Previous studies have found that individuals react differently to online customer service interactions with chatbots than human agents. However, it remains unclear which agent is more effective at mitigating negative consumer evaluations in the context of service failures. This article explores whether online customer service agents, informed by social perception theory, can influence consumer evaluations by simulating service failure scenarios for future reference. In this study, human participants responded to a service failure interaction task with the Wizard-of-Oz method. This study conducted the pretest to verify the validity of stimuli and the main experiment to verify the proposed hypotheses. Results indicated that participants in human conditions were more likely to mitigate negative evaluations brought about by service failures, while a chatbot aggravated this negative effect. These findings contribute to a deeper understanding of the dynamics between online chatbots and human agents in the context of service failures. Interactions with human agents can mitigate consumers' negative evaluations of services, as human agents are perceived to demonstrate higher competence than chatbots.