1. Introduction
E-commerce has utilized virtual streamers for live streaming to optimize cost-efficiency and provide round-the-clock services to consumers, with unprecedented growth expected in the near future [
1]. According to iiMedia Research [
2], nearly 90% of the respondents expressed their intention to make purchases through virtual streamers, and 47.5% of consumers were highly optimistic about the prospects of virtual live-streaming commerce in China. Virtual streamers, also referred to as digital or AI streamers, are computer-generated and AI-powered characters that possess a human-like appearance [
3]. They are designed to mimic human-like features, possessing the social interaction abilities to facilitate transactions and enhance the overall experience in live-streaming commerce [
4]. Compared with human streamers, virtual streamers are recognized for their consistency, scalability, and cost-effectiveness [
5].
However, despite the technical advances, the practical application of virtual streamers has yet to achieve satisfactory performance and frequently fails to meet consumer expectations [
5,
6]. For instance, virtual streamers may provide unsuitable responses to consumer requests, lack genuine human emotion, and offer limited information to consumers [
6]. These issues may significantly undermine consumers’ shopping experiences, resulting in negative outcomes, such as reluctance to use virtual streamer services or negative word of mouth [
7]. The study of Gao et al. [
7] suggests that AI streamers lack perceived intimacy and responsiveness compared with human streamers, thus diminishing consumers’ purchase intention. Therefore, it is crucial to understand and avoid the potential negative consumer behavior that may arise during human–virtual streamer interactions. Previous studies on virtual streamers have revealed the advantages of virtual streamers with anthropomorphic traits [
8,
9], warmth factors [
1], and sociability [
4] in bolstering a consumer’s trust, engagement, experiential value, and purchase intention. Although these studies have mostly reported the factors affecting the effectiveness of virtual streamers, few have paid attention to the unpleasant experiences, consequences, and mechanisms that arise when consumers encounter poor performance from virtual streamers.
To bridge this gap, we draw on expectation violation theory (EVT) and computers as social actors (CASA) theory to explore the negative outcomes of consumers’ interactions with virtual streamers, an aspect that has been less discussed in prior research. According to CASA theory, people tend to treat human-like virtual streamers as social actors and behave socially in interactions with them [
10,
11]. When entering a live broadcast room, the consumer first notices the appearance of a virtual streamer before engaging in formal interaction. The initial impression of the virtual streamers directly influences consumers’ subsequent expectations [
12]. Consumers may perceive anthropomorphic virtual streamers as having the same efficiency and competencies as human streamers [
13,
14]. Miyan Liao et al. [
15] pointed out that the competencies of human streamers encompass cognitive, emotional, and social dimensions, which play a crucial role in attracting a substantial viewership and driving live-streaming sales. Therefore, based on CASA theory and the work of Miyan Liao et al. [
15], we propose that consumers in the initial stage of interacting with virtual streamers may expect them to possess the same cognitive, emotional, and social competence as human streamers.
Nevertheless, if the actual performance of virtual streamers fails to meet consumer expectations regarding their competencies, negative outcomes may occur as the interaction deepens [
5]. Previous research has demonstrated the positive effect of streamers’ competencies on consumers’ experiences [
1,
7]. However, the consequences of a lack of these competencies (or being lower than expected) on consumers’ negative behavior in the virtual live-streaming context have remained underexplored. Therefore, we draw on EVT to explain the negative outcomes when consumers’ expectations are violated while interacting with virtual streamers. Specifically, based on the “expectation violations–psychological state–behavior outcome” framework, we examined the mediation effect of distrust and dissatisfaction on the relationships between competency expectation violations and consumers’ discontinuance behavior. Two research questions are addressed in this study: (1) How do expectation violations affect consumers’ distrust and dissatisfaction while interacting with virtual streamers? (2) How do distrust and dissatisfaction with virtual streamers influence consumers’ negative behavior?
Our study contributes to the existing literature in several ways. First, existing research has extensively studied human streamers in live-streaming commerce, while the attention given to virtual live streamers remains limited. By drawing on EVT, this study focuses on virtual streamers in the context of live-streaming commerce and examines the underlying mechanism for negative user behaviors resulting from expectation violations. Thus, this study not only enriches the live-streaming commerce research involving virtual streamers, but also extends the EVT in novel fields. Second, this study contributes to human–AI interaction by exploring the negative consequences of expectation violation regarding human-like competencies. Specifically, we conceptualize and decompose the expectation violations into three dimensions: cognitive (professionalism expectation violation), emotional (empathy expectation violation), and social (responsiveness expectation violation). We then uncover how these violations influence consumers’ feelings of distrust and dissatisfaction, leading to discontinuance behavior. Third, this study advances the literature by revealing that the distrust and dissatisfaction with virtual streamers mediate the effect of expectation violations on discontinuance behavior. From a practical perspective, this study encourages live-streaming practitioners and information technology managers to appropriately adjust the implementation strategies of virtual streamers according to the three dimensions of expectation violations.
The next section presents a literature review on virtual streamers in live-streaming commerce and expectancy violation theory.
Section 3 presents our research model and discusses the corresponding hypotheses.
Section 4 describes the methodology and research design. The analysis results are presented in
Section 5. In
Section 6, we discuss the implications for both research and practice and the limitations of this study.
3. Research Model and Hypotheses
Based on the above literature review, we developed our research model, as presented in
Figure 1. Drawing upon expectation violation theory and the “expectation violations–psychological states–behavior outcome” framework, we attempted to explain the mechanism of the negative outcomes of expectation violations in the virtual live-streaming commerce context. CASA theory posits that consumers will interact with virtual streamers with certain expectations due to their anthropomorphism [
10]. However, when consumers’ perceptions fail to align with their expectations, i.e., negative expectation violations occur, their affective states will be negatively impacted, resulting in distrust and dissatisfaction with virtual streamers [
11]. Subsequently, these negative affective states exert an influence on consumers’ behavior.
In addition, prior research has suggested that demographic factors, such as gender, age, education, and income, may affect consumers’ adoption and post-adoption behaviors [
9,
19,
38]. For instance, Huang and Yu [
39] revealed differences between men and women in terms of their continuance intention to watch AI news anchors. Yuguang Xie et al. [
40] showed that younger users were more likely than older users to adopt AI assistants. Therefore, we included demographic factors as control variables to examine their potential influence on consumers’ discontinuance behavior.
3.1. Distrust, Dissatisfaction, and Discontinuance Behavior
In the context of live-streaming commerce, discontinuance behavior can be defined as individuals reducing their level of engagement with virtual streamers or temporarily or permanently ceasing to watch them [
41,
42].
Distrusting belief is defined as “the degree to which one believes, with feelings of relative certainty, that the other person or entity does not have characteristics beneficial to one” [
43] (p. 44). Previous research has argued that individual distrust belief can lead to distrust-related behaviors, such as a decrease in the intensity of information system usage [
44], resistance to AI chatbots [
32], and discontinuance intention of WhatsApp [
45]. In the context of this study, when consumers develop mistrust toward virtual streamers, it can result in diminished confidence in virtual streamers’ capabilities, leading to decreased engagement and a lower likelihood of continued watching. Accordingly, we propose that:
H1. Distrust in virtual streamers is positively related to discontinuance behavior.
Dissatisfaction reflects negative feelings, such as irritation or frustration, that individuals may have toward their overall experience with a product or service [
41,
46]. Dissatisfaction with virtual streamers, in particular, indicates that consumers feel discontented after engaging with virtual streamers. Previous research has found that such negative feelings would lead to unfavorable consumer behaviors, such as discontinuing usage, restricting use, or switching to alternatives [
41,
47]. For example, Zhang et al. [
41] found a positive relationship between dissatisfaction and discontinuance intention. Therefore, consumers who experience negative feelings toward virtual streamers are less likely to invest time and effort in engaging with them in the future. Instead, they may seek real human streamers as a substitute for virtual ones. Hence, we hypothesize the following:
H2. Dissatisfaction with virtual streamers is positively related to discontinuance behavior.
3.2. Negative Expectation Violations and Distrust in Virtual Streamers
According to Darke et al. [
48], when negative expectation violation occurs, it leads to customer distrust. Consumers evaluate their experience of watching virtual streamers and ultimately assess the knowledge, effectiveness, and intelligence of these virtual streamers. If virtual streamers fail to meet consumers’ expectations, this may result in an amplifying or boomerang effect, ultimately leading to a level of distrust in them [
49]. The impacts of PEV, EEV, and REV on distrust are formulated as follows.
The level of professionalism of streamers reflects their capability to effectively communicate appropriate knowledge, experience, and skills to consumers [
50,
51]. Empirical research on real human streamers has shown that professional streamers engender greater trust among viewers [
31,
52]. Individuals often perceive AI-based entities as having advanced cognitive capabilities, thus setting high expectations regarding the quality and accuracy of live content from virtual streamers [
6,
53,
54]. When virtual streamers perform poorly in demonstrating their professionalism, such as lacking essential product knowledge, frequently making factual errors, or providing inaccurate information, it can lead consumers to doubt their ability and reduce trust in them [
32]. Therefore, we hypothesize the following:
H3a: Professionalism expectation violation (PEV) is positively related to distrust in virtual streamers.
Empathy refers to streamers’ capacity to sense and react to an individual’s thoughts, feelings, and experiences [
32]. This capacity is crucial for streamers as it enables them to demonstrate personalized care and attention toward their viewers [
55]. Empathy has been shown to play a central role in the trust-building process [
32,
56], whereas its absence can lead to distrust. For example, Yang et al. [
32] found that the absence of empathy can increase distrust in AI-based customer service. Consumers generally expect a certain level of emotional intelligence and understanding from virtual streamers [
11,
57]. If these expectations are not met, particularly in moments where consumers seek emotional support or validation, it can lead consumers to feel detached and unimportant, thereby increasing their distrust in virtual streamers. Hence, we hypothesize the following:
H4a: Empathy expectation violation (EEV) is positively related to distrust in virtual streamers.
Responsiveness reflects streamers’ ability to assist consumers and deliver timely service [
58,
59]. Previous studies have shown that the immediacy of responses from real human streamers helps to form a relational bond between consumers and streamers, thus enhancing trust in streamers [
33,
59]. On the other hand, slow responsiveness and inefficient communication may hinder consumers’ trust [
60]. In this study, individuals who engage with virtual streamers expect to receive fast and effective replies from them [
1]. When consumers encounter delayed responses or feel ignored by virtual streamers, they may perceive them as less capable and efficient in problem resolution. Therefore, failure to address consumer issues in a timely manner may damage consumer trust in virtual streamers and subsequently lead to distrust in them. Hence, we propose the following:
H5a: Responsiveness expectation violation (REV) is positively related to distrust in virtual streamers.
3.3. Negative Expectation Violations and Dissatisfaction with Virtual Streamers
According to the expectancy disconfirmation theory, when the actual performance of a product or service fails to meet consumers’ expectations, it will lead to consumer dissatisfaction [
61]. Previous studies have extensively demonstrated the impact of negative expectancy violations on consumer dissatisfaction [
62,
63,
64]. The effects of PEV, EEV, and REV on dissatisfaction are formulated as follows.
Previous studies have indicated that the information provided by streamers with high professionalism greatly reduces consumers’ costs in product search and evaluation, and notably enhances their satisfaction during the purchasing process [
15,
30]. Drawing upon the expectation disconfirmation theory, if consumers’ expectations regarding the credibility, accuracy, or professionalism of virtual streamers are not met—for instance, when a virtual streamer provides inaccurate information or engages in misleading behaviors—consumers might experience dissatisfaction with the virtual streamer [
64]. Corroborating this notion, Peng et al. [
6] discerned that when the information quality of AI-oriented live streaming goes against consumers’ expectations, it could result in consumer dissatisfaction. Therefore, we propose the following hypothesis:
H3b: Professionalism expectation violation (PEV) is positively related to dissatisfaction with virtual streamers.
Existing literature on live-streaming commerce has demonstrated that the empathy shown by real human streamers during interactions and their concern for consumers’ needs and interests make consumers feel more comfortable and satisfied [
3,
59]. Furthermore, research on human–AI interactions has indicated that empathy improves satisfaction in service robots and AI devices [
32,
65]. Therefore, similar to empathetic human streamers who can increase the satisfaction and relationship quality between consumers and streamers, the presence of empathy in virtual streamers can increase consumers’ trust and satisfaction toward them. According to CASA theory [
65], consumers expect virtual streamers to be empathetic (similar to human streamers). If virtual streamers fail to sense and react to a consumer’s thoughts, feelings, and experiences, it often leads to a sense of frustration and disappointment in the consumer [
6], thereby fostering dissatisfaction. Thus, we hypothesize the following:
H4b: Empathy expectation violation (EEV) is positively related to dissatisfaction with virtual streamers.
The relationship between a lack of responsiveness and dissatisfaction has been examined in relationship marketing and the service industry [
66]. Specifically, Mostafa et al. [
64] concluded that the negative parasocial interaction disconfirmation of virtual agents, such as failing to respond suitably to consumers’ requests, is positively associated with dissatisfaction with virtual agents. Responsiveness emphasizes the streamers’ capacity to provide instant responses [
1]. This quick response service can diminish the psychological distance between consumers and streamers, enhancing their connection and promoting consumer comfort and satisfaction with their interaction experience [
58]. In contrast, the absence of responsiveness can engender a sense of neglect or disregard among consumers, resulting in a perception of an inefficient and unfriendly shopping process, ultimately leading to dissatisfaction [
6]. Therefore, we propose the following hypothesis:
H5b: Responsiveness expectation violation (REV) is positively related to dissatisfaction with virtual streamers.
6. Discussion and Implications
6.1. Discussion of the Results
Despite the majority of research on live-streaming commerce focusing on human streamers, there is limited knowledge about virtual streamers. This study sought to explore the adverse consequences arising from negative expectation violations when consumers interact with virtual streamers. In particular, this research investigated how PEV, EEV, and REV impact consumers’ discontinuance behavior, mediated by distrust and dissatisfaction. Through this study, we argue that live-streaming practitioners and information technology managers should take proactive steps to meet or exceed consumer expectations of virtual streamer competencies in order to mitigate potential negative outcomes.
First, both distrust (β = 0.262,
p < 0.001) and dissatisfaction (β = 0.507,
p < 0.001) were found to have a positive association with discontinuance behavior, thus supporting H1 and H2. This supports the previous research conducted by Yang et al. [
32], which showed that customer distrust in chatbots reduces their resistance to them. Additionally, this also aligns with prior research on AI-based entities, such as chatbots and virtual agents, which concluded that satisfaction is a vital determinant of intention to reuse [
60,
64]. In addition, this study revealed that dissatisfaction had a stronger impact on discontinuance behavior compared to distrust. Our study has extended the understanding of the relationship between distrust/dissatisfaction and discontinuance behavior in the novel context of human–virtual streamer interactions.
Second, the empirical results verified that the violations of professionalism (β = 0.328,
p < 0.001), empathy (β = 0.312,
p < 0.001), and responsiveness (β = 0.169,
p < 0.05) were positively associated with distrust, thus supporting H3a, H4a, and H5a. These findings are consistent with those of Yang et al. [
32], which showed that the drawbacks of AI-based chatbots, such as a lack of empathy and the provision of irrelevant or biased information, can lead to user distrust. The study results suggested that if consumers perceive virtual streamers as lacking the expected communication abilities, such as understanding, empathy, and responsiveness, that consumers anticipate, it may lead to an increase in distrust in virtual streamers.
Third, the findings demonstrated that the violations of professionalism (β = 0.354,
p < 0.001), empathy (β = 0.282,
p < 0.001), and responsiveness (β = 0.252,
p < 0.05) were positively associated with dissatisfaction, thus supporting H3b, H4b, and H5b. The findings revealed that when virtual streamers fail to meet consumers’ expectations of human-like competencies, it can result in dissatisfaction with the virtual streamers. The findings align with previous research, which implied that customers’ interaction with AI-enabled non-human entities that go against their expectations may lead to dissatisfaction [
6,
64]. Our findings complement previous research (e.g., [
29,
53]), which emphasized the importance of human-like competencies for AI agents in influencing consumer satisfaction. Furthermore, the results align with expectation disconfirmation theory [
61], demonstrating that consumers interact with virtual streamers with pre-determined expectations regarding their human-like competencies. If these expectations are not met, they tend to be dissatisfied with virtual streamers.
Finally, our study found that distrust and dissatisfaction mediated the relationship between the three aspects of expectation violations and discontinuance behavior, further validating the central role of affective factors in translating expectation violations into negative behaviors. Specifically, when consumers evaluate their experience of interacting with virtual streamers that fail to meet their expectations, their affective attitudes toward the virtual streamers will be negative, which in turn leads to discontinuance behavior. This result corresponds with previous studies [
37,
41,
63], which highlighted the mediating effect of psychological states in the cognition–behavior nexus. This finding helps us understand the series of changes in individual psychological reactions when expectation violations occur in human–virtual streamers interactions. It underscores the importance of addressing consumer dissatisfaction and distrust as a key strategy for retaining consumer engagement and preventing discontinuance behavior in the virtual streaming context.
6.2. Theoretical Implications
Our study made several contributions to the literature. First, our findings extended the live-streaming commerce literature by delving into the potential negative effects when virtual streamers’ performance violates consumers’ expectations. Research on live-streaming commerce has largely focused on consumers’ positive perceptions of virtual streamers, such as increasing perceptions of social presence and telepresence [
1], impacting individuals’ sense of engagement [
9], and enhancing consumer brand forgiveness [
24]. Drawing from CASA theory, we acknowledged that consumers engage with virtual streamers with certain expectations due to their anthropomorphism [
10]. Contrary to prior studies that primarily emphasized the positive aspects of virtual streamers, our study proposed that discrepancies between consumers’ expectations and the actual performance of virtual streamers may lead to negative consequences, and we further explored the underlying mechanism behind it, thereby broadening the research boundaries of live-streaming commerce. Furthermore, our study contributed to the understanding of discontinuation behavior and the factors that induce such behavior in AI-enabled live-streaming commerce contexts.
Second, this study provided valuable insights into the literature on human–AI interactions by exploring the mechanisms of expectation violation regarding human-like competencies within the domain of virtual live-streaming commerce. Drawing upon the multidimensional nature of streamers’ competencies, as proposed by Miyan Liao et al. [
15], this study offered insightful perspectives on the effects of expectation violation of human-like competencies (cognitive, emotional, and social) in virtual streamers. Our findings revealed that when virtual streamers are perceived as lacking human-like competencies, such as providing inadequate responses, lacking authentic emotion, or offering limited information to consumers, this can lead to consumer distrust and dissatisfaction toward the virtual streamers. Consequently, we underscored the importance of fostering consumers’ trust and satisfaction toward virtual streamers by meeting expectations about three human-like competencies in the context of human–virtual streamer interaction.
Third, this study applied the EVT to the field of live-streaming commerce and proposed three types of competencies’ expectation violations, namely, PEV, EEV, and REV. This expanded the scope of EVT, which was previously applied in conversational agents [
35], chatbots [
12], and AI speakers [
76]. Further, this study investigated how these three violations led to consumers’ distrust and dissatisfaction with virtual streamers, ultimately resulting in discontinuance behavior. These findings contribute to the literature on negative perceptions and responses in the context of AI-enabled live-streaming commerce.
6.3. Practical Implications
This paper has several practical implications for live-streaming marketing personnel, as well as for managers in live-streaming platforms and information technology (IT). First, it is necessary to clearly inform consumers that they are interacting with AI-based streamers. By doing so, their expectations for the interaction can be adjusted to be lower than those typically held for human streamers [
35]. For example, it should be explicitly stated in the live room that the streamer is AI-powered, and a concise explanation of its capabilities and limitations should be provided. Setting modest expectations for these virtual streamers allows consumers to be pleasantly surprised when their expectations are exceeded, rather than feeling frustrated when their expectations are unmet. This approach can benefit the company by reducing the instances of negative expectation violations that lead to consumer distrust and dissatisfaction when utilizing AI-powered streaming services.
Second, managers of IT and live-streaming platforms need to be keenly aware of the inherent limitations of virtual streamers in fully meeting consumers’ expectations. These limitations often stem from the technological and algorithmic constraints of AI, which may lead to unexpected behaviors or responses that violate consumers’ expectations. To mitigate these issues, IT managers should continuously optimize virtual streamers’ capabilities, adopting more advanced artificial intelligence and machine learning technologies that enable virtual streamers to better respond to consumers’ needs and preferences. They should train virtual streamers to enhance their ability to provide expert descriptions of products to target consumers, effectively address consumers’ requirements, and express empathy in communication with consumers. For example, virtual streamers should be programmed to recognize and respond appropriately to emotional cues from consumers, thereby creating a more engaging and satisfying experience. In addition, live-streaming platforms should solicit feedback on consumer interactions with AI-based streamers, identifying potential expectancy violations as they occur. For example, platforms can implement a real-time sentiment analysis system that monitors consumer comments and reactions during live streaming. When such sentiment analysis detects potential expectancy violations—such as consumers expressing confusion over a product’s features, dissatisfaction with the virtual streamer’s response time, or frustration with a lack of empathy in communication—managers can then quickly assess the situation, determine whether an expectancy violation has indeed occurred, and take appropriate corrective actions.
Finally, marketers and IT managers can build a comprehensive management mechanism to minimize the negative feelings that consumers may experience (such as distrust and dissatisfaction) and adverse consumer behaviors that can result from expectation violations. For example, they can establish a dedicated customer feedback loop that integrates AI-driven analytics with human oversight, allowing consumers to voice their concerns and feedback through multiple channels while enabling the AI systems to identify negative comments and potential expectancy violations. In addition, marketers could offer appropriate compensation or incentives to affected consumers to mitigate the negative impacts of expectancy violations and strengthen consumer loyalty.
6.4. Limitations and Future Research
This study had several limitations, which suggest directions for future research. First, the sample of the current study was limited to Chinese consumers who had previously interacted with virtual streamers. Thus, the results are limited to a single cultural background and may not be directly applicable to other cultural contexts. Future studies could expand the scope of the research by including samples from various countries and cultural contexts to improve the generalizability of our findings.
Second, our study relied on self-reported survey data collected from consumers regarding their actual feelings and experiences with virtual streamers. While this study showed that CMB was not a significant concern, we did not completely exclude the possibility of its existence. Future research could incorporate in-depth interviews, semantic analysis, or experimental methods to provide a more accurate picture of consumers’ experiences and behaviors in the context of virtual streamers’ live-streaming commerce.
Third, our study validated the complex relationship between expectation violations, distrust, dissatisfaction, and discontinuance behavior. However, the reality may be even more complex, with potential moderating factors that could influence the strength and direction of these relationships. For example, personality attributes, such as an affinity for human–computer interactions [
5] and moral licensing [
37], may moderate the impacts of expectation violations on distrust and dissatisfaction. In addition, there may be some possible interactions between product types (e.g., hedonic products and utilitarian products) and expectations regarding virtual streamer competencies [
23]. The impact of cultural differences on consumer expectations and perceptions of virtual streamers remains an open question [
77]. Therefore, future research should explore additional moderating factors, such as personality traits, product types, the timing and frequency of interactions, and cultural differences, to validate the robustness of the study or extend the theoretical model.
Finally, although our study presented a rigorous analysis of the phenomenon of virtual streamers in live-streaming commerce, the broader implications of our findings across other domains where digital human interactions are becoming increasingly prevalent have yet to be fully explored [
3]. Future research should explore the generalizability to different cultural settings and diverse industry applications, such as AI-driven customer service, virtual healthcare providers, and digital education platforms. Additionally, as AI technology continues to evolve, it is crucial to examine how advancements in natural language-processing and machine learning algorithms might further shape consumers’ expectations and experiences with virtual streamers. By doing so, researchers can uncover new dimensions of consumers’ expectation violations and contribute to the development of more effective strategies for managing consumers’ trust and satisfaction in human–AI interactions.