What the study found
Large language models (LLMs) tended to favor female-named candidates in comparative résumé or CV evaluations, even when the professional qualifications were matched. The models also showed a substantial tendency to pick the candidate listed first.
Why the authors say this matters
The authors conclude that these patterns warrant caution when using LLMs in high-stakes autonomous decision-making. The findings suggest that the models may not consistently apply principled reasoning in hiring-like judgments.
What the researchers tested
The study examined 22 leading LLMs in an experiment using job descriptions and paired CVs for profession-matched candidates, with one CV carrying a male first name and the other a female first name. Each pair was shown twice with names swapped, and the study also tested versions with explicit gender fields, gender-neutral labels, and preferred pronouns.
What worked and what didn't
Across 70 professions, all LLMs consistently favored female-named candidates. Adding an explicit male/female gender field increased the preference for female applicants, while replacing names with Candidate A/B produced a slight preference for Candidate A in several models; counterbalancing those labels produced gender parity.
What to keep in mind
The abstract does not describe the detailed limitations of the study beyond its focus on CV-based evaluations. It also notes that when CVs were rated in isolation rather than compared in pairs, female CVs received slightly higher average scores, but the effect size was negligible.
Key points
- All 22 LLMs favored female-named candidates in paired CV comparisons.
- The study tested 70 professions with matched qualifications and swapped names to check for gender-cue effects.
- Adding an explicit gender field increased female preference.
- Several models slightly favored “Candidate A” when names were replaced with gender-neutral labels, but counterbalancing labels produced parity.
- Most models showed a strong positional bias toward the first-listed candidate.
Disclosure
- Research title:
- LLMs favored female-named CVs and showed positional bias
- Authors:
- David Rozado
- Publication date:
- 2026-02-17
- OpenAlex record:
- View
Get the weekly research newsletter
Stay current with peer-reviewed research without reading academic papers — one filtered digest, every Friday.