The increasing use of AI-powered chatbots in service industries has altered customer experiences. While chatbots offer efficiency, they are not perfect replacements for human agents and are often blamed for service failures. Service failures evoke negative emotions in consumers, potentially leading to chatbot aversion and reduced satisfaction. Existing research has focused on the positive aspects of chatbot communication in successful service encounters, but less is known about consumer reactions to chatbot service failures. This study addresses this gap by exploring how chatbot communication styles impact consumer responses during service failures. Specifically, it examines whether task-oriented or social-oriented chatbot communication styles influence consumer satisfaction, trust, and behavioral intentions. The researchers propose that social-oriented communication, fostering warmth, will improve consumer outcomes, and that expectancy violations (the discrepancy between expectations and actual chatbot performance) will moderate this relationship. The study's importance lies in providing insights into how to mitigate negative consumer experiences and improve customer relationships following chatbot-related service failures.

The literature review examines prior research on human-computer interaction, chatbot communication styles, and expectancy violations. Studies show that human-like qualities in chatbots can enhance positive experiences, but service failures often lead to harsher evaluations of chatbots and even the company. The review highlights the importance of warmth and competence perception as dimensions of mind perception, impacting user attitudes. Expectancy violation theory suggests that unmet expectations lead to negative reactions. Several studies are reviewed that support the hypothesis that social-oriented chatbot communication styles can enhance warmth perception and mitigate the negative impact of service failures. A table (Table 1) summarizes key findings from previous studies on the impact of AI and chatbots on consumer experiences in various sectors.

This study employed a between-subjects experimental design. Stimuli consisted of screenshots of chatbot conversations, manipulating communication style (task-oriented vs. social-oriented). A pre-test validated the stimuli, ensuring that the designed conversations effectively represented task-oriented and social-oriented communication styles. 141 participants were randomly assigned to one of the two conditions (task-oriented or social-oriented chatbot interaction). Participants first rated their interaction efficacy expectations for the chatbot and then read a scenario describing an online shopping failure. Following the scenario, they interacted with the assigned chatbot (via the screenshot). Finally, they completed a questionnaire measuring interaction satisfaction, trust, patronage intention, perceived warmth, perceived competence, and expectancy violations (the difference between pre- and post-interaction expectations). The questionnaire items were adapted from relevant scales in prior research (e.g., Bhattacherjee, 2002; Cadotte et al., 1987; etc.) The researchers used five-point Likert scales for responses, and Cronbach's alpha was used to assess the reliability of the measurement scales. The experimental process is shown in Figure 2, and Table 2 displays the design of the chatbot communication styles.

The results supported the hypotheses. First, the social-oriented chatbot significantly enhanced interaction satisfaction, trust, and patronage intention compared to the task-oriented chatbot (H1). Second, perceived warmth mediated the relationship between communication style and all three dependent variables (interaction satisfaction, trust, and patronage intention; H2a). Competence perception, however, did not show a significant mediating effect (H2b). Third, expectancy violation moderated the indirect effect of communication style on the dependent variables through perceived warmth (H3a). When expectancy violations were high, the positive indirect effect of social-oriented communication through warmth perception was stronger. The moderation effect of expectancy violation on the relationship through competence perception was not significant (H3b). These findings are supported by statistical analyses including one-way ANOVA, mediation analysis using SPSS PROCESS model 4, and moderated mediation analysis using PROCESS model 7 with 5000 bootstrapped samples. Table 4 shows the total and indirect effects of communication style on the outcome variables. Table 5 presents results of the moderated mediation analyses. Figure 3 illustrates the effect of communication style on interaction satisfaction, trust, and patronage intention.

The findings highlight the importance of social-oriented communication in chatbot design, especially when dealing with service failures. The mediating role of warmth suggests that creating a friendly and supportive chatbot interaction can significantly improve consumer responses. The lack of a mediating role for competence suggests that consumers may prioritize emotional support over technical efficiency during service failures. The moderating role of expectancy violations underlines the need for adaptive communication strategies. Chatbots should adopt a more social approach when dealing with customers whose expectations have been highly violated. The dominance of warmth perception over competence in this context is discussed and compared to existing literature. The study's findings are discussed in relation to the machine heuristic concept and consumer emotional responses to service failures. The researchers acknowledge some limitations of relying solely on chatbot screenshots in the experiment.

This study demonstrates the significant impact of chatbot communication style on consumer experiences during service failures. Social-oriented communication, fostering perceived warmth, is key to improving satisfaction, trust, and patronage intention. Expectancy violation moderates this effect, with social orientation being particularly important when expectations are unmet. The study offers valuable implications for chatbot design and customer service strategies. Future research should explore real-time chatbot interactions, investigate the effect of varying degrees of social orientation, examine chatbot-initiated interactions, and consider the role of demographic variables in shaping consumer responses.

The study's limitations include the use of static screenshots instead of real-time chatbot interactions, the dichotomous nature of the communication style manipulation, the focus on consumer-initiated interactions, the limited consideration of demographic factors, and the relatively small sample size. These limitations affect the generalizability of findings and suggest avenues for future research. Future work could incorporate real-time chatbot interaction, more nuanced manipulation of communication style, and investigation of chatbot-initiated interactions and different demographic groups.