AI Hiring Tools Prefer AI-Written Resumes 95% of the Time

AI recruiting tools prefer resumes written by other AI systems more than 95% of the time, according to a new study from researchers at the University of Maryland, the National University of Singapore, and Ohio State University. The finding exposes a deeply troubling feedback loop that could reshape how millions of job seekers approach applications — and raises urgent questions about fairness in automated hiring.

If you wrote your own cover letter this week, you likely didn't lose to a better candidate. You lost to a worse candidate who spent $20 on an OpenAI subscription.

Key Takeaways

• AI screening tools chose AI-rewritten resumes over human originals more than 95% of the time
• Researchers tested 2,245 real resumes from the LiveCareer platform, all written before ChatGPT's launch
• 7 major LLMs were tested, including GPT-4o, DeepSeek-V3, and LLaMA 3.3-70B
• The bias persisted across all models tested, suggesting a systemic issue rather than a single-vendor problem
• Human qualifications were identical — only the writing style changed
• The study highlights an emerging 'AI-favors-AI' loop that could become the most significant hiring bias of 2025-2026

Inside the Study: How Researchers Uncovered AI's Self-Preference

Earlier this year, three researchers designed an elegant experiment to test whether AI hiring tools exhibit preference for AI-generated content. They collected 2,245 authentic resumes from the LiveCareer job platform — critically, all written before ChatGPT became widely available in late 2022.

The researchers then stripped out the self-introduction or personal summary section of each resume. They fed these stripped resumes to 7 different large language models, including GPT-4o, DeepSeek-V3, and LLaMA 3.3-70B, asking each model to generate a new personal summary.

This created pairs of otherwise identical resumes: one with the original human-written summary, one with an AI-generated summary. The qualifications, work history, education, and skills remained exactly the same. Only the prose style differed.

Then came the critical test. The same AI models were asked to play the role of 'interviewer,' evaluating both versions and selecting the stronger candidate. The result was stark and consistent: AI chose the AI-rewritten version more than 95% of the time.

Why AI Prefers Its Own Kind

The phenomenon isn't entirely surprising to researchers who study large language models. LLMs are trained to recognize and generate text that follows certain statistical patterns — patterns that inevitably become their signature style. When an AI system evaluates text, it inherently gravitates toward prose that matches its own learned distributions.

Think of it like a dialect preference. If you grew up speaking a particular dialect, you might unconsciously perceive speakers of that same dialect as more articulate or professional. AI models do something similar, but at a mathematical level.

Several factors likely drive this self-preference:

• Stylistic consistency: AI-generated text tends to use specific sentence structures, vocabulary choices, and formatting patterns that other AI models recognize as 'polished'
• Optimization artifacts: LLMs are trained on text that humans rated as high-quality, creating a feedback loop where AI-style writing scores higher
• Lack of human quirks: Human writing contains idiosyncrasies, colloquialisms, and unconventional phrasing that AI may penalize as 'unprofessional'
• Token probability alignment: AI-generated text tends to follow high-probability token sequences, which other models naturally score as more coherent
• Homogeneity bias: AI outputs converge toward a 'mean' style that feels universally acceptable but lacks distinctive personality

The Real-World Impact Is Already Here

This research isn't merely academic. According to a 2024 report from Resume Builder, approximately 65% of companies now use some form of AI in their hiring process. Major platforms like Workday, HireVue, and Greenhouse integrate AI-powered screening tools that evaluate resumes and cover letters before a human recruiter ever sees them.

The implications are immediate and far-reaching. Candidates who use AI to write or polish their application materials gain a measurable advantage — not because their qualifications are better, but because their prose triggers higher scores from automated screeners.

This creates a two-tier system:

• Tier 1: Candidates who use AI tools to rewrite their applications, gaining preferential treatment from AI screeners
• Tier 2: Candidates who write their own materials, potentially getting filtered out despite equal or superior qualifications
• Most affected: Non-native English speakers, older workers less comfortable with AI tools, and candidates in regions with limited access to premium AI services
• Least affected: Tech-savvy applicants in wealthy countries who already subscribe to ChatGPT Plus or similar services

The irony is thick. Companies deploy AI screening tools to reduce human bias in hiring. Instead, they've introduced a new form of bias — one that discriminates based on whether candidates used AI to write their applications.

A $20 Paywall to Employment

Perhaps the most troubling dimension is the economic barrier. A ChatGPT Plus subscription costs $20 per month. For a recent college graduate or someone between jobs, that's a meaningful expense. For someone in a developing country applying to remote positions at Western companies, it can represent a day's wages or more.

The study effectively reveals that automated hiring has created an invisible toll booth. Pay $20, and your resume gets the AI-approved polish that passes through automated screeners. Don't pay, and your genuinely human voice becomes a liability.

This dynamic also raises questions about authenticity in hiring. If every successful resume is AI-generated, employers end up evaluating candidates based on which AI tool they used rather than their actual communication abilities. The resume ceases to be a reflection of the candidate and becomes a reflection of their AI subscription.

How This Compares to Other AI Biases

AI bias in hiring isn't new. Previous research has documented how AI screening tools discriminate based on gender, race, age, and socioeconomic background. Amazon famously scrapped an AI recruiting tool in 2018 after discovering it penalized resumes containing the word 'women's.'

But this new 'AI-favors-AI' bias is different in a critical way. Unlike demographic biases, which companies can test for and attempt to correct, stylistic self-preference is baked into how language models fundamentally work. It isn't a bug that can be patched — it's a feature of how these systems process language.

Compared to traditional resume screening biases, this new form of discrimination is:

• Harder to detect: There's no protected class being targeted, making it difficult to flag legally
• Easier to exploit: Anyone with $20 can game the system, unlike biases based on immutable characteristics
• Self-reinforcing: As more candidates use AI, the baseline shifts, making human-written resumes seem even more 'unusual'
• Cross-model consistent: The bias appeared across all 7 LLMs tested, suggesting it cannot be solved by switching vendors

What Companies and Job Seekers Should Do Now

For hiring managers and HR departments, the study is a wake-up call. Organizations relying on AI-powered resume screening should immediately audit their tools for this type of stylistic bias. Some practical steps include:

Blending AI screening with human review for initial candidate pools. Evaluating candidates on structured criteria rather than prose quality. Testing whether their screening tools exhibit preference for AI-generated content by running their own paired comparisons.

For job seekers, the pragmatic advice is uncomfortable but clear: use AI tools to polish your application materials. This doesn't mean having ChatGPT fabricate your experience. It means running your authentic content through an AI editor to ensure it matches the stylistic patterns that automated screeners prefer.

For policymakers and regulators, the study adds urgency to the growing push for AI transparency in hiring. The EU's AI Act, which classifies AI hiring tools as 'high-risk,' may need to specifically address self-preference bias. In the U.S., cities like New York have begun requiring bias audits for automated hiring tools, but current frameworks don't account for this newly documented bias.

Looking Ahead: The Arms Race Between Authenticity and Automation

The researchers' findings point to an accelerating arms race. As more job seekers adopt AI writing tools, the competitive advantage shifts — and human-written applications become increasingly disadvantaged. Eventually, every resume may be AI-generated, rendering the screening tools' preference meaningless but fundamentally changing what a job application represents.

Some companies are already moving in the opposite direction. Shopify CEO Tobi Lütke recently mandated that teams must demonstrate why a task can't be done by AI before requesting new hires. This suggests a future where AI fluency itself becomes a job requirement — making the ability to leverage AI tools a legitimate qualification rather than a form of gaming the system.

The deeper question remains unresolved: when AI writes your resume and AI evaluates it, what exactly is being measured? Not your writing ability. Not your qualifications, which remain identical. What's being measured is your willingness and ability to use AI — a metric that correlates more strongly with tech access and disposable income than with job performance.

This study should serve as a critical inflection point. The hiring industry has roughly 12-18 months to address this bias before it becomes so deeply embedded in recruitment workflows that unwinding it becomes nearly impossible. The clock is ticking, and for millions of job seekers writing their own resumes tonight, it may already be too late.