Beyond the Hero: The Heroine’s Journey in Game Narratives Generated by LLMs

Abstract

In light of the ethical debates surrounding fairness in Artificial Intelligence (AI) systems and their growing use in digital games, it becomes essential to investigate the presence and the propagation of structural biases in the content generation for this field. This study analyzes how six of the most widely used LLMs construct characters and narratives, contrasting the structures of the Hero's and Heroine's Journeys. Through a qualitative analysis of 270 generated narratives, 45 produced by each LLM, character trait codes and thematic axes were extracted. For each model, these 45 narratives were generated in three prompt formats: 15 with no gender identification for the protagonist, 15 with a female protagonist, and 15 with a male protagonist, to explore whether their formulation influences bias. The results reveal that although LLMs generate more female protagonists, masculinity is coded as proactive and resourceful. In contrast, female characters are restricted to emotional containment, the romantic support stereotype, and journeys of protection.