AI Self-Preferencing in Hiring: The Hidden Bias in Resume Screening

Your AI resume screener might be favoring candidates for a reason that has nothing to do with talent. It could simply be rewarding resumes that were written by itself.

What is AI self-preferencing in hiring?

AI self-preferencing is the tendency of an AI model to favor writing generated over equally strong writing produced by a person. In hiring, this can happen when AI is used on both sides of the process: a candidate uses an AI tool to write or optimize a resume, and the employer uses a similar AI-powered tool to screen applications. The screener recognizes its own writing style and ranks that resume higher than an equally qualified human-written version.

Here’s the unsettling part: nothing appears to be wrong. Resumes are screened, scores are assigned, and candidates are shortlisted. Unless you specifically test for it, the bias can remain completely hidden.

AI screeners prefer AI-written resumes

A recent study from researchers at the University of Maryland and Ohio State tested whether AI self-preferencing occurs in hiring. They collected thousands of real resumes written before generative AI became common, created AI versions of each one, and then had the same AI models evaluate both versions for the same jobs.

The study found:

• AI models preferred their own writing 67% to 82% of the time, even when the human-written resume was equally strong.
• Across 24 occupations, candidates who used the same AI tool as the screener were 23% to 60% more likely to make the shortlist than equally qualified candidates who wrote their own.
• The effect was strongest in business roles, including sales and accounting.
• A simple prompt instructing the AI to ignore whether a resume was AI-generated cut the bias by more than half.

Resumes can’t predict job performance

AI self-preferencing highlights a much bigger issue: resumes are becoming a weaker predictor of job performance.

A resume tells you where someone has worked and how they present their experience. It doesn’t directly measure whether they can perform the job. Resumes have always had limitations as predictors of job performance, and generative AI makes them even less reliable.

Today, a “strong” resume can reflect AI prompting skill and access to the latest tools as much as a candidate’s qualifications. If a resume can receive a higher score simply because it was optimized using the same AI model as the screener, it tells you less about who can do the job and more about who created the better AI-assisted resume.

Why AI self-preferencing raises adverse impact concerns

Beyond making resumes less reliable, AI self-preferencing also raises important questions about fairness in hiring and has the potential to create adverse impact.

Access to generative AI tools, digital literacy, and AI prompting skills are not distributed equally across applicants. These differences often correlate with factors such as income, geography, language, education, and familiarity with technology.

If an AI screener systematically rewards resumes optimized with a particular AI model, candidates without access to those tools may be disadvantaged for reasons unrelated to their ability to perform the job. The hiring process may appear objective, but it cannot be considered truly fair if success depends on access to AI tools rather than job-related qualifications, leading to unequal outcomes among groups protected under anti-discrimination laws.

The four-fifths rule is designed to detect this type of adverse impact. However, a standard audit looks at groups like race, sex, and age — not whether candidates used AI or which AI model they used. As a result, AI self-preferencing can easily go undetected.

Why won’t your four-fifths audit catch self-preferencing?

The challenge comes down to data. Nothing on a candidate’s application records whether AI was used to write the resume or which AI model was used, so an adverse impact audit has nothing to measure.

That’s what makes AI self-preferencing so difficult to detect. The bias can be real and still go unnoticed because you can’t catch what you aren’t tracking. Building a bias-free hiring process means looking beyond traditional audits to identify the new blind spots AI can introduce, not just the ones you already know to look for.

Measure people, not resumes

AI can help streamline hiring, but it shouldn’t be the primary factor determining who advances. The goal is to evaluate the person, not how well they used AI to write a resume.

Pre-employment assessments measure job-related capabilities directly, including cognitive ability, personality, work ethic, and other behavioral characteristics shown to predict performance. Unlike resumes, these capabilities remain reliable predictors regardless of how a candidate used AI during the application process.

At Prevue, our assessments are scientifically validated and monitored for adverse impact across gender, ethnicity, and age. Candidates are evaluated against a customized benchmark built for the role, helping employers focus on what matters most: who can perform the job and who is likely to succeed over the long term.

The answer isn’t to abandon AI. It’s to use it where it adds value while relying on validated assessments to evaluate job-related qualifications. Together, they provide the best of both worlds: the efficiency of AI and the confidence of knowing who can actually do the job.

Frequently asked questions

What is AI self-preferencing?

It’s the tendency of an AI model to rate its own writing higher than equally strong writing from a person or another AI. In hiring, an AI screener favors resumes that were written with the same kind of tool.

Can AI self-preferencing change hiring decisions?

Yes. Across 24 different jobs, researchers found candidates who used the same AI tool as the screener were 23% to 60% more likely to be shortlisted than equally qualified candidates who wrote their own resume.

Can a four-fifths or adverse impact audit catch AI self-preferencing?

Not on its own. Standard audits compare protected groups and don’t track which AI tool wrote a resume, so self-preferencing can slip through.

How do you reduce AI self-preferencing in hiring?

Rely less on resume screening and use pre-employment assessments that measure job-relevant ability directly. Researchers also found that telling the AI to ignore whether a resume was AI-written cut the bias by more than half.