How corrective messages persuade in AI era: message sidedness, source attribution, and psychological pathways in health misinformation correction

Vollständiger Abstract

Worum geht es in dieser Arbeit?

Introduction In contemporary AI-mediated digital environments, corrective health information is encountered through diverse sources and message formats, raising questions about how message design and source attribution shape belief endorsement. Drawing on persuasion theory, this study examines the effects of corrective message sidedness and source attribution on endorsement of a corrected health claim. Methods A 2 × 3 factorial experiment was conducted with 543 participants. Corrective message sidedness was manipulated as one-sided or two-sided, while source attribution identified the correction as originating from an LLM chatbot, a government health authority, or an unspecified source. The correction addressed a health claim concerning influenza vaccination during pregnancy. Factorial analysis of variance was used to test the main and interaction effects of message sidedness and source cues. Mediation analyses were conducted to examine the roles of perceived persuasiveness and perceived source credibility. Results Significant main effects were found for both message sidedness and source cues. Two-sided corrective messages generated higher belief endorsement than one-sided messages, while corrections attributed to a government health authority generated higher belief endorsement than those attributed to an LLM chatbot or an unspecified source. The interaction between message sidedness and source cues was not significant. Mediation analyses further showed that perceived persuasiveness mediated the effect of message sidedness on belief endorsement, whereas perceived source credibility mediated the effect of source attribution. Discussion Corrective health messages influence belief endorsement not only through the information they provide but also through distinct psychological processes associated with argument evaluation and credibility inference. The findings advance understanding of health misinformation correction in AI-mediated environments and provide practical guidance for designing more effective corrective messages in digital health communication.