High- versus low-anthropomorphic-cue packages and health-information trust: social presence and source credibility under explicit AI disclosure

Abstract

AI virtual opinion leaders are increasingly used in health communication, but it remains unclear whether bundled anthropomorphic-cue packages foster initial trust and information-adoption intention when AI identity is explicitly disclosed. Guided by the Computers Are Social Actors paradigm, Social Presence Theory, Source Credibility Theory, and the Information Adoption Model, we conducted a single-factor between-subjects randomized experiment comparing overall high- and low-anthropomorphic-cue packages. The packages differed simultaneously in avatar realism, gesture behavior, interpersonal/linguistic framing, and video duration; AI disclosure and the response-button structure were held constant. A total of 536 valid responses were analyzed using partial least squares structural equation modeling. Social presence, source credibility, trust, and adoption intention were measured by participant self-report immediately after a single branched-video exposure. The overall high-versus-low package contrast increased social presence. Social presence was positively associated with source credibility; source credibility was positively associated with trust in AI-generated health information; and information trust was positively associated with health information adoption intention. The direct effect of the package contrast on information trust was not significant, whereas the bootstrapped serial indirect effect through social presence and source credibility was significant. Within the present Chinese sample and single-exposure setting, the randomized contrast supports an effect of the overall cue package on initial social presence, while the mediation estimates are consistent with the hypothesized indirect pathway to information trust. The design cannot identify the independent contribution of any visual, gestural, linguistic, vocal, interactive, or duration component. The outcome is self-reported adoption intention, not observed health behavior.