Interviewing AI-Generated Personas: Talking to Your Data to Generate Qualitative Text from Users

We introduce a qualitative data generation method, Persona Interviews, interviewing AI-generated personas created from large-scale survey data. Surveys often provide quantitative breadth but limited interpretive depth. User interviews have limited breadth but can elicit deep insights. Persona Interviews bridges these paradigms by generating data-driven personas from survey data, each representing a distinct user segment, and instantiating them as LLM-powered chatbots for semi-structured interviews. As a critical case study, we apply this method to a survey of more than 8,000 respondents across 16 countries in the Middle East and North Africa (MENA) focused on social media use and privacy concerns. From the survey data, we construct 16 representative personas, one per country, and interview each using a consistent set of eighteen qualitative questions. We analyze the AI-persona responses for distinctiveness and accuracy. Results show that word counts were generally comparable across personas and that responses exhibited high accuracy for the underlying survey data, with factual data accuracy of 90.4% and perceptual data accuracy of 94.4%. Findings show that interviews with AI-personas can extend traditional survey analysis by providing contextual and accurate user narratives for qualitative analysis and insights.