From scripted responses to emotional reactions: exploring customer incivility behavior and emotions during chatbot capability failures—a machine learning approach

Purpose This study examines how chatbot capability failures lead to customers' incivility in AI-enabled service delivery through customers' emotional responses. It also investigates whether chatbot-expressed empathy, as a recovery-oriented capability, weakens or intensifies the relationship between customers' negative emotions and incivility. Design/methodology/approach Drawing on social response theory, we analyzed 32,586 text conversations from a university library chatbot using a framework that combines natural language processing and supervised machine learning to extract the constructs of interest. We then estimated a moderated mediation model using PROCESS to examine the proposed relationships. Findings The study identified chatbot capability failures and empathy, customer digital incivility and customer emotional profiles. The findings showed that technical and functional chatbot capability failures elicited negative customer emotions, and that these negative emotions fully mediated the relationship between chatbot capability failures and customer incivility behaviors. Although we anticipated that chatbot empathy, as a recovery-oriented capability, would mitigate the effect of negative emotions on incivility, the findings revealed the opposite: empathy strengthened rather than attenuated this effect. We termed this paradox the “empathy backfire,” indicating that empathic expressions alone were insufficient and could even intensify customer incivility when they were not matched by adequate chatbot capability. Originality/value The study advances social response theory by revealing a counterintuitive empathy backfire effect in which chatbot empathy intensifies, rather than mitigates, customer incivility behaviors. Although empathy often helps mitigate the negative effects of service failure on customer behavior in human interactions, managers should not assume that empathetic language embedded in chatbot scripts will produce comparable benefits.