Advancements in AI and natural language processing have enabled chatbots to engage in highly human-like conversations. Additionally, many chatbots are intentionally equipped with various social cues, such as human avatars, to further enhance their human-like appearance. While most existing research highlights the social and beneficial perception of humanized chatbots, this thesis critically examines potential limitations. Through three empirical papers applying both qualitative and quantitative methods, this thesis demonstrates that users typically enter chatbot interactions with computer-like cognitive schemas, leading to different expectations compared to interactions with humans. For instance, users might expect chatbots to respond quickly, objectively, and without emotion. Social cues that clearly conflict with these expectations (e.g., empathetic expressions or response delays) can have adverse effects on central outcome dimensions like perceived authenticity, perceived usefulness, trust, and usage intentions. However, results provide evidence that these effects are moderated by individual and contextual factors, such as the user's inherent tendency to anthropomorphize chatbots. This thesis thus contributes to the ongoing debate on the (non-)social nature of chatbots by providing numerous theoretical insights and practical implications for anthropomorphic chatbot design. It also outlines future research directions and discusses emerging ethical challenges related to AI and chatbots, considering their societal impact and the potential rapid obsolescence of research findings.