With the rapid development of artificial intelligence technology, conversational AI is increasingly widely used in daily life. However, technical limitations and scenario complexity make AI service failure inevitable. After service failure, how to repair it effectively becomes the key to maintaining user trust. This study takes conversational AI as the research object to explore the impact of different apology strategies and service types on user forgiveness. Through two experiments, the results show that general humility and intellectual humility positively affect user forgiveness; Compared with no apology, both can significantly improve user forgiveness; Service type moderates the effect of apology strategy. In tool-type service, an intellectual humble apology is more effective, while in chat-type service, a humble apology is more favored by users. The results of this study provide empirical evidence and practical guidance for developers of conversational AI to optimize user experience and improve user satisfaction. Developers need to pay attention to users’ perceptions and understandings of different apology strategies and constantly optimize their design and expression to align with users’ expectations and emotional needs.