Talking the talk: how AI voice assistants’ apology styles shape consumer satisfaction in service recovery

Purpose The objective of this study is to explore the impact of AI voice assistants’ anthropomorphic apology styles on consumer recovery satisfaction. Design/methodology/approach The study adopts three scenario-based experiments to examine the effects of anthropomorphic apology styles on consumer recovery satisfaction. Experiment 1 involved 108 university students, while Experiments 2 and 3 collected data from 222 and 288 AI voice assistant users, respectively, recruited via an online survey platform (sojump.com). SPSS was used for data analysis. Findings The results confirmed that cute apology styles result in higher consumer recovery satisfaction than literal ones. Both cute and literal apology styles improve consumer satisfaction by enhancing perceived competence and warmth; however, warmth plays a more critical role, particularly in the case of cute apologies. Moreover, the positive impacts of anthropomorphic apology styles on perceived warmth are stronger in cases of lower severe service failures, confirming the moderating role of failure severity. However, the moderation in the relationship between anthropomorphic apology style and perceived competence is not significant. Practical implications This study provides important managerial guidelines for AI voice assistant design and AI recovery service. Originality/value This study extends the application of social cognitive theory to the domain of AI service recovery and fills the gap in the literature by examining how different apology styles of AI voice assistants impact consumer satisfaction and the moderating effect of the severity of failure, thus addressing a significant gap in the literature.