ABSTRACT Drawing on the Uncanny Valley theory, we conducted an online experiment to examine how avatars with varying human‐likeness (i.e., real human, deepfake, human‐like, and cartoon) affect homophily, message credibility, and concept understanding in anti‐scam public service announcements (PSAs). Results showed that the human‐like avatar was perceived as less homophilous but more credible. Additionally, the deepfake condition closely mirrored the outcomes of the real human condition, suggesting it may cross the Uncanny Valley. This research offers timely insights for the deployment of AI‐generated avatars, particularly in guiding how to calibrate avatar realism.
Huang et al. (Wed,) studied this question.