GenAI Social Actors in Customer Services: How Customer Perceptions Vary Across Interaction Formats
Abstract: Generative AI has the potential to fundamentally transform customer service by enabling artificial social actors to provide individualized frontline interactions at unprecedented scale. Realizing this potential requires firms to decide through which formats customers should encounter these agents. We focus on three synchronous formats in which customers communicate with the GenAI social actor through speech: voice-only interaction, interaction with an onscreen avatar, and interaction in immersive virtual reality. These formats differ substantially in the amount and intensity of the social cues they provide, yet it remains unclear how these differences affect customers’ perceptions and service outcomes.
Drawing on media-as-social-actors theory, we examine customer responses to the same GenAI social actor across these three formats. We conducted a two-stage randomized framed field experiment with 534 students in a supplementary educational service setting. Participants were first assigned either to a lecture-only baseline condition (n = 239) or to an interactive GenAI tutorial (n = 295). Tutorial participants were subsequently randomized to voice (n = 99), screen (n = 100), or virtual reality (n = 96). The GenAI social actor conducted a fully interactive tutorial lasting 10 to 15 minutes, during which students could ask questions while the agent kept the conversation focused on the course content.
The tutorial improved objective performance relative to the baseline, establishing the value of the supplementary service itself. Yet adding social cues did not consistently improve customer service outcomes. Compared with voice, the onscreen avatar increased eeriness without increasing social presence or sensory intensity. Compared with the screen format, virtual reality increased social presence and sensory intensity, but also heightened eeriness and exhaustion. These countervailing mechanisms produced no gains in objective performance or satisfaction across interaction formats, while virtual reality reduced reuse intentions relative to both voice and screen. The findings show that enriching GenAI interactions with additional social cues does not necessarily improve customer responses. Effectiveness depends on whether the experiential benefits of added cues outweigh the discomfort and demands associated with more humanlike and immersive interaction formats.
Thorsten Hennig-Thurau