Artificial intelligence agents are responsive to identity cues in persuasive health communication
Year: 2026
Type: article
Abstract: Introduction Large language models (LLMs), such as GPTs, are increasingly being considered as tools for simulating human responses to communication. This paper investigates whether LLM-based AI agents can reproduce nuanced, identity-based responses to persuasive health messages, with a focus on political and religious identity cues in COVID-19 vaccine communication. Methods Across two studies, AI agents were assigned political or religious identities and exposed to COVID-19 vaccine messages that varied in the presence and congruence of identity-based cues. Study 1 examined whether politically identified AI agents responded differently to vaccine messages endorsed by politically congruent versus incongruent figures. Study 2 examined whether AI agents assigned varying levels of Christian religiosity responded differently to vaccine messages that included versus omitted religious identity cues. Results In Study 1, AI agents' responses aligned with their assigned political affiliations, with greater persuasion occurring when vaccine messages were endorsed by politically congruent figures. In Study 2, AI agents identified as moderately or highly religious Christians responded more favorably to vaccine messages incorporating religious identity cues than to messages without such cues. Discussion The findings suggest that LLMs can simulate nuanced, human-like responses to identity-based health communication, highlighting their potential value for pretesting and refining public health messages. The results also reinforce the persuasive potential of identity-based cues in health communication. More broadly, the studies demonstrate the promise of LLM-based AI agents as a methodological tool for persuasion and health communication research while pointing to important directions for future research on the validity and appropriate use of simulated human responses.
Source: Frontiers in Communication
Authors: Xiaoli Nan, Xingman Wu
Institution: University of Maryland, College Park
Language: English
Topic: Media Influence and Health
Subfield: Communication
Field: Social Sciences
Domain: Social Sciences
Open Access status: gold
Estimated APC paid: $1,418
OpenAlex ID: W7212011254