INVISIBLE BIAS: GENDERED REPRESENTATIONS IN AI-GENERATED TEXT

Abstract

This paper explores gender bias in AI-written texts using ChatGPT (GPT 5.2) and focus on how both male and female subjects are represented in the context of professional, social and personal life. Based on Feminist Critical Discourse Analysis (FCDA), the study discusses how language and discursive patterns depict the underlying gender ideologies and power relations. The AI generated descriptions of men and women were compared, using ten parallel prompts, in the role of CEOs, political leaders, teachers, parents, scientists and friends. The results show that although both genders are portrayed as competent, male subjects are portrayed as decisive, rational and action-oriented, and female subjects are portrayed as empathetic, relational and ethically conscious. These trends demonstrate gender conventions in society in AI output, which points to the ethical and social consequences of biased generative technologies. The research adds qualitative, FCDA-informed evidence to the discussion of AI bias, which can be used to detect bias, reduce it, and develop AI in a socially responsible way.