The perception of humanness in a conversational agent (CA) has been shown to strongly impact users' processing and reaction to it. However, it is largely unclear how this perception of humanness influences users' processing of errors and subsequent intention for negative word-of-mouth (WoM). In this context, we propose two pathways between perceived humanness and negative WoM: a cognitive pathway and an affective pathway. In a 2x2 online experiment with chatbots, we manipulated both the occurrence of errors and the degree of humanlike design. Our findings indicate that perceived humanness effects users' intentions towards negative WoM through the cognitive pathway: users' confirmation of expectations is increased by perceived humanness, reducing negative WoM intentions. However, it has no effect on users' anger and frustration and does not interact with the effects of errors. For practice, our results indicate that adding humanlike design elements can be a means to reduce negative WoM.