Who Gets the Callback? Generative AI and Gender Bias
Sugat Chaturvedi, Rochana Chaturvedi
TL;DR
This work assesses gender bias in several mid-sized open-source LLMs deployed for candidate shortlisting, using 332,044 real job ads from India and eliciting gendered callback decisions to quantify bias, segregation, and wage effects. It integrates SOC mapping via semantic embeddings, wage regressions, lexical/LIWC analyses, and prompt-based persona steering (Big Five traits and historical figures) across over 40 million prompts to reveal how model decisions align with traditional gender stereotypes. The findings show substantial bias in many models, manifested as higher female penalties in wages and stronger occupational segregation, with some models showing bias in favor of women but often still entangled with occupation-specific disparities; intriguingly, recruiter-like personas and certain prompts can modulate or even undermine safeguards. The study offers a scalable auditing framework and actionable guidance for designing fair AI hiring tools, highlighting the need for policy-aware deployment to mitigate labor-market biases.
Abstract
Generative artificial intelligence (AI), particularly large language models (LLMs), is being rapidly deployed in recruitment and for candidate shortlisting. We audit several mid-sized open-source LLMs for gender bias using a dataset of 332,044 real-world online job postings. For each posting, we prompt the model to recommend whether an equally qualified male or female candidate should receive an interview callback. We find that most models tend to favor men, especially for higher-wage roles. Mapping job descriptions to the Standard Occupational Classification system, we find lower callback rates for women in male-dominated occupations and higher rates in female-associated ones, indicating occupational segregation. A comprehensive analysis of linguistic features in job ads reveals strong alignment of model recommendations with traditional gender stereotypes. To examine the role of recruiter identity, we steer model behavior by infusing Big Five personality traits and simulating the perspectives of historical figures. We find that less agreeable personas reduce stereotyping, consistent with an agreeableness bias in LLMs. Our findings highlight how AI-driven hiring may perpetuate biases in the labor market and have implications for fairness and diversity within firms.