Gender Bias in Generative AI-assisted Recruitment Processes
arXiv:2603.11736v1 Announce Type: new Abstract: In recent years, generative artificial intelligence (GenAI) systems have assumed increasingly crucial roles in selection processes, personnel recruitment and analysis of candidates' profiles. However, the employment of large language models (LLMs) risks reproducing, and in some cases amplifying, gender stereotypes and bias already present in the labour market. The objective of this paper is to evaluate and measure this phenomenon, analysing how a state-of-the-art generative model (GPT-5) suggests occupations based on gender and work experience background, focusing on under-35-year-old Italian graduates. The model has been prompted to suggest jobs to 24 simulated candidate profiles, which are balanced in terms of gender, age, experience and professional field. Although no significant differences emerged in job titles and industry, gendered linguistic patterns emerged in the adjectives attributed to female and male candidates, indicating a
arXiv:2603.11736v1 Announce Type: new Abstract: In recent years, generative artificial intelligence (GenAI) systems have assumed increasingly crucial roles in selection processes, personnel recruitment and analysis of candidates' profiles. However, the employment of large language models (LLMs) risks reproducing, and in some cases amplifying, gender stereotypes and bias already present in the labour market. The objective of this paper is to evaluate and measure this phenomenon, analysing how a state-of-the-art generative model (GPT-5) suggests occupations based on gender and work experience background, focusing on under-35-year-old Italian graduates. The model has been prompted to suggest jobs to 24 simulated candidate profiles, which are balanced in terms of gender, age, experience and professional field. Although no significant differences emerged in job titles and industry, gendered linguistic patterns emerged in the adjectives attributed to female and male candidates, indicating a tendency of the model to associate women with emotional and empathetic traits, while men with strategic and analytical ones. The research raises an ethical question regarding the use of these models in sensitive processes, highlighting the need for transparency and fairness in future digital labour markets.
Executive Summary
This study addresses a critical emerging issue in the intersection of AI and employment: the potential for generative AI systems to reproduce or amplify gender bias in recruitment. Using GPT-5 to evaluate occupation suggestions for 24 simulated Italian graduate profiles—balanced by gender, age, experience, and field—the research uncovers subtle but significant linguistic patterns. While job titles and industries remained neutral, the model’s adjective assignments revealed a persistent bias: female candidates were disproportionately linked with emotional and empathetic descriptors, and male candidates with strategic and analytical ones. This subtle reinforcement of gendered stereotypes, even in algorithmic recommendations, signals a systemic risk in AI-assisted hiring. The findings underscore the urgent need for algorithmic transparency and bias mitigation protocols in digital labor markets. The study is methodologically sound, leveraging controlled simulations to isolate bias effects without external confounding variables, making its results credible and actionable.
Key Points
- ▸ GPT-5 exhibits gendered linguistic bias in occupation suggestions despite balanced candidate profiles
- ▸ Adjective attribution patterns reveal stereotypical associations: women with emotional traits, men with analytical ones
- ▸ Bias persists in algorithmic recommendations even when demographic variables are controlled
Merits
Methodological Rigor
The use of simulated, balanced candidate profiles allows for controlled isolation of bias effects, enhancing validity and generalizability.
Demerits
Limited Scope
The study focuses on a specific demographic (Italian graduates under 35) and a single model (GPT-5), limiting applicability to broader international or multi-model contexts.
Expert Commentary
The paper presents a compelling and timely contribution to the field of AI ethics in labor. What is particularly noteworthy is the subtlety of the bias: it does not manifest as overt exclusion or discriminatory language, but rather through the normalization of gendered descriptors—a phenomenon that is arguably more insidious because it operates beneath the threshold of conscious detection. This aligns with broader feminist critiques of AI, which have long warned that bias in machine learning is not always explicit but often embedded in training data and linguistic paradigms. The authors rightly pivot from measurement to ethics, framing their findings not as a technical glitch but as a moral imperative. Their call for transparency is not merely procedural; it is epistemological—demanding that stakeholders understand how AI constructs perceptions of competence and suitability. This work sets a benchmark for future research in algorithmic labor governance and should inform both academic discourse and policy drafting on AI in HR.
Recommendations
- ✓ 1. Integrate bias detection algorithms into HR AI platforms to flag gendered linguistic patterns in real-time recommendations
- ✓ 2. Develop industry-wide certification standards for AI recruitment tools, requiring third-party audits for gender bias mitigation