With the rapid development of artificial intelligence (AI), many companies are investing in AI and deploying AI chatbots in customer service provision. AI customer service agents are expected to effectively help companies reduce labor costs. However, companies may also face pushback from users during the application of AI customer service agents, especially during the occurrence of service failure incidents, which can result in customers’ switching (or reverting) from AI customer service agents to human agents. This can result in an increase of labor costs and a loss of profits. Drawing on social response theory, the current study explores how users' perceptions during their interactions with anthropomorphic AI customer service agents will influence their service recovery expectancy in the context of service failure. We also examine the impact of users’ service recovery expectancy on their decisions regarding whether to proactively re-engage in the interactions with AI agents or switch to the interactions with human agents.