Files
Abstract
Artificial Intelligence (AI) is reshaping advertising practices by enabling AI-powered chatbots to become persuasive communicators that directly interact with human users through two-way conversations. Despite their growing capabilities, a key challenge is the ontological and psychological gap between chatbots (‘machines’) and users (‘humans’). This dissertation addresses how to close the human-chatbot gap by examining how the match and mismatch between chatbot-displayed social cues presented at different points in chatbot conversations influence persuasion outcomes. Based on Expectancy Violations Theory (EVT), the present study investigates how different combinations of emotional expression cues (i.e., lesser, greater) and shared identity cues (i.e., higher-order, lower-order) influence persuasion outcomes through negative violation and perceived closeness toward the chatbot. This was examined in the context of interactive public service announcements (PSAs) that promote buying-less behavior. Furthermore, this study examines under what conditions people become more sensitive to chatbots’ negative violations based on the message-to-recipient matching principles (Study 2).Using real-time interactive chatbots, this research conducted two online experimental studies that adopted a 2 (emotional expression level: lesser vs. greater) x 2 (shared identity level: higher-order vs. lower-order) between-subjects design. Study 1 used geography-based identities and found that when a lower-order identity was shared with a chatbot, the chatbot’s greater (vs. lesser) emotional expression led to lower negative violation, which enhanced perceived closeness and persuasion outcomes (i.e., message attitude, continuance intention). Contrary to expectations, these effects did not emerge when a higher-order identity was shared. Study 2 extended this investigation by shifting the shared identity context to family-derived identities and introducing the user characteristic, family involvement level. For people with higher family involvement, when a lower-order identity was shared, the chatbot’s greater (vs. lesser) emotional expression led to lower negative violation, which enhanced perceived closeness and the persuasion outcomes. However, contrary to prediction, for people with lower family involvement, such significant effects did not emerge, regardless of the shared identity level. This dissertation highlights the importance of the alignment between sequentially presented social cues and theoretically contributes to the current human-machine communication literature. Practical implications and direction for future research are discussed.