Gender Bias in Generative AI: An Analysis of Recruitment Processes

In recent years, generative artificial intelligence (Gen AI) systems have assumed increasingly crucial roles in selection processes, personnel recruitment and analysis of candidates’ profiles. However, the employment of large language models risks reproducing, and in some cases amplifying, gender stereotypes and bias already present in labour market. This research aims to evaluate and measure this phenomenon, analysing how a state-of-the-art generative model (ChatGPT-5) suggests occupations and represents the ideal candidate based on gender and work experience background, focusing on under 35 years old Italian graduates. The study is organised into two complementary phases. In Phase I, the model has been trained to provide job suggestions to 24 simulated candidates profiles of female and male genders, balanced in age, experience and professional field. The output variables - as job title, industry and descriptive adjectives - were coded using open coding and tested statistically with χ^2 test. Results show that, although no significant differences emerged in job titles and industry, gendered linguistic patterns exist in the adjectives attributed to female and male candidates, indicating a tendency of the model in associating women to emotional and empathetic traits, while men to strategic and analytic ones. Phase II employed 114 LinkedIn job advertisements, used as prompts for generating textual and visual representations of ideal candidates. The analysis of the outputs highlighted a clear gender polarisation: the model assigned 71% of profiles to male and 29% to female gender. The strongest association emerge in technology and engineering sectors, where male candidates prevail and in HR and commercial functions, where female representation prevails. The visual analysis confirms the perpetuation of gender stereotypes: female profiles are more frequently depicted smiling, in approachable postures and dressed elegantly, while men portraits result more focused and assertive with formal clothing style. These results prove that Gen AI models do not simply reflect the gender biases of the training data, but they also amplify them, as the experiment in the labour market context clearly shows. The research raises an ethical question regarding the use of these models, highlighting the need for transparency and bias mitigation strategies to ensure fairness and inclusive representation.