Personality Shapes Gender Bias in Persona-Conditioned LLM Narratives Across English and Hindi: An Empirical Investigation
Tanay Kumar, Shreya Gautam, Aman Chadha, Vinija Jain, Francesco Pierri
2026-04-28
Summary
This research investigates how giving Large Language Models (LLMs) different 'personalities' affects whether they show gender biases when writing stories. It looks at stories generated in both English and Hindi, focusing on professionals in India.
What's the problem?
LLMs are being used more and more in applications where they pretend to *be* someone – like a teacher or customer service rep. While this can make interactions better, it’s unclear how these assigned personalities might worsen existing gender stereotypes. The core issue is that LLMs might reinforce harmful biases when asked to act as different people, and we don't fully understand how personality traits play into this.
What's the solution?
The researchers created a controlled experiment where they asked six different LLMs to write over 23,000 stories about working professionals in India. They specifically varied the gender, job, and personality of the characters in the stories, using established personality frameworks like HEXACO (which measures traits like honesty and emotionality) and the Dark Triad (which measures traits like narcissism and manipulation). Then, they analyzed the stories to see if certain personality traits were linked to stronger gender stereotypes.
Why it matters?
This study shows that gender bias in LLMs isn't fixed; it changes depending on the personality the model is given. Specifically, 'darker' personality traits tend to lead to more stereotypical portrayals of genders. This is important because it means that applications using these models could unintentionally reinforce harmful stereotypes in things like educational materials, professional reports, or even social media content, creating unfair representations.
Abstract
Large Language Models (LLMs) are increasingly deployed in persona-driven applications such as education, customer service, and social platforms, where models are prompted to adopt specific personas when interacting with users. While persona conditioning can improve user experience and engagement, it also raises concerns about how personality cues may interact with gender biases and stereotypes. In this work, we present a controlled study of persona-conditioned story generation in English and Hindi, where each story portrays a working professional in India producing context-specific artifacts (e.g., lesson plans, reports, letters) under systematically varied persona gender, occupational role, and personality traits from the HEXACO and Dark Triad frameworks. Across 23,400 generated stories from six state-of-the-art LLMs, we find that personality traits are significantly associated with both the magnitude and direction of gender bias. In particular, Dark Triad personality traits are consistently associated with higher gender-stereotypical representations compared to socially desirable HEXACO traits, though these associations vary across models and languages. Our findings demonstrate that gender bias in LLMs is not static but context-dependent. This suggests that persona-conditioned systems used in real-world applications may introduce uneven representational harms, reinforcing gender stereotypes in generated educational, professional, or social content.