Generating Public Health Responses using Survey-Augmented Large Language Models

This study explores the use of large language models (LLMs) to generate synthetic survey responses that reflect real-world health-related decision-making patterns, specifically in vaccination behaviors. By leveraging longitudinal data from the FluPaths surveys and employing a cluster-informed prompting approach, the researchers evaluated various LLMs, finding that while the synthetic data closely matched demographic and belief distributions, they struggled to accurately capture the interplay of these factors within individual respondents. The results indicate that while LLM-generated data can augment exploratory analyses in epidemiological modeling, they require further validation before being considered a replacement for traditional survey data.