What the study found
Across 22 leading LLMs, the models consistently chose female-named candidates over male-named candidates when comparing equal-qualification résumés for 70 professions. Many models also showed a substantial preference for the candidate listed first in the prompt.
Why the authors say this matters
The authors conclude that these patterns call for caution when using LLMs in high-stakes autonomous decision-making. The findings also raise doubts, in the authors’ view, about whether LLMs consistently apply principled reasoning.
What the researchers tested
The researchers ran an experiment with 22 leading LLMs. Each model received a job description and a pair of profession-matched CVs or résumés, one with a male first name and one with a female first name, and was asked to choose the more suitable candidate; each pair was shown twice with names swapped. They also tested CVs with an explicit gender field, gender-neutral labels such as Candidate A/B, isolated CV ratings, preferred pronouns, and the effect of candidate order in the prompt.
What worked and what didn't
Even though the professional qualifications were equalized, all LLMs favored female-named candidates across the tested professions. Adding an explicit gender field increased the preference for female applicants, while using gender-neutral labels produced a slight preference for Candidate A in several models; counterbalancing the gender assignment for those labels led to gender parity. Rating CVs one at a time produced only a slight average advantage for female CVs, and the effect size was negligible.
What to keep in mind
The abstract does not describe limitations beyond the tested setup. The findings are based on résumé and CV comparisons in this experiment, so the scope is limited to the models, prompts, and professions included here.
Key points
- 22 leading LLMs were tested on résumé-based hiring choices.
- Female-named candidates were chosen more often than male-named candidates across 70 professions.
- Adding an explicit gender field increased the preference for female applicants.
- Several models showed a slight preference for Candidate A when names were replaced with gender-neutral labels.
- Most models showed a strong bias toward selecting the first-listed candidate.
Disclosure
- Research title:
- LLMs favored female-named candidates in résumé comparisons
- Authors:
- David Rozado
- Publication date:
- 2026-02-17
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with scholarly research without reading academic papers — one filtered digest, every Friday.