "self-preferencing" - AI systems favoring AI-generated content in ways that correlate with how those systems were trained, not with actual candidate quality.
What the Research Actually Found
The study tested AI hiring tools against matched pairs of resumes - same candidate, same qualifications, one version written naturally and one polished or rewritten with AI assistance. The AI-written versions scored higher across multiple platforms, even when the underlying qualifications were identical. The effect wasn't small or borderline - it was consistent enough to produce measurably different hiring outcomes at the screening stage.
This matters because algorithmic screening now filters out most applicants before any human sees their file. If you're applying to a company using AI recruitment software, your resume may never reach a recruiter unless it passes an automated score threshold. Employers often don't know exactly how these systems score, and candidates have almost no visibility into it.
The Feedback Loop Problem
The self-preferencing dynamic creates a compounding problem. Job seekers who learn (or guess) that AI-polished resumes score better will start using AI tools to write their applications. This shifts the baseline upward, pressuring more candidates to do the same. Meanwhile, hiring tools trained on outcomes from this new pool will become even better calibrated to AI-style content.
The research also raises a harder question for employers: if your hiring tool prefers AI-written resumes, you're not screening for the best candidates - you're screening for the candidates most willing or able to run their resume through a language model. For roles where writing ability matters (marketing, communications, content, customer-facing work), that's a significant distortion. For technical roles, it may simply be noise that increases application volume without improving hire quality.
For job seekers, the practical takeaway is uncomfortable but real: using AI to improve your resume is no longer just a productivity choice, it may be a competitive necessity when applying to companies that use automated screening. Tools like Ashby and similar recruiting platforms are deeply embedded in how mid-size and larger companies manage hiring pipelines, and the AI layers built on top of them are rarely auditable by applicants.
The harder fix falls on employers. Running periodic audits on whether your screening tool systematically scores AI-generated content above human-written content - independent of quality signals - is something most HR teams aren't doing yet. The research suggests they should start.