AI job screeners prefer AI-written resumes over human ones, researchers find

A study titled *AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights* revealed that AI-powered applicant tracking systems (ATS) systematically favor resumes written by the same large language models (LLMs) they use. Researchers Jiannan Xu of the University of Maryland, Gujie Li of the National University of Singapore, and Jane Jiang of Ohio State University found these systems were 23% to 60% more likely to shortlist candidates whose resumes were generated by identical LLMs, such as GPT-4o or Deepseek-V3.1. The bias was most pronounced in accounting, sales, and finance roles, raising fairness concerns for job seekers and potential risks for employers overlooking strong candidates. The study, published on arXiv.org in February, warns that unaddressed self-preferencing could distort hiring outcomes across industries, including education and publishing. The researchers tested 2,245 human-written resumes, creating AI-generated counterfactual versions to simulate hiring pipelines for 24 occupations. The findings suggest AI evaluators prioritize resumes that align with their own output, rather than assessing true qualifications. Boston University’s Emma Wiles, an expert in AI’s labor market impact, cautioned that AI tools may not accurately reflect applicants’ abilities but instead favor outputs resembling the AI itself. She advised job seekers to use AI as a writing aid rather than a replacement, emphasizing that human effort should remain central to resume creation. The study highlights broader implications for industries facing AI-driven hiring, where over 300,000 job cuts were reported in tech alone between January and April 2026, according to Challenger, Gray & Christmas. Researchers urge addressing this bias to ensure fair and equitable hiring processes.