If you have ever wondered whether the resume you spent a weekend polishing is even getting read, here is the uncomfortable update: it probably is not, and the thing reading it has a quiet preference. That preference is not for you. It is for resumes written by another AI.
That is the takeaway from a piece by Suzanne Lucas, the longtime HR columnist behind Evil HR Lady, published in Inc. She walks through new academic research showing that the large language models powering modern resume-screening tools systematically favor resumes that other large language models wrote. The bias is not subtle. In one experiment, GPT-4o picked the AI-written summary 82 percent of the time when the underlying candidate was identical.
The practical effect is that hiring in 2026 has quietly turned into a contest between AI writers and AI screeners, and the human candidate in the middle is the one most likely to lose.
The study behind the headline
The research Lucas cites is a 2025 preprint from researchers at the University of Maryland, the National University of Singapore, and Ohio State, who ran a controlled correspondence experiment across 24 different job roles. They held the underlying candidate constant and varied only the resume summary, swapping a human-written version for an AI-generated one.
When the same models were then used as screeners, they consistently rewarded resumes that matched their own writing style:
- GPT-4o preferred the AI-written summary roughly 82 percent of the time.
- LLaMA 3.3-70B was close behind at 79 percent.
- Other frontier models, including GPT-4-turbo, Qwen 2.5-72B, and DeepSeek-V3, showed self-preference above 65 percent.
Lucas notes that the boost in shortlisting rates ranged from about 23 percent in agriculture roles to 60 percent in sales roles. In other words, the same person, with the same experience, became dramatically more or less hireable depending on who, or what, wrote the top of their resume.
Why this is happening now
The AI screener market and the AI writing market grew up together, and they are starting to look like the same market.
On the screening side, ResumeBuilder reports that 83 percent of companies are using or planning to use AI to evaluate resumes. Most applicant tracking systems now layer LLM-driven ranking on top of older keyword filters.
On the candidate side, HireVue’s 2026 Global AI in Hiring report found that 71 percent of candidates use AI to help write their resumes, and 77 percent of hiring managers admit they do too. Tools like ChatGPT, Gemini, and our own resume builder have moved from optional to baseline.
When most candidates write with AI and most companies screen with AI, you get a feedback loop. The screener was effectively trained on the same patterns the writer is producing, so it scores those patterns higher. The candidates who do not use AI are not just behind on tools. They are writing in a dialect the gatekeeper does not value.
The double bind for job seekers
This is where it gets uncomfortable. AI helps you clear the screener, but raw AI output also gets you rejected by humans.
Resume Now found that 62 percent of employers reject resumes that read as generic AI output, and 78 percent of managers say specific, personalized details are the strongest signal that a candidate actually fits the role. HireVue’s data shows only 44 percent of managers trust AI tools in hiring at all. So the screener loves your AI summary, then a recruiter opens it and bounces it for sounding like every other AI summary.
The candidates winning right now are not the ones using the most AI. They are the ones using AI surgically, then writing the parts that matter themselves.
The 2026 playbook
If you take Lucas’s argument seriously, and the data behind it, there is a clear three-part strategy.
1. Use AI for structure, not substance
Let AI handle the mechanical parts of resume writing. That includes:
- Pulling exact-match keywords from the job description.
- Producing a clean, ATS-readable layout.
- Suggesting section ordering and length.
- Catching obvious gaps or inconsistencies.
This is the part the screener cares about. Aim for an ATS compatibility score above 80 before you send anything out. Below that, you are gambling on the bot getting the benefit of the doubt.
2. Write the experience section yourself
This is the line in the sand. Every bullet under your work history should have three things:
- A concrete metric.
- A timeframe.
- The scope of what you owned.
Not "improved customer satisfaction." Try "raised CSAT from 78 to 91 over six months across a 12-person support team." Not "led marketing initiatives." Try "ran a six-channel campaign that drove 2,400 qualified leads in Q3 against a $40K budget."
AI cannot fabricate these details credibly. That is exactly why they work. They signal to the screener that the resume is high-information, and they signal to the human reviewer that you actually did the job. They are also the easiest thing to defend in an interview, which is where AI-polished candidates fall apart.
3. Run a parallel referral track
The best way to beat the screener is to skip it. Industry data consistently shows referred candidates are roughly four times more likely to be hired than candidates who come in through a job board.
Dedicate maybe a third of your search energy to direct outreach: targeted messages to hiring managers, conversations with people already at the company, and warm intros from your network. The other two-thirds can run through the standard application funnel where the AI-vs-AI dynamic plays out.
What we see in our own data
We build resumes for a living, and the pattern Lucas describes shows up in what we see too. Resumes that pass our automated checks but read as generic in the experience section consistently underperform on interview conversion. The candidates who land offers are the ones whose bullets read like field notes from a real job, not like a polished summary of a real job.
The AI screener is a filter. It is not a judge. The judge is still a person, and that person is reading hundreds of resumes in a row and looking for a reason to put yours on the short pile.
FAQ
Should I stop using AI to write my resume?
No. Not using AI puts you at a measurable disadvantage at the screener stage. Use it for structure, formatting, and keyword fit, then rewrite the parts that describe you specifically.
How do I know if my resume sounds too AI-generated?
Read it out loud. If it sounds like a press release, rewrite it. If your bullets could describe any person who has ever held the job title, rewrite them with metrics, scope, and timeframes that are unique to your work.
Does this affect candidates in every industry equally?
No. The research showed AI-vs-AI bias was strongest in roles like sales, where the boost from AI polish reached 60 percent, and weakest in roles like agriculture, where it was closer to 23 percent. The more a job involves written communication, the more this matters.
What is the single fastest improvement I can make today?
Go to your most recent role and add a metric to every bullet that does not have one. That single change will move your resume from the maybe pile to the yes pile faster than any AI rewrite.
Source: AI Resume Screeners Have a Favorite Candidate: Other AI by Suzanne Lucas, Inc.