Notes
The influence of interaction mechanism and social error responses on the acceptance of customer service chatbots
30As improvements in artificial intelligence will continue to drive the proliferation of chatbots, it is crucial to systematically investigate how users perceive and evaluate them. Chatbots are automated agents that interact with users via natural language (Shawar & Atwell, 2007). Free-text interactions with chatbots often cause users to abandon conversations (Klopfenstein et al., 2017). Structuring elements like buttons might clarify dialogues but can negatively affect perceived humanness (Diederich et al., 2019). Adequate error responses from chatbots can prevent users from abandoning conversations (Benner et al., 2021). In this study, we use actual chatbot interactions to examine how interaction mechanism and social error responses, i.e., offering an apology, influence users’ (a) warmth and (b) competence perceptions, (c) satisfaction, (d) intention to use, and (e) emotional connection with the company.
31According to the CASA paradigm and social response theory, humans apply the same rules in interactions with machines as they do with other humans. Chatbots provide social cues that elicit social responses. The more social cues a chatbot presents, the more likely humans are to act socially toward it (Nass & Moon, 2000). Thus, we expected users to perceive free-text (vs. button) interactions and social (vs. neutral) error responses as more human-like, leading to higher scores on (a)-(e) (H1, H2). Moreover, we expected the effect of social error responses on all outcomes to be stronger in the free-text interaction conditions (H3). In addition, we expected warmth and competence to mediate the effects of free-text interaction and social error response on (c)-(e) (H4, H5).
32A preregistered online experiment with a 2 (interaction mechanism: buttons vs. free-text) x 2 (error response: neutral vs. social) between-subjects design was conducted in October 2022 (N=416). We asked participants to solve customer service tasks using a state-of-the-art chatbot. Participants then completed a questionnaire containing outcomes and sociodemographic variables.
33Artificial intelligence / chatbot / interaction mechanism / error response / user acceptance