The hierarchy-of-effects pathways of virtual influencers in metaverse: unpacking the path to luxury brand engagement

The global metaverse market is expected to increase from USD 17.5 billion in 2023 to approximately USD 54.5 billion by 2028 (Hughes and Barbour, 2024). The metaverse is defined as a VR-based digital environment that merges online connectivity with AR elements (Dwivedi et al., 2022). Metaverse live streaming has become more and more popular, which delivers interactive events in real time within virtual settings (Daniel, 2023).

Virtual influencers are computer-generated personas developed through digital graphics software and operated by humans or artificial intelligence systems (Xie-Carson et al., 2023). Unlike human influencers, they possess distinct characteristics that set them apart (Lee et al., 2024). Brands can strategically design these virtual influencers to appeal to their target audiences (Lee et al., 2024; Yu et al., 2024). Song et al. (2024) said that virtual influencers can express emotions and share personal stories. This improves the efficiency of a virtual influencer in communication with consumers. In other words, Kim and Kim (2025) pointed out that virtual influencers can use the idea of friendship. This helps get consumers involved on social media. In addition, Consumers are attracted when the influencer aligns with their values (Belanche et al., 2024). Thus, connectivity will be a characteristic of the virtual influencers.

Luxury brands are increasingly exploring the metaverse to create unique and engaging consumer experiences (Chitrakorn, 2023). Levkov et al. (2023) suggested that virtual influencers can make a kind of one-sided social connection with their followers. For instance, self-disclosure can significantly enhance the perceived closeness and trustworthiness of influencers and produce a parasocial bond with their audience (Chung and Cho, 2017; Ma et al., 2024). However, interactions in the metaverse may not always have strong emotions. A study shows that if virtual experiences are not planned well, they could make people feel lonely (Oh et al., 2022). Then, the question arises: How can luxury brands create compelling and emotionally resonant relationships between virtual influencers and consumers in the metaverse?

While luxury brands are increasingly integrating virtual influencers into the metaverse, Chitrakorn (2023) suggested that the research about how virtual influencers can effectively enhance consumer relationships is still limited. In addition, the characteristics of virtual influencers in the metaverse are still valued for research (Kumar and Shankar, 2024). Concerning the parasocial relationships within virtual environments, the particular characteristics of virtual influencers are necessary to investigate (Liu and Wang, 2025). To address the above research gap, this study applied commitment-trust theory to investigate whether trust and commitment serve as key relational mechanisms that shape consumer engagement with virtual influencers.

In order to solve these gaps, this research investigates how friendship and connectivity influence consumer-influencer relationships in metaverse live streaming and how these relationships, in turn, impact consumer engagement with luxury brands. This research started with a review of previous research and theory to formulate hypotheses. Subsequent sections will discuss the research methodology, data analysis, implications, limitations, and suggestions for future research.

Virtual influencers have recently gotten more attention (Ju et al., 2024). Luxury brands are collaborating with virtual influencers to adapt to market changes (Dwivedi et al., 2023; Mo and Wang, 2024). Virtual influencers provide brands with creative ways to communicate with their target consumers (Audrezet and Koles, 2023). According to research, virtual influencers’ human-like attributes and ability to interact with people positively affect the formation of parasocial relationships, which in turn improve company motivation and commitment (Liu et al., 2023).

How the virtual influencer successfully attracted consumers remains insufficiently understood (Laszkiewicz and Kalinska-Kula, 2023). Research on investigating the consumer trust of virtual influencers is still limited (Muniz et al., 2024). Furthermore, it is challenging to balance the virtual nature of influencers with the heritage of luxury brands (Mo and Wang, 2024). Notably, there is evidence suggesting that virtual influencers enhance brand narratives when appropriately aligned with luxury brands’ heritages (Guzzetti et al., 2024). Therefore, the characteristics of virtual influencers have become a hot topic of research. This research will start with the characteristics of the virtual influencers and investigate their impact on consumer-influencer relationships.

In accordance with the Hierarchy-of-Effects (HOE) model, consumers go through a step-by-step process that leads to a final behavior (Rehman et al., 2014). This model has three main stages: cognition, affect, and conation (Barry and Howard, 1990; Zhu et al., 2019). For virtual influencers, the cognition stage is the first step. In this stage, consumers learn about the influencer and process information about them. This study identifies two key characteristics relevant to consumer cognition: friendship and connectivity. The second stage is the affect stage. It involves the development of emotional responses toward a virtual influencer. Thus, consumer-influencer relationships are important in this stage. The last stage is conation. It reflects the consumer’s behavioral intentions and actions. Thus, this stage includes active engagement with both the influencer and the affiliated luxury brand. Active engagement is defined as “the behavioral manifestations that represent the level of users’ investment, participation, and efforts in activities” (Xue et al., 2020), which reflects the consumer's actions. Thus, the consumer's active engagement is in this stage.

Parasocial relationships are defined as the one-sided psychological bonds that people form with media figures (Horton and Richard Wohl, 1956; Ma et al., 2024). Virtual influencers can be used to create human-like emotions and relationships because of parasocial relationships (Liu and Wang, 2025). A study has illuminated how followers' perceptions of virtual influencers, including aspects of desirability and similarity, can lead to high levels of parasocial interaction (Melnychuk et al., 2024). This study uses the parasocial relationship framework to explore how virtual influencers build relationships with consumers.

The commitment–trust theory posits that trust and commitment are fundamental to building relationships (Morgan and Hunt, 1994). This theory serves as the foundation of the consumer-influencer relationship concept of this research. The commitment–trust theory has been applied to explain relationship marketing (Morgan and Hunt, 1994) and social media marketing (Bao and Wang, 2021). The commitment–trust theory is suitable for exploring consumer behavior (Morgan and Hunt, 1994); thus, this research uses consumer commitment and trust propensity to reflect the consumer-influencer relationship.

Friendship is conceptualized as a reciprocal bond marked by closeness and affinity (Chung and Cho, 2017; Tukachinsky, 2010). These are unilateral connections where individuals perceive a sense of nearness and acquaintance with virtual influencers, even without physical interactions. Fehr (1996) emphasizes that friendship development is heavily shaped by contact regularity. Consequently, consistent exposure to an influencer cultivates profound intimacy among consumers (Chung and Cho, 2017). Similarly, this characteristic can also apply to a virtual influencer. Additionally, the work of Delbaere et al. (2021) underscores that social media influencers are instrumental in engaging consumers with brands, fostering consumer-brand relationships, and ultimately driving brand sales. Based on the previous review, the following hypotheses are presented.

In this research, connectivity refers to the relationship, based on similarities, between virtual and real-world social influencers (Malik et al., 2023). According to the research by Levkov et al. (2023), the theoretical framework of parasocial relationships can be applied to explain how virtual influencers develop connections with consumers. Virtual influencers can create a sense of one-sided social interaction with their followers. For example, through regular posts, stories, and comments on social media platforms, a virtual influencer can make its followers feel like they are part of its “life” (Levkov et al., 2023). Research has found that virtual influencers with human-like features are viewed as more trustworthy and credible than those designed as animals (Kim et al., 2023). Thus, the following hypotheses are proposed.

Active engagement refers to user behavior that demonstrates investment, participation, and effort in activities (Lee et al., 2023; Xue et al., 2020). As mentioned before, the commitment-trust theory posits that successful relationships require both trust and commitment (Morgan and Hunt, 1994); thus, this research uses consumer commitment and trust propensity to reflect consumer-influencer relationships. Past research emphasizes the importance of parasocial interactions, one-sided relationships where consumers feel connected to influencers (Chung and Cho, 2017). A study highlights that perceived similarity and human likeness in influencers enhance these connections, leading to greater consumer engagement and loyalty (Stein et al., 2024). According to research, engagement and purchase intentions are driven by the emotional connections existing between the influencer and the consumer (Koay et al., 2023). Moreover, commitment, satisfaction, and trust serve as key variables in linking consumer engagement to loyalty (Rather, 2019). Then, the hypotheses are developed.

Figure 1 presents the research model and the hypothesized paths.

A questionnaire will be developed in this research, with measurement items for the constructs adapted from previous studies. The personal luxury goods market in China was about US$77 billion in 2024 (Thomala, 2024). By 2025, China's luxury goods market is projected to expand to 816 billion yuan ($112 billion), accounting for roughly 25% of the global total (Yang, 2023). In addition, the collaboration between luxury brands and virtual humans is a significant development in the marketing landscape in China (Daswani, 2021; Gastel, 2023). Thus, the Chinese consumers are selected as the target population. The target population is the Chinese consumer who has experience watching virtual influencers. Judgment sampling was applied. The minimum sample size for this research is 146, calculated using G*Power version 3.1, with six predictors, f² = 0.15, α = 0.05, and efficacy = 0.95 (Faul et al., 2009).

As mentioned in the research design, the target population for this research is Chinese consumers who have experience following virtual influencers. To verify respondent eligibility, a screening question was designed: “Have you watched a virtual influencer participating in a metaverse live-streaming session in the past six months?” Only respondents who answered “yes” were permitted to complete the full survey. This criterion ensured that participants possessed meaningful exposure to the phenomenon under investigation. The questionnaire was distributed in China to achieve the target population through online (https://www.wjx.cn/). 384 valid respondents were reserved and analyzed in this study.

The measurement items for the variables are listed in Table 1. Studies suggest that a 7-point Likert scale is more effective than a 5-point scale for capturing participants' true evaluations, especially in usability studies (Hapuarachchi et al., 2023). Thus, this research adopts a 7-point Likert scale to measure respondents' opinions. Friendship has three items, which are adopted from Chung and Cho (2017). The items of Connective are adopted from Malik et al. (2023). Consumer Commitment has 3 items, and Trust Propensity, which has 3 items, are all adopted from Kim and Chan-Olmsted (2022). Both Active engagement with the influencer and Active engagement with the Luxury Brands have 4 items that were adopted from Lee et al. (2023).

The research collected data from 384 respondents, including 283 females (73.7%) and 101 males (26.3%). According to Baklanov (2022), 61.45% of the virtual influencers' audiences are women. Thus, the female is the main consumer of virtual influencers (El Hedhli et al., 2023). Therefore, data collection is objective. According to Table 2, about 75.3% of them are between 18 and 21 years old, 308 (80.2%) of them have used Metaverse live streaming for less than 1 year, and 76 (19.8%) of the respondents have more than 1 year of experience on Metaverse live streaming. According to Hryziuk (2022), the user of the metaverse is supposed to be 80% of users beyond 16 years old. Thus, as the majority of the participants in this study are young, the data is considered to be representative.

This study uses Partial Least Squares Structural Equation Modeling (PLS-SEM) with SmartPLS version 4 to analyze the data. PLS-SEM has strong prediction power for complex models (Hair et al., 2019). It is useful for studies that aim to predict outcomes because it focuses on the explained variance of dependent variables (Hair et al., 2019). This study aims to predict how virtual influencer characteristics affect consumer behavior in the metaverse. Therefore, PLS-SEM is suitable for this research.

Common method bias (CMB) occurs when measurements of both independent and dependent variables are obtained through a single data collection method (Kock et al., 2021). This research uses Liang's approach to test the common bias (Liang et al., 2007). According to Table 3, the ratio of average Rb² to average Ra² is 155.870, a high value, suggesting that significant common method bias is not present.

This study assesses the reflective measurement model to examine the latent constructs and their respective measures (Hair et al., 2019). The first step is the estimation of loading and significance. As Table 4 shows, all the standardized loadings are higher than 0.708, and all the associated t-statistics exceed 1.96; thus, the loadings and significance of the measurement model are acceptable. Secondly, according to Table 5, outer loading is higher than 0.708; thus, the indicator reliability is acceptable (Hair et al., 2019). Third, internal consistency reliability was assessed using Cronbach's alpha and composite reliability (rho_c), with both values required to be above 0.7 to confirm reliability (Hair et al., 2019). Fourthly, convergent validity was checked through outer loadings and Average Variance Extracted (AVE). According to Hair et al. (2019), outer loadings should be above 0.708, and AVE should exceed 0.5. Fifthly, discriminant validity was assessed using the Heterotrait-monotrait (HTMT) ratio (Table 6), based on 5,000 bootstrap samples (Tan and Ooi, 2018).

The model fit was assessed using the Standardized Root Mean Square Residual (SRMR). The SRMR values for the saturated model (0.037) and the estimated model (0.061) were both below the 0.08 cut-off, indicating a good model fit (Hair et al., 2019). Collinearity was examined using the variance inflation factor (VIF). As Table 7 shows, all VIF values were below 5. Thus, there are no multicollinearity issues in this research (James et al., 2021).

Hypothesis testing results are summarized in Table 8. First, FD significantly influenced CC (β = 0.569, p < 0.001) and TP (β = 0.503, p < 0.001), while CN also had significant positive effects on CC (β = 0.318, p < 0.001) and TP (β = 0.328, p < 0.001). Second, CC (β = 0.231, p < 0.001) and TP (β = 0.637, p < 0.001) both had significant positive impacts on AEI. Lastly, AEI significantly influenced AEL (β = 0.761, p < 0.001).

A Q² value greater than 0 suggests high predictive relevance for endogenous constructs (Hair et al., 2019). Table 9 confirms this, as all Q² values exceed 0, demonstrating the strong predictive relevance of the exogenous constructs. Regarding the structural model, R² values of 0.75, 0.50, and 0.25 represent substantial, moderate, and weak explanatory power, respectively. In this study, all R² values are above 0.5, indicating moderate explanatory power for the endogenous variables. In addition, effect size evaluation measures the influence of independent variables on dependent ones, with f² values categorized as small (0.02), medium (0.15), or large (0.35) (Cohen, 1988). Table 10 shows that FD exhibits large effects on CC (0.514) and TP (0.309), while CN shows medium effects on CC (0.164) and small effects on TP (0.131). Additionally, CC has a small effect on AEI (0.057), TP exerts a large effect on AEI (0.432), and AEI strongly influences AEL (1.374).

H1a and H1b are supported, suggesting that friendship significantly influences consumer commitment and trust propensity, which aligns with the result by Zhang et al. (2024), that in live streaming, friendship has a significant impact on customer behavior. This indicates that the relationship between influencers and customers, resembling a friendship, can positively impact consumer behavior. The result also offers support for applying the theory of parasocial relationships to virtual influencers (Stein et al., 2024). This result may challenge the previous studies, which state that while virtual influencers may capitalize on striking visual representations and perform efficiently without the unpredictable nature of human behavior, they often invoke a sense of discomfort among audiences, particularly among older consumers or segments skeptical of non-human interactions (Xin et al., 2024; Zhou et al., 2024). This finding shows how important it is to build friendships to strengthen consumer–virtual influencer relationships.

In addition, connectivity (H2a and H2b) has a significantly positive impact on consumer commitment and trust propensity. This finding aligns with the earlier research conducted by Malik et al. (2023), which indicated that virtual influencers and social influencers have things in common. These common points can create a sense of familiarity for consumers. Also, the result reflects that the consumer prefers to trust human-like virtual influencers (Kim and Wang, 2023). Thus, connectivity should be an important factor in virtual influencers. These results show that connectivity is important in building trust and commitment toward virtual influencers. This can work together with the unique advantages of virtual influencers. A virtual influencer has special benefits. For example, it can represent a brand perfectly. It also has no risk of personal scandals or human mistakes (Mrad et al., 2022). It is essential for brands to assess their target demographic's preference towards virtual influencers versus human influencers, as younger consumers, in particular, might be more inclined towards engaging with virtual influencers due to their novelty and integration with digital culture (Kholkina et al., 2024).

H3a and H3b are supported in this research; thus, consumers’ active engagement with the influencer is significantly positively impacted by the consumer-influencer relationships (consumer commitment and trust propensity). This result is aligned with previous research, which indicated that purchase intentions and engagement are driven by the influencer's and the customer's emotional connection (Koay et al., 2023). The consumer-influencer relationships in this research are like emotional connections in some way. In addition, H4 was supported in this study, which indicated that active engagement with the influencer significantly positive impact on active engagement with the brand at a high level. This result aligns with past research, which suggests that effective engagement strategies on social media can lead to favorable consumer responses toward luxury brands (Kumar et al., 2022).

This study develops the concept of virtual influencer characteristics and identifies two distinct dimensions: friendship and connectivity. It empirically tests the impact of these characteristics on consumer–influencer relationships. The results suggest that both friendship and connectivity have a positive impact on consumer commitment and trust propensity. These findings enrich the literature on metaverse live streaming by showing how virtual influencer characteristics impact consumers’ relationships.

This research further extends the parasocial relationship theory by applying it to virtual influencers within metaverse live-streaming environments. While parasocial relationship theory has traditionally been applied to human media figures (Horton and Richard Wohl, 1956), this study advances parasocial relationship theory by extending its application from traditional media contexts to immersive metaverse environments and virtual influencers. While the original theory emphasizes one-sided emotional bonds formed through repeated mediated exposure, our findings show that parasocial connections can also develop in computer-generated environments where the influencer is a virtual influencer. This offers a novel theoretical insight by showing that parasocial relationships can be effectively applied to the interaction between human and virtual influencers, thereby expanding their conceptual boundary.

Additionally, this study contributes to commitment–trust theory (Morgan and Hunt, 1994). This research confirms that consumer commitment and trust propensity influence active engagement with virtual influencers. Also, this research proves that the trust and commitment parts work in a similar way for the relationships between consumers and virtual influencers. Thus, this research further advances the commitment–trust theory by confirming that commitment-trust theory can also be applied in the relationship between consumer and virtual influencers.

The findings show that perceived friendship and connectivity are key factors in consumer commitment and trust. A study has pointed out that metaverse live streaming allows real-time interactions (Barta et al., 2023). These interactions allow consumers to engage with virtual influencers through questions and discussions. This kind of interaction builds a sense of belonging (Barta et al., 2023). Therefore, companies should prioritize designing virtual influencers with strong elements of friendship and connectivity to enhance consumer-influencer relationships. In particular, luxury brands can leverage the interactive nature of metaverse live streaming to cultivate perceived friendship through frequent and authentic interactions between virtual influencers and consumers (Gao et al., 2023). Moreover, companies can improve virtual influencers’ visual design, such as giving them human-like features. This may help strengthen the sense of connection.

More and more luxury brands are using metaverse technologies. These brands include Burberry, Ralph Lauren, and Louis Vuitton (Tugba, 2024). This study contributes practical insights for fostering strong relationships between consumers and virtual influencers. The results show that such relationships positively influence consumer engagement with influencers during metaverse live streaming, which subsequently increases engagement with the associated brands.

The findings suggest that designing virtual influencers with characteristics of friendship and connectivity is important to develop consumer-influencer relationships in metaverse environments. Notably, brands such as Louis Vuitton, Burberry, and Gucci have been pioneers in the metaverse space (Boyd, 2023). Thus, luxury brands can design virtual influencers with the characteristics of friendship and connectivity to strengthen the consumer-brand relationship.

Data for this study were collected in China because China has approximately 183.13 million users on the metaverse platform, accounting for 16.7% of the global user base (Elad, 2024). However, the findings may not be fully generalized in other cultural contexts. Moreover, cultural values strongly affect consumer behavior in online contexts (Brand et al., 2022). Future studies should explore how cultural differences shape consumer engagement in different national and cultural settings. Artificial intelligence continues to advance, and virtual influencers are expected to undergo significant evolution in their capabilities, roles, and societal impact, which deserve further attention (Aw and Agnihotri, 2024)