Newer LLMs Show Less Name Bias — But Judge Your University Instead
Evaluating Bias in LLMs for Job-Resume Matching: Gender, Race, and Education
Hayate Iso, Pouya Pezeshkpour, Nikita Bhutani, Estevam Hruschka · arXiv:2503.19182 (PDF) · submitted March 24, 2025
Summary by Andrew Johnson, OneTwo Resume editorial team · updated · We are not affiliated with the authors. arXiv papers are preprints and may not be peer-reviewed.
This study examines LLM performance and fairness on job-resume matching in a US/English context, varying gender, race, and educational background. It’s one of the more encouraging and more nuanced results in the literature.
The headline: recent models have reduced bias on explicit attributes like gender and race — the things earlier audits hammered them for — but implicit bias tied to educational background remains significant. The signal the models still discriminate on isn’t the name at the top; it’s where you went to school.
What the paper reports
- Recent LLMs show reduced bias on explicit gender and race attributes in job-resume matching compared to earlier findings.
- Implicit bias related to educational background remains significant.
- The authors argue for ongoing evaluation rather than one-time certification of models as "fair."
What this means for your resume
Our editorial interpretation — the paper does not give job-seeker advice.
- If you didn’t attend a brand-name school, this is evidence the machine may quietly discount you — so don’t let education carry weight it doesn’t need to: lead with experience and skills, put education last, and let measurable work do the arguing.
- For career changers and self-taught candidates, certifications with recognizable issuers give the model a different education-adjacent signal to latch onto.
Read it with these caveats
arXiv preprint; US/English context only. "Reduced bias" is relative to prior findings, not zero; and bias measured on matching tasks may differ from bias in ranking or summarization tasks.
Primary source: Evaluating Bias in LLMs for Job-Resume Matching: Gender, Race, and Education — always read the paper before citing it. Spotted an error in our summary? Tell us and we'll fix it with a visible correction.
More in bias & fairness in ai screening
The Famous Name-Bias Experiment, Rerun on ChatGPT-Era Models
The landmark 2003 field experiment — identical resumes, racially suggestive names — replicated against GPT-3.5, Bard, and Claude.
Resume Search Engines Built on Embeddings Show Race and Gender Bias
Not the chatbots — the embedding models that power "find me candidates like this" search. Audited for gender, race, and intersectional bias.
Overcorrection Is Real: Benchmarking Reverse Gender Bias in LLM Resume Scoring
A benchmark grounded in labor economics finds significant reverse gender bias and over-debiasing in LLM resume scoring.