Nobody Talks About the Prompt Club: Evaluating Diverse Persona Generation Through System Prompts and Guidelines
Personas are widely used to represent users' motivations, goals, and needs, yet the perspectives of underrepresented groups are rarely addressed in human-centered design. Large Language Models (LLMs) promise to generate personas from vast datasets, but achieving genuine diversity and accurate representation requires deliberate prompting. In this work, we investigate two strategies, namely system prompts and guidelines, to support users in crafting diverse personas and evaluate their effects in a controlled user study $(\mathbf{N}=\mathbf{6 6})$. Our results show that both strategies help users generate more diverse personas, with their combination achieving the highest coverage of diversity aspects. Consequently, this work aims to go beyond evaluating LLMs as tools for persona creation. We show and discuss strategies to support requirements engineers during their interaction with the LLM.