Do AI résumé screeners favor candidates whose writing sounds like the AI itself over what they can do?

This explores how surface features of an application, rather than what the candidate can actually do, end up deciding who gets through a hiring screen. The collection has no study of resume length as such. What it does have is sharper and a little more unsettling: the surface feature that matters most to an AI screener may be whether the resume sounds like the AI itself.

In a controlled experiment on 2,245 resumes, eight of nine language models preferred their own rewrites over matched human-written versions. Preference rates ranged from 26% to 98%, and the bias got stronger in larger models. It came from stylistic alignment, not from any difference in content quality Do language models favor resumes they rewrote themselves?. When this was simulated as a full hiring pipeline across 24 occupations, applicants who wrote with the same model the employer used to screen were shortlisted 23% to 60% more often than equally qualified human-written applicants. The largest gaps were in business roles like sales and accounting Do LLM evaluators favor resumes written by their own model?. The practical upshot is odd: the best resume strategy may be to guess which chatbot the employer uses.

Humans fall for a version of the same thing. Evaluators have been shown to mistake AI-generated documents for human writing and to rate them higher than real human submissions. That is why one review argues polish should stop being read as a sign of merit Does polished writing actually signal better quality work?. Recruiters also reward labels without checking them. In a conjoint experiment (recruiters choosing between profiles that differ in controlled ways) with 1,725 recruiters, simply listing AI skills raised interview invitations by 8 to 15 percentage points. Certificates added only a little beyond self-declaration, so the claim alone carries most of the weight Do AI skills help candidates get more job interviews?. One finding suggests substance still leaks through: on Freelancer.com, workers who spent longer editing their AI-drafted cover letters were hired more often, even though most submitted drafts with little revision Does editing time on AI drafts predict hiring success?. That is a correlation, not proof that editing causes hiring success.

At the system level, surface-gaming and surface-filtering seem to feed each other. Greenhouse's survey reports that 41% of job seekers use prompt injections (hidden instructions in a resume meant to manipulate AI filters) and 49% send more applications than before. Meanwhile 34% of recruiters spend half their week filtering spam. The survey supports each part of this loop but does not show which side drives the other Are job applicants and employers locked in an escalating AI arms race?. Perceptions are split too: 70% of hiring managers say AI helps them decide faster, but only 8% of job seekers think it makes hiring fairer. Only 21% of recruiters are very confident their systems don't reject qualified candidates Do hiring managers and job seekers agree on AI fairness?.

A useful lens comes from outside hiring. Recommendation research shows that ranking systems trained on their own past choices drift toward self-reinforcing equilibria unless they explicitly model selection bias Why do ranking systems need to model selection bias explicitly?. A hiring screener that favors its own style, and that later learns from the people it selected, risks the same loop. And fixes that only touch the prompt may not be enough: persona prompts shift where bias shows up in a model's output, but the gaps between groups persist underneath Can persona prompts actually reduce bias in language models?. Telling a screening model to 'ignore style' may change what it says without changing what it rewards.