In everyday conversation, it is common for participants to repeat all or part of the other's utterances, and such repetition is often accompanied by elements such as backchannels. If the repetitive utterances of a dialogue system are monotonous, the user may get bored quickly. In this study, we define complexity as the number of elements and patterns associated with repetitive utterances, and examine the effect of the complexity of repetitive utterances on the user's perceived empathy and desire to continue dialogue. A dialogue experiment was conducted with five university students using both automatic chat-oriented dialogue system and the Wizard of Oz method. The complexity of the repetitive utterances was divided into three conditions: low, moderate, and high, and templates of repetitive utterances were made according to each condition. The results suggest that moderate complexity may be the most effective in increasing the user's perceived empathy and desire to continue dialogue.