AI or human? Exploring the effects of user awareness in conversational dynamics with virtual avatars

In plain English

Forty-two participants held a conversation with the same virtual avatar, driven either by ChatGPT-4o or by a human experimenter; half were told which, half were not. Those who were not told could not tell the difference above chance. Those who were told reported stronger positive emotion, greater social presence and stronger attribution of emotions and intentions when the partner was human. Disclosure, in other words, did more to shape the experience than the avatar's surface realism — useful ammunition in the recurring argument about whether labelling an agent as AI ruins the interaction. The study used 42 female participants in a single topic domain, so treat it as a directional finding rather than a settled one.