Evolving Risk Communication: Analyzing Risk Response Effectiveness of Personalized Narrative Messages

Effective risk communication is critical in motivating to take protective actions during emergencies. In this paper, we evaluate the impact of personalized narrative frameworks generated using large language models (LLMs) on individuals’ willingness to respond to threats across three domains: phishing, flooding, and active shooting. Building on our previous work, we tested five message types, including general narrative, strict science-based, and three individualized variants, using a withinsubjects survey of $\mathbf{7 3}$ participants. Our results indicate that individualized messages, especially those incorporating demographic, interest, and value data, consistently outperformed generic and factual formats in motivating action. Analyses revealed significant variations by race and gender, with personalized narratives showing stronger internal consistency and domain-specific effectiveness. The most customized framework (called as ind npf: all) emerged as the most reliable and engaging across demographics and domains. This work highlights the promise of LLM-powered personalized messaging in enhancing public preparedness and tailoring emergency communication strategies to diverse audiences.