From Behavioral Instructions to Perceptual Goals: Manipulating Anthropomorphism in Human-AI Communication Experiments

Researchers commonly manipulate chatbot anthropomorphism through social cues, whose effects vary and may introduce unintended perceptual differences. Generative AI enables an alternative: manipulating perceived characteristics through natural-language prompting. Two online experiments (N = 84 each) evaluated behavior-prescriptive and perception-goal prompting for manipulating anthropomorphism. Study 1 found no significant differences in anthropomorphism. Study 2 found significant differences in general, mindful, and mindless anthropomorphism (ds = .86–.95), without significant differences in response coherence or perceived intelligence. These findings provide initial empirical support for perception-goal prompting and suggest broader methodological possibilities for manipulating perceived chatbot characteristics in human–AI communication research.