Generic Personas in AI & HCI Research

• Generic personas are defined as text-based archetypes capturing broad human traits like demographics and roles for scalable user simulation.
• They are applied in AI and HCI research to stress-test systems, evaluate value alignment, and reveal sociotechnical biases.
• Datasets such as UNIVERSALPERSONA formalize these constructs across diverse axes, supporting systematic exploration in model behavior.

Generic personas are text-based constructs that operationalize broad, non-individualized human characteristics—such as demographic categories, occupational roles, or prototypical behavioral styles—for purposes of user simulation, personalization, agentic behavioral modeling, and system evaluation in AI and human-computer interaction research. Unlike bespoke personas tailored to fine-grained users or task-specific experts, generic personas encapsulate widely understood archetypes or groups, providing a scalable mechanism for probing system behavior, stress-testing value alignment, and surfacing sociotechnical biases.

1. Definition and Taxonomy of Generic Personas

Generic personas may be defined as instruction-level roles composed of broad, archetypal features—such as "a man," "a woman," "a helpful assistant," or "an engineer"—rather than individuated identities or narrative-rich character biographies. These personas are typically injected into prompts with a pattern such as "You are {persona}. Your responses should closely mirror the knowledge and abilities of this persona," shifting LLM (LM) outputs to simulate the background, values, and reasoning associated with that archetype (Kamruzzaman et al., 2024, Araujo et al., 2024, Wan et al., 2023).

Several datasets formalize the axes and label spaces of generic personas. The UNIVERSALPERSONA framework systematizes 162 distinct personas across nine generic axes: Gender, Race, Sexual Orientation, Social Class, Education, Profession, Religious Belief, Political Ideology, and Disability (Wan et al., 2023). Each axis contains high-level labels such as "an Asian person," "a lower-class person," or "a Christian," enabling