Uncanny…But Convincing? Inconsistency Between a Virtual Agent’s Facial Proportions and Vocal Realism Reduces Its Credibility and Attractiveness, but Not Its Persuasive Success

Abstract

Embodied agents—i.e. digital systems represented by a virtual or robotic body—are used as persuasive tools in many different contexts. Still, psychological research indicates that for an agent to successfully influence its audience, many design factors have to work together to create a likable and trustworthy impression. Tapping into literature on the uncanny valley phenomenon, which has received only little attention in the field of persuasion research, we advance a consistency perspective that proposes matching levels of modality realism as a main requirement for users’ acceptance. In an online experiment, we invite 107 participants to watch the persuasive speech of a virtual agent, manipulating both its facial proportions and vocal realism in a 2 × 2 between-subject design. Indeed, a mismatch between the realism of both features significantly reduces the agent’s perceived credibility and attractiveness; yet, we observe that neither manipulation actually influences persuasive success in terms of attitude change. A potential explanation for this result pattern is offered by the Elaboration Likelihood Model, assuming that participants focused more on the agent’s message than on peripheral cues to adjust their attitudes.