How and When Do Virtual Influencers Work? A Meta-Analysis of Mechanisms and Moderators in Digital Commerce

Abstract

In recent years, virtual influencers (VIs) have been increasingly used in digital commerce. Despite the rise in VI research, past studies have yet to comprehensively examine the effectiveness of VIs, often focusing only on isolated partial models rather than an integrated framework and boundary conditions that drive consumer responses. This meta-analysis fills this gap by synthesizing 186 effect sizes from 76 studies (N = 64,545) to examine the mechanisms and moderators of purchase intention in VI marketing. The results indicate that human-likeness is a central antecedent that directly and indirectly affects purchase intention through source credibility, customer engagement, and attitude. More importantly, this study challenges prior social proof assumptions by showing that follower size has no significant impact on purchase intention in VI marketing. In addition, purchase intention is independent of a nation’s AI readiness, suggesting a borderless potential for commerce regardless of a country’s technological maturity. This study also examined the moderating effects of product type, consumer age, and uncertainty avoidance culture. Although these moderators showed initial significance, none remained significant after the Benjamini–Hochberg false discovery rate (FDR) correction. Therefore, these effects were viewed as exploratory rather than confirmatory, providing directions for future research. These findings offer new insights for e-commerce managers: success in the metaverse era depends on anthropomorphism and targeted alignment rather than metrics such as follower counts or a nation’s AI readiness.

1. Introduction

The evolution of artificial intelligence (AI) has made virtual influencers (VIs) or computer-generated influencers popular on social media platforms. Similarly to celebrities and human influencers, VIs are typically used in marketing to promote products, brands, and audiences through storytelling, lifestyle, and interactive content. These digital characters can connect with audiences, work with brands, and have choreographed social media identities [1]. In addition, they are easier to control, less risky, and cheaper than traditional influencers [2]. As a result of their effects, VIs have gained attention in recent research, which has focused on identifying key drivers such as anthropomorphism [3], source credibility, and customer engagement [4,5,6]. Despite the increasing prevalence of research, two gaps remain in the literature. First, various studies have been conducted to explain the mechanisms underlying the impact of VIs on purchase intention. For instance, it has been shown that customer engagement and purchase intention are enhanced by human-likeness, source credibility, and attractiveness [7]. Furthermore, Wang et al. [8] have elucidated the relationship between attitude and purchase intention. Although previous studies support various partial models, an integrated framework is required to fully capture the impact of VIs on consumer behavior. Second, prior research has endeavored to investigate moderators such as post types [9] and loneliness [6] to resolve the varying impacts of VIs. However, several moderators, such as VIs and consumer characteristics, cultural factors, and AI readiness, have not received much attention. Consumer responses have been shown to be significantly influenced by these moderators. For example, regarding the influencer, Dong et al. [10] demonstrated that mega influencers, such as celebrities, can enhance brand-related outcomes, stimulate emotional engagement, and lead to purchase intentions. Weismueller et al. [11] also indicated that the number of followers influences source attractiveness and trustworthiness, thereby increasing purchase intention. On the consumer side, consumer age plays an important role in purchase intention owing to differences in cognitive processing, emotions, and interests. For instance, younger adults are more affected by cognitive social presence, whereas older age groups are more affected by affective social presence in livestream shopping [12]. Contextual factors such as culture, technological readiness, and product type also shape these outcomes. Rizzo et al. [13] posited that collectivist cultures with low levels of uncertainty avoidance have positive attitudes towards influencers, while a positive frame directs e-consumers’ attention to a functional attribute for functional search products [14]. Furthermore, technology readiness has a positive effect on online shoppers’ attitudes and increases their purchase intentions [15,16].

Building on the gaps in the existing literature, this study investigates the mechanism and a set of moderators, including VI and consumer characteristics, cultural factors, AI readiness, and product type. The following research questions are addressed:

RQ1. 

Through which mechanism does human-likeness influence purchase intention in virtual influencer marketing?

RQ2. 

How do follower size and product types (information-based and motivation-based) moderate the relationship between key antecedents and purchase intention?

RQ3. 

How do consumer age and cultural factors moderate these effects?

RQ4. 

Does country-level government AI readiness moderate these effects in virtual influencer marketing?

To achieve these aims, a meta-analysis was conducted to understand the mechanisms and moderating effects of the factors affecting consumers’ purchase intention. In particular, this study collected data on purchase intention in virtual influencer marketing (e.g., research on human-likeness, source credibility, customer engagement, and attitudes). Next, this study evaluated three different models and an integrated model that combined various mechanisms and theories to explain how human-likeness influences customer response. Building on Stimulus-Organism-Response (S-O-R) theory, this study found that human-likeness directly and indirectly affects purchase intention through source credibility, customer engagement, and attitude. In addition, this study identified the effects of moderators (VI characteristics, product type, consumer characteristics, cultural factors, and government AI readiness) on purchase intention. By achieving these aims, this study contributes to the theory and provides practical implementation for e-commerce management.

Table 1. Prior literature reviews of virtual influencers.

Article	Year	Method	Search Period	Total Studies	Study Purpose	Main Results
[17]	2026	Systematic literature review	~June 2025	117 studies	To propose a conceptual framework.	Proposed conceptual framework and identified research gaps and suggested future research.
[18]	2026	Systematic literature review	2020–2025	75 studies	To synthesize AI/generative AI applications in social media and consumer behavior.	AI enables personalization, engagement, and analytics but raises concerns about ethical risks.
[19]	2025	Systematic literature review	2020–2025	51 studies	To explore VIs’ impact on trust, ethical considerations, and engagement in different cultures and regulatory environments.	Proposed conceptual framework which showed that trust is considered a mediator, and highlighted the importance of cultural differences.
[20]	2025	Systematic literature review	2016–2024	88 studies	To synthesize the literature, resolve contradictions, and propose future research directions.	Identified three core dimensions (affordances, actualization, outcomes) and proposed 13 strategic research propositions.
[21]	2025	Meta-analysis	~July 2024	71 studies	To assess the effectiveness, mechanisms, and moderation of human social media influencers compared to other endorsements (e.g., celebrities, virtual influencers).	SMIs outperform alternative endorsers in terms of engagement and purchase intention
[22]	2024	Systematic literature review	2012–2023	106 studies	To synthesize literature on the impact of VIs on consumer decisions using the TCM framework and ADO	Examined the determinants of consumer decisions and provided key gaps and insights for future research.
[23]	2023	Systematic literature review	2012–2023	35 studies	To synthesize and evaluate the literature on VIs in marketing.	Revealed the multidimensionality of the topic and proposes critical concerns for future studies.
This study	2026	Meta-analysis	2018–2025	76 studies	To synthesize empirical literature and test an integrative S-O-R structural model (VI-focused framework) explaining how human-likeness drives consumer purchase intention. To identify moderators (VIs and consumer characteristics, cultural factors, and AI readiness).	Human-likeness significantly drives purchase intention through source credibility, customer engagement and attitude. Furthermore, FDR adjustments validated the specific moderating effects (e.g., follower size and product type) on VI effectiveness.

presents a comparison of the methods, search periods, study purposes, and main results of the present review with those of earlier reviews, highlighting the novelty of the present study.

2. Literature Review

In this study, the variables were categorized as human-likeness, source credibility, customer engagement, attitude, and purchase intention. These are the most frequently encountered variables in prior research. The percentages of studies in the dataset that examined each response variable are shown in

Table 2. Empirical representation of variables in previous studies.

Variables	Numbers of Studies	Percentages of Studies
Human-likeness	42	55.3% *
Source credibility	47	61.8%
Customer engagement	27	35.5%
Attitude	21	27.6%
Purchase intention	48	63.2%

* The proportion of studies in the dataset that evaluated each response variable is represented as a percentage. The total percentage is greater than 100 because a study may contain multiple response variables.

. Although parasocial interaction and other variables that influence purchase intention have been investigated previously (e.g., Shah et al. [24]), these variables were excluded from this meta-analysis because of an inadequate number of studies reporting the required pairwise correlations.

Based on these variables, this study adopts the Computer as Social Actors (CASA) paradigm [25] and Stimulus-Organism-Response (S-O-R) theory [26] as the primary lenses. According to the CASA paradigm, consumers apply social rules to computers and robots and treat them as human entities. Recent evidence suggests that consumers socially engage with VIs and have emotional responses to them as if they were real people [27]. While CASA explains why VIs can influence consumers, it does not explain the psychological processes. Therefore, the S-O-R framework was adopted to explain the process by which human-likeness translates into behavioral outcomes. Guided by the S-O-R framework, human-likeness functions as the primary stimulus, which influences the internal cognitive and affective states of the consumer. Specifically, source credibility refers to a cognitive state that reflects perceptions of credibility. Customer engagement reflects affective states, which are defined by cognitive, emotional, and behavioral engagement, whereas consumer attitudes capture affective evaluation. Finally, purchase intention functions as a behavioral response triggered after consumers have positive attitudes.

To provide a robust theoretical foundation for our framework, we integrated four theoretical perspectives within the Stimulus-Organism-Response (S-O-R) model. Initially, Anthropomorphism Theory explains the stimulus (S) by detailing the consumer’s inclination to ascribe human traits to VIs, thereby converting human-like attributes into social agents. Second, the Stereotype Content Model (SCM) structures organism (O) evaluation by mapping the dimensions of competence and warmth to source credibility and customer engagement, respectively. Furthermore, Source Credibility Theory provides additional insights into how perceived source credibility fosters customer engagement and subsequently shapes attitudes. Finally, the Theory of Planned Behavior (TPB) posits that positive attitudes directly predict purchase intentions. This integration creates a consistent logical basis for connecting human-like characteristics with the ultimate behavior of consumers, as illustrated in Figure 1.

3. Hypothesis Development

To comprehensively understand the impact of VIs on purchase intention, this meta-analysis formally develops hypotheses for both mediation pathways and moderating conditions.

3.1. Human-Likeness

Human-likeness refers to attributing human characteristics to a VI or AI influencer [28]. Earlier studies have classified human-likeness into three dimensions. First, according to Pan et al. [29] and Shen [30], human-likeness is perceived through appearance and hyper-realistic visual imagery. Accordingly, this dimension illustrates the highly realistic visual designs of VIs, such as lifelike skin, natural hair, and accurate face shapes. Second, human-like characteristics have pointed towards a factor other than physicality, which is behavioral realism [28,31]. This dimension defines how an AI influencer moves and acts. For example, it includes smooth body movements and natural hand gestures. These human-like actions make the digital character feel more real and present. Third, human-likeness entails emotional expressions. Basic examples include smiling, frowning, or changing eye contact [27]. Drawing from prior research, this study defines human-likeness as having three dimensions: (1) human-like appearance, (2) human-like behavior, and (3) human-like emotions.

3.2. The Mediating Role of Cognitive and Affective Mechanisms

The mechanisms underlying the effect of human-likeness on purchase intention have been explored through several distinct pathways in the existing literature. The first mediator is source credibility. Source credibility is defined as the positive qualities of a sender that affect the receiver’s acceptance of the message [32]. The three-dimensional source credibility framework (expertise, trustworthiness, and attractiveness) has been extensively employed and demonstrated in various cultures [33,34]. Source expertise refers to the ability, proficiency, and expertise of the source. Trustworthiness refers to the extent to which a source is perceived to be honest, genuine, credible, believable, and truthful. Source attractiveness refers to physical attraction and social appeal [32,35]. Later, Munnukka et al. [36] identified similarity as a dimension of source credibility [37]. In the context of VIs, source credibility is an important factor that affects purchase intention. Some studies have highlighted that when a VI exhibits realistic physical, emotional, and behavioral characteristics, it enhances users’ cognitive and sensory experiences as relevant drivers of source credibility (e.g., attractiveness, trustworthiness, expertise, and similarity) [4,38,39]. Grounded in Source Credibility Theory, this study used dimensions such as attractiveness, expertise, trustworthiness, similarity, and familiarity to examine their effects in this framework.

When VIs showcase source credibility, consumers are more likely to trust their recommendations, which increases customer engagement. For example, Jiang, Qin, Deng and Zhou [27] suggested that customers experience empathy as an emotional state. Zhang et al. [40] also revealed that consumers tend to be more satisfied when interacting with VIs. Jayasingh, Sivakumar and Vanathaiyan [7] also demonstrated that a like or comment can be a signal that increases purchase intention. In addition, customers are willing to follow, share, and search for information about AI influencers to show their engagement with them [41,42]. Customer engagement also encompasses the flow experience, which is characterized by involvement, heightened focus, and immersion in an activity [40]. Based on a literature review, this study examines the effects of customer engagement, which includes emotional interaction (e.g., satisfaction), affective response (e.g., empathy), and willingness (e.g., willingness to follow), on purchase intention.

These cognitive and affective states among consumers lead to the development of a specific evaluation (e.g., attitude). Research findings on VIs have classified customer attitudes into three domains: attitude towards endorsers, brands, and advertising. Consumers’ attitudes towards VIs guide their decisions to purchase products endorsed by VIs [38,43]. In addition, consumers are likely to follow a VI’s advice and purchase a product when they have a good attitude toward a self-aligned character [44]. Likewise, attitude towards the brand, which arises from the congruence between VI and the brand, serves as a driver of intention [38]. In addition, attitude toward the advertisement is the consumers’ evaluation of the relevance, appeal, and execution of advertisements. Therefore, creating positive consumer attitudes is a key factor in converting perceptions into purchase intentions [45,46]. In this study, attitudes towards VIs, brands, and advertising were used to investigate their effects on purchase intention. Drawing on the S-O-R framework, we formally propose the following mediation hypothesis:

H1. 

Source credibility, customer engagement, and attitude sequentially mediate the relationship between human-likeness and purchase intention.

3.3. The Moderating Effects of VIs Effectiveness

This study focuses on moderators, including the sender (VIs size), message (hedonic vs. utilitarian and search vs. experience), receiver (age and culture), and national AI readiness (Figure 2). The use of these moderators provides an in-depth understanding of how VIs influence purchase intention.

3.3.1. The Moderating Role of Virtual Influencer’s Size

There are different types of influencers based on their follower count. According to Haenlein and Libai [47], mega-influencers are well-known individuals who are perceived to be experts or opinion leaders with over 1 million followers (e.g., famous bloggers or digital celebrities). In contrast, micro-influencers are ordinary people who have an impact on small social networks, typically ranging from 10,000 to 100,000 followers. Prior studies have also proposed another type of influencer, macro-influencers, which fill the gap between mega- and micro-influencers [48,49]. In addition, Campbell and Farrell [50] mentioned nano-influencers, defined as influencers with as few as 1000 followers [51]. Previous studies have demonstrated that smaller influencers drive more sales than larger influencers [48,52]. For example, micro-influencers are viewed as trustworthy and similar to peers; thus, their endorsements become credible with persuasive power, which eventually persuades followers’ actions [49,53]. In contrast, mega-influencers have a high status, making their endorsements more likely to be perceived as commercial. Such consumer perceptions can negatively affect authenticity and personal connections, which in turn affects purchase intentions, especially for hedonic products [48,53]. Based on previous studies, size-based differences may affect how persuasive determinants translate into purchase intentions in the VI context. Specifically, it is expected that the effect is stronger as the number of followers decreases and weaker as the number of followers increases.

H2. 

The positive effects of (a) human-likeness, (b) source credibility, (c) customer engagement, and (d) attitude on purchase intention are stronger for VIs with a small follower size than for those with a large follower size.

3.3.2. The Moderating Role of Product Characteristics

Nelson [54] classified products into experience and search. Accordingly, search products contain attributes whose values can be evaluated before purchase (e.g., electronics and clothes). In contrast, experiential products contain attributes whose value can only be evaluated after consumption (e.g., food and cosmetics). Many studies have demonstrated the significance of distinguishing between search and experience products in influencing consumers’ intentions to purchase online and through digital commerce. According to Yang et al. [55], descriptive information helps consumers make purchase decisions regarding search products. In contrast, consumers are more reliant on affective cues and metaphorical expressions to infer the quality and value of experiential products [55]. In addition, the style used in online reviews affects purchase behavior according to product type. For example, literal reviews can increase purchase intention for search products, whereas figurative reviews and prior user experience can increase purchase likelihood for experiential products [56]. According to these studies, search products are generally assessed based on their technical specifications, whereas experiential products necessitate the customer’s subjective perception and imagination. In the context of VIs, because VIs can elicit a high level of emotional resonance [40], consumers can effortlessly envision a product owing to the human-like nature of VIs. Consequently, the purchase intentions for experiential products are expected to be more positively influenced by human-likeness, customer engagement, and attitude than those for search products.

H3a. 

The positive effects of (a) human-likeness, (b) customer engagement, and (c) attitude on purchase intention are stronger for experiential products than for search products.

H3b. 

The positive effect of source credibility on purchase intention is stronger for search products than for experiential products.

In addition to information-based product types, products can be primarily divided into two types based on consumption motivation: utilitarian and hedonic products. Utilitarian products have practical characteristics and perform functional roles, whereas hedonic products possess pleasure-seeking characteristics and include sensory pleasure [57]. Research has shown that both hedonic and utilitarian products significantly affect consumers’ purchase intentions [58]. The purchase intention for hedonic products is primarily stimulated by emotional arousal, curiosity, and mental imagery. According to Chang et al. [59], social cues and prior experiences can stimulate emotions via livestreaming for hedonic products. In contrast, utilitarian products are related to cognitive evaluations of value, perceived usefulness, and risk reduction in terms of functional benefits, such as efficiency, quality, and convenience [59]. As VIs possess visual appeal and the ability to deliver high-quality entertainment content [7], there is a natural fit between the nature of VIs and consumer motivations for hedonic consumption. In other words, when promoting this product category, the human-likeness, customer engagement, and attitude may easily stimulate positive consumer emotions, which results in higher purchase intention.

H4a. 

The positive effects of (a) human-likeness, (b) customer engagement, and (c) attitude on purchase intention are stronger for hedonic products than for utilitarian products.

H4b. 

The positive effect of source credibility on purchase intention is stronger for utilitarian products than for hedonic products.

3.3.3. The Moderating Role of Consumer Age

Consumer age also plays a role in moderating purchase intention. Previous studies have demonstrated that younger and older consumers respond differently when they are exposed to the same messages. Older buyers are more analytical and information-driven than younger buyers. They have more favorable views of the professionalism of influencers than aesthetics [60,61]. The key drivers of purchase intention for older consumers are product information, objective reviews, and trustworthy recommendations, whereas affective and parasocial cues have weak effects [61]. In contrast, younger consumers are more responsive to homogeneity, attractiveness, interactivity, and parasocial interaction because they are more socially active and identify with the influencer [61].

In the context of VIs, consumer age relates to differences in digital literacy, trust formation, and social engagement, which influence consumers’ information-processing abilities. Younger consumers, who are already accustomed to mediated and virtual interactions, view VIs as a social presence [62]. The attractiveness of a VI, together with the perceived similarity with the consumer, generates a positive response toward the influencer, leading to the acceptance of VIs [63]. Therefore, we propose that the impacts of human-likeness, customer engagement, and attitude on purchase intention are stronger for younger consumers than for older consumers.

H5a. 

The positive effects of (a) human-likeness, (b) customer engagement, and (c) attitude on purchase intention are stronger for younger than for older consumers.

In contrast, older consumers are more skeptical about VIs and rely more on professionalism, credibility, and informativeness cues than younger consumers regarding their purchase intentions [61]. Owing to variations in cognitive and affective processing based on age, the effect of source credibility on purchase intention may be stronger for older customers.

H5b. 

The positive effect of source credibility on purchase intention is stronger for older consumers than for younger consumers.

3.3.4. The Moderating Role of Cultural Factors

Cultural values may moderate the effects of antecedents on consumers’ purchase intentions. VIs may be effective in certain environments that govern consumer perceptions, trust, and engagement with social phenomena. For example, consumers in cultures with high levels of uncertainty avoidance and collectivism are more likely to have social cues, group trust, and a shared identity [13]. As a result, this study proposes that cultural dimensions moderate the relationship between antecedents and purchase intention. According to Hofstede and Bond [64], there are six cultural dimensions: (1) individualism/collectivism, (2) power distance, (3) uncertainty avoidance, (4) motivation towards achievement and success, (5) long-term/short-term orientation, and (6) indulgence.

First, individualism or collectivism refers to the extent to which people value personal goals over group goals [64]. In collectivist cultures, social networks, group norms, and trust play important roles in determining consumers’ purchase intentions. Hence, human-likeness, source credibility, and customer engagement of VIs may have a strong impact on purchase intentions in highly collectivist cultures because these attributes illustrate the trustworthy social connections and relational bonds that consumers rely on to make purchase decisions. In contrast, the purchase intention of individualist consumers is based on personal benefits and rational information processing [65,66,67]. In this culture, customers intend to purchase a product based on their independent consumer judgment rather than on social signals. Previous studies have demonstrated that customer attitudes have a greater effect on purchase intentions in individualistic cultures than in collectivist cultures [68]. Hence, this study proposes the following hypothesis:

H6a. 

The positive effects of (a) human-likeness, (b) source credibility, and (c) customer engagement on purchase intention are stronger for consumers in collectivist cultures than in individualistic cultures.

H6b. 

The positive effect of attitude on purchase intention is stronger for consumers in individualistic cultures than in collectivist cultures.

Second, power distance indicates whether individuals accept inequalities as avoidable or functional [64]. In cultures with a high power distance, a one-way message system positively annotates purchase intention. In contrast, in cultures with low power distance, purchase intention remains positive when an interactive system is used [69,70]. Therefore, when VIs show their expertise in high power distance cultures, customers may follow the advice of a VI positioned as an expert. Cultures with low power distances favor equality; therefore, human-likeness, customer engagement, and attitude may maximize their effectiveness.

H7a. 

The positive effect of source credibility on purchase intention is stronger in cultures with a high power distance.

H7b. 

The positive effects of (a) human-likeness, (b) customer engagement, and (c) attitude on purchase intention are stronger in cultures with low power distance.

Third, uncertainty avoidance is the extent to which people within a culture are nervous about situations that they perceive and experience as unstructured, unclear or unpredictable [64]. Thus, they try to avoid these by means of strict codes of behavior and a belief in the absolute truth. Consumers from high-uncertainty-avoidance cultures rely on reliable information in a clear and structured manner before making purchase decisions. These consumers have a strong preference for reliability. As such, the intention to purchase is significantly enhanced by transparency and secure platforms [67,70]. In the context of VIs, in cultures with high uncertainty avoidance, enhancing human-likeness, source credibility, customer engagement, and attitudes may function as key factors to mitigate skepticism. Specifically, human-likeness may reduce ambiguity, source credibility may ensure reliable information, customer engagement may create emotional interaction, and a positive attitude may help consumers overcome their initial skepticism. Hence, we propose the following hypothesis:

H8. 

The positive effects of (a) human-likeness, (b) source credibility, (c) customer engagement, and (d) attitude on purchase intention are stronger in cultures with high uncertainty avoidance than in cultures with low uncertainty avoidance.

Fourth, motivation towards achievement and success, formerly called masculinity/femininity, is a cultural orientation representing a society’s preference for competition and achievement as a means to overall quality of life [64]. The achievement orientation dimension distinguishes between performance- and relationship-oriented consumer behaviors [71]. The former group is responsive to competitive and success-oriented messages. The latter group prefers empathy and social relationship. In masculinity-oriented cultures, source credibility, especially the expertise of VIs, may be the most important prerequisite. Their persuasion, based on information value, helps shape purchase intentions in this culture [7]. Conversely, femininity-oriented cultures prioritize relationships over achievement and empathy over competition. In this context, human-likeness, customer engagement, and attitude may affect purchase intention more effectively. Hence, this study proposes the following hypothesis:

H9a. 

The positive effect of source credibility on purchase intention is stronger in achievement-oriented cultures (masculinity).

H9b. 

The positive effects of (a) human-likeness, (b) customer engagement, and (c) attitude on purchase intention are stronger in quality-of-life-oriented cultures (feminine).

Fifth, long-and short-term orientations are cultural dimensions that indicate the extent to which people in a culture prioritize future- or present-oriented values [64]. Buyers from long-term-oriented cultures who are exposed to functional information messages report higher purchase intentions when marketers highlight future-oriented benefits [70]. In the context of VIs, source credibility and customer engagement play a central role in creating long-term relationships with VIs [72]. In short-term cultures, customers prioritize immediate gratification. Human-likeness and a positive attitude have a greater influence in this context because purchase intentions are often impulsive owing to the attractive appeal of VIs [73]. Hence, this study proposes the following hypothesis:

H10a. 

The positive effects of (a) source credibility and (b) customer engagement on purchase intention are stronger in cultures with a long-term orientation.

H10b. 

The positive effects of (a) human-likeness and (b) attitude on purchase intention are stronger in cultures with a short-term orientation.

Ultimately, indulgence determines how freely people pursue pleasure [70]. Indulgent cultures respond more strongly to emotionally stimulating and pleasurable messages than restrained cultures do. Cultures that are restrained have a lower intention to purchase when exposed to such advertising appeals [71]. Because a culture of indulgence prioritizes pleasure, entertainment, and personal emotional expression, human-likeness, positive attitudes, and customer engagement may resonate with the need for experiential enjoyment [7]. Conversely, customers in restrained cultures are prohibited from using strict social norms. Therefore, source credibility may be a decisive factor in overcoming psychological defenses and stimulating purchase intentions. Hence, this study proposes the following hypothesis:

H11a. 

The positive effects of (a) human-likeness, (b) customer engagement, and (c) attitudes on purchase intention are stronger in indulgent cultures.

H11b. 

The positive effect of source credibility on purchase intention is stronger in cultures with restraint.

3.3.5. The Moderating Role of Government AI Readiness

The Government AI Readiness Index examines a country’s AI readiness through an analysis of 40 indicators within ten dimensions, which all fall under the three key pillars: government, technology, and data and infrastructure [74]. Previous studies have demonstrated that technology readiness, such as innovation orientation and technology optimism, improves attitudes toward and willingness to purchase technology [15,16]. For VIs, a country’s AI readiness may act as a moderating factor that influences consumers’ interpretations and reactions to their purchase intentions. Consumers in countries with high AI readiness may interact with AI systems and believe that VIs provide credible, innovative, and effective product information, which increases their purchase intention. In contrast, low government AI readiness can decrease purchase intention. Therefore, this study proposes that a country’s AI readiness moderates the relationship between antecedents and purchase intention.

H12. 

The positive effects of (a) human-likeness, (b) source credibility, (c) customer engagement, and (d) attitude on purchase intention are stronger in a country with high AI readiness than in one with low AI readiness.

4. Research Methods

4.1. Data Collection

To ensure methodological rigor and transparency, this meta-analysis was conducted according to the 2020 Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. A PRISMA 2020 checklist that has been fully completed is included in Appendix A.

Several databases, including Scopus, Web of Science, and Google Scholar, were used to identify all findings related to VIs. Key search terms were “artificial intelligence influencer*,” “AI influencer*,” “CGI influencer*,” “computer-generated influencer*” or “virtual influencer*”. All databases were searched using the title, abstract, and keywords of the studies. The search period was from 2018 to December 2025. In addition, to ensure that the literature retrieved was comprehensive, we manually searched four key journals (Journal of Retailing and Consumer Services, Journal of Business Research, Psychology & Marketing, and Computers in Human Behavior). This process yielded no new empirical studies, confirming the robustness of our primary database queries. Next, the authors used different inclusion criteria to purify the database using the PRISMA method (Appendix B) [75]. Studies including conceptual papers, qualitative studies, experimental studies comparing VIs to human influencers, and literature reviews were excluded from the review. This study retained only those studies that reported sufficient statistical information (e.g., sample and effect size).

4.2. Data Coding

This study created and utilized a coding sheet to obtain all data based on the operational definitions of the constructs (Appendix D). The manual provided information for coding, including all key variables of the conceptual model, moderators, sample size, and effect size (see Appendix E). The moderators included VI size (nano-, micro-, macro-, or mega-influencers), product type (search vs. experience; hedonic vs. utilitarian), consumer age, cultural factors (individualism vs. collectivism, power distance, uncertainty avoidance, motivation towards achievement and success, long-term orientation, and indulgence), and the AI Readiness Index (Appendix F). To enhance coding reliability, in addition to one of the authors, an independent coder (not a member of the research team and blinded to the study’s objectives and hypotheses) was invited to code all data independently. Discussions were held to resolve any differences between the two coders to ensure that similar results were obtained. The agreement rate, which was coded by two independent coders, was >90%.

4.3. Effect Size Integration

The effect sizes extracted from the chosen studies were transformed into correlation coefficients and corrected for measurement error. If there were no reliability coefficients, the mean reliability of the construct was used [76]. A random–effects model was used to integrate the effect sizes. To test the degree of heterogeneity of the effect sizes, this study used the I² and Q-statistic tests [77]. The Q values should be greater than the critical values, and the I² value should exceed 75% to demonstrate heterogeneity. Additionally, Rosenthal’s fail-safe N (FSN) was calculated to assess publication bias. FSNs should be larger than 5 × k + 10, where k is the number of studies [78]. Funnel plots were used to check for all relationships to adjust for publication bias. Egger’s test was also used to assess funnel plot asymmetry. This study also screened for outliers using standardized residuals greater than 3. All models were re-estimated with and without outliers to examine their effects on the results. All analyses were performed using R version 4.4.3 and the metafor package.

4.4. Meta-Analytic Structural Equation Model

Meta-analytic structural equation modeling (MASEM) was performed using Mplus 9 to test the proposed model. We compared the proposed model with three alternative models and one integrated model. The sample size for the MASEM was a harmonic mean of 3760. This study used several goodness-of-fit indices, including the Comparative Fit Index (CFI), Root Mean Square Residual (RMSEA), and Standardized Root Mean Square Residual (SRMR), to assess the fit of the models.

4.5. Moderator Analysis

To assess the effects of eight continuous moderators (e.g., culture and AI readiness) on purchase intention in VI marketing, a meta-regression analysis was conducted, as mentioned by Steel et al. [79]. Regarding categorical variables (e.g., VI size and product type), subgroup analysis was conducted to test the moderation effects [80]. Three moderators were coded: VI size (1 = mega- and macro-influencers, 0 = micro- and nano-influencers), product types (1 = experience products, 0 = search products; 1 = hedonic, 0 = utilitarian). Importantly, to control the false discovery rate (FDR), we applied the Benjamini–Hochberg (BH) procedure. This correction was calculated independently within the four groups of pairwise comparisons.

5. Results

5.1. Study Characteristics

The final database included 76 papers after applying all criteria, resulting in 186 effect sizes with a total sample of 64,545 (Appendix C). Since 2022, the number of research papers has steadily increased. In particular, 2024 and 2025 accounted for the largest proportions, with 28 papers (37%). Between 2022 and 2025, the number of publications increased by 7.5 times (Figure 3).

Among the publications analyzed, 67 were journal articles, while the rest included conference proceedings (6) and theses (3). A total of 46 different journals were analyzed, with the most prolific being with Journal of Retailing and Consumer Services (n = 6), Computers in Human Behavior (n = 4), Journal of Consumer Behavior (n = 3), and Journal of Theoretical and Applied Electronic Commerce Research (n = 3) (see

Table 3. Top journals in the reviewed studies.

Journal Name	Publisher	Publication Count
Journal of Retailing and Consumer Services	Elsevier	6
Computers in Human Behavior	Elsevier	4
Journal of Consumer Behavior	Wiley	3
Journal of Theoretical and Applied Electronic Commerce Research	MDPI	3
Asia Pacific Journal of Marketing and Logistics	Emerald Publishing	3
Journal of Business Research	Elsevier	2
Psychology & Marketing	Wiley	2
International Journal of Consumer Studies	Wiley	2
International Journal of Human–Computer Interaction	Taylor & Francis	2
Journal of Interactive Advertising	Taylor & Francis	2
Telematics and Informatics	Elsevier	2
Sustainability	MDPI	2

).

5.2. Integration of Effect Sizes

Table 4. Results of the effect size integration.

Relationship	k	N	rrcvw	CI	Q	I2	FSN
Human-likeness → Source credibility	29	10,810	0.43 ***	[0.34, 0.51]	797	96.8	715
Human-likeness → Customer engagement	5	2670	0.39 ***	[0.32, 0.46]	17.6	79.2	163
Human-likeness → Attitude	12	3282	0.33 ***	[0.16, 0.47]	249	96.1	108
Human-likeness → Purchase intention	32	10,858	0.55 ***	[0.38, 0.68]	4151	99.2	1410
Source credibility → Customer engagement	15	4402	0.48 ***	[0.31, 0.62]	464	97.8	376
Source credibility → Attitude	10	3058	0.47 ***	[0.24, 0.65]	604	98.1	314
Source credibility → Purchase intention	41	14,853	0.55 ***	[0.44, 0.64]	2942	98.8	2015
Customer engagement → Attitude	3	1208	0.67 ***	[0.33, 0.86]	82.6	98.4	317
Customer engagement → Purchase intention	27	9191	0.74 ***	[0.62, 0.82]	2680	99.0	2865
Attitude → Purchase intention	12	4213	0.75 ***	[0.58, 0.86]	1825	99.1	1581

k: number of studies; N: sample size; r_rcvw: reliability-corrected, variance-weighted average correlation; p < 0.001 ‘***’.

shows the effect size integration for different relationships. The calculated Q-statistics were greater than the critical values, and the I² indices were greater than 75%, indicating heterogeneity among the studies. Therefore, it is methodologically appropriate to conduct moderator analysis on these relationships. In addition, the FSN was larger than the thresholds, suggesting that the observed effects were robust. Egger’s regression tests did not indicate significant funnel plot asymmetry, suggesting no evidence of publication bias (Appendix G). Outlier analyses indicated that the results were robust; therefore, all studies were retained (Appendix H).

5.3. Structural Model Evaluation

We conducted the MASEM using a pooled correlation matrix (

Table 5. Correlation matrix of antecedents, mediators, and purchase intention.

	1.	2.	3.	4.	5.
1. Human-likeness (HL)	[0.87]	29	5	12	32
2. Source credibility (SR)	0.43	[0.88]	15	10	41
3. Customer engagement (CE)	0.39	0.48	[0.88]	3	27
4. Attitude (AT)	0.33	0.47	0.67	[0.89]	12
5. Purchase intention (PI)	0.55	0.55	0.74	0.75	[0.88]

Diagonal values represent weighted mean Cronbach’s alpha coefficients; the lower triangle shows inverse variance-weighted reliability-adjusted correlations, and the upper triangle shows the number of effect sizes.

). As detailed in Appendix I, the proposed model (Model 4) outperformed the alternative models (Models 1–3) and exhibited a satisfactory fit (χ²(2) = 218, p < 0.001, CFI = 0.98, RMSEA = 0.17, and SRMR = 0.07). In evaluating the models, we prioritized the CFI and SRMR values over the χ² and RMSEA because of two statistical limitations. Specifically, the χ² test is generally sensitive to large sample sizes (N = 3760) [81,82], while the RMSEA frequently falsely indicates poor fit when the degrees of freedom are small (df = 2) [83]. These findings were consistent with those of prior meta-analyses (e.g., Pan et al. [84]; Horstmeier et al. [85]; Maricuțoiu et al. [86]).

However, to ensure that the inflated χ² value stems from these limitations rather than structural omissions, we conducted a post hoc analysis using Modification Indices (MIs). The MIs suggested adding a direct path from human-likeness to customer engagement (M.I. = 209.39). Incorporating this path (Model 5) yielded a near-perfect fit (χ²(1) = 2.51, RMSEA = 0.020, CFI = 1.000, SRMR = 0.005). Despite this superior fit, we retained Model 4 as the primary framework because of two methodological constraints. First, Model 5 was near-saturated (df = 1, CFI = 1.000), raising substantial concerns regarding overfitting and theoretical parsimony. Furthermore, the relationship between human-likeness and customer engagement relies on a small number of studies (k = 5). Altering a model based on a few studies would yield more spurious results rather than reflecting the true theoretical relationships.

Furthermore, we conducted a sensitivity analysis to evaluate the structural necessity of the hypothesized engagement–attitude pathway, as its estimated effect size relies on a small sample of studies (k = 3). As detailed in

Table 6. Sensitivity Analysis: Full vs. Reduced Model Comparison.

Relationships	Full Integrative Model	Reduced Model
Human-likeness → Credibility	0.43 ***	0.43 ***
Credibility → Engagement	0.48 ***	0.48 ***
Credibility → Attitude	0.19 ***	0.47 ***
Engagement → Attitude	0.58 ***	–
Attitude → Purchase Intention	0.43 ***	0.46 ***
Engagement → Purchase Intention	0.32 ***	0.34 ***
Credibility → Purchase Intention	0.10 ***	0.11 ***
Human-likeness → Purchase Intention	0.25 ***	0.27 ***
Model Fit Indices
(df)	218.0 (2)	1750.4 (3)
CFI	0.98	0.81
SRMR	0.07	0.16
RMSEA	0.17	0.39

p < 0.001 ‘***’.

, constraining this path to zero resulted in a severe deterioration in the model fit (Δχ² (1) = 1532.4; CFI decreased to 0.81; RMSEA increased to 0.39). This degradation confirms that this relationship is structurally indispensable to the proposed S-O-R framework. Consequently, we retained the full integrative model (Model 4) as the most theoretically and empirically robust representation for the hypothesis testing.

5.4. Hypothesis Testing

The findings indicate that human-likeness positively influenced source credibility (β = 0.43, p < 0.001). Source credibility positively predicted customer engagement (β = 0.48, p < 0.001) and attitude (β = 0.19, p < 0.001), and customer engagement positively predicted attitude (β = 0.58, p < 0.001). Ultimately, purchase intention was positively driven by human-likeness (β = 0.25, p < 0.001), source credibility (β = 0.10, p < 0.001), customer engagement (β = 0.32, p < 0.001), and attitude (β = 0.43, p < 0.001). Figure 4 illustrates the results of the proposed model.

To assess the mediation effects, we analyzed the direct, indirect, and total effects on purchase intention (

Table 7. Direct effects.

Relationships	β	S.E.	z	p
Human-likeness → Purchase intention	0.25	0.010	25.44	<0.001
Source credibility → Purchase intention	0.10	0.011	9.10	<0.001
Customer engagement → Purchase intention	0.32	0.012	26.31	<0.001
Attitude → Purchase intention	0.43	0.012	36.53	<0.001

and

Table 8. Direct, indirect, and total effects on purchase intention.

Path	Direct Effect	Indirect Effect	Total Effect	VAF
HL → PI	0.25 ***	0.20 ***	0.45 ***	0.44
SC → PI	0.10 ***	0.36 ***	0.46 ***	0.78
CE → PI	0.32 ***	0.25 ***	0.57 ***	0.44
AT → PI	0.43 ***	–	0.43 ***	0

*** p < 0.001; VAF = Indirect/Total.

). As shown in Table 8, the overall Variance Accounted For (VAF) of 44% suggests that the mediating constructs act as important catalysts in translating human-likeness into purchase intention. Regarding mediation effects, source credibility had a significant direct effect on purchase intention (β = 0.10, p < 0.001); however, its indirect effect was even larger (β = 0.36, p < 0.001), accounting for 78% of the total effect. These findings demonstrate that source credibility alone is insufficient to increase purchase intention unless it stimulates customer engagement and fosters positive attitudes. These results suggest that an effective mechanism for purchase intention is through human-likeness, source credibility, customer engagement and attitude (

Table 9. Specific indirect effects on purchase intention.

Indirect Path	Std. Effect (β)	S.E.	z-Value	p-Value
HL → SC → PI	0.04	0.005	8.75	<0.001
HL → SC → CE → PI	0.06	0.004	17.97	<0.001
HL → SC → AT → PI	0.04	0.003	12.37	<0.001
HL → SC → CE → AT → PI	0.05	0.003	18.81	<0.001

).

5.5. Results of Moderator Analysis

There were 44 tests of moderators (4 pairwise relationships × 11 moderators) to evaluate their effects on purchase intention. The results of these analyses are presented in

Table 10. Results of sub-group analysis.

Relationship		Follower Size		Product Type
		k	1 = High Followers, 0 = Low Followers	k	Information Availability (1 = Experience, 0 = Search)	k	Usage (1 = Hedonic, 0 = Utilitarian)
HL → PI	r1	8	0.50	11	0.75 **	6	0.81 *
	r0	24	0.56	21	0.40 **	26	0.46 *
SC → PI	r1	6	0.69	13	0.50	6	0.45
	r0	35	0.52	28	0.57	35	0.56
CE → PI	r1	7	0.79	9	0.68	3	0.87
	r0	20	0.71	18	0.76	24	0.71
AT → PI	r1	6	0.79	3	0.75	1	0.74
	r0	6	0.71	9	0.76	11	0.76

p < 0.01 ‘**’, p < 0.05 ‘*’.

and

Table 11. Meta-regression analysis of national culture and age.

Relationship	Covariate	β	SE	p	95% CI
HL → PI	IND	0.004	0.011	0.72	−0.017	0.025
	POW	0.004	0.007	0.58	−0.010	0.018
	UNC	−0.004	0.007	0.56	−0.017	0.009
	MOT	0.004	0.013	0.77	−0.021	0.029
	LONG	0.009	0.007	0.17	−0.004	0.022
	INDU	−0.004	0.007	0.65	−0.017	0.010
	AIR	0.006	0.016	0.70	−0.025	0.036
	AGE	0.023	0.023	0.31	−0.022	0.068
SC → PI	INDI	0.006	0.005	0.21	−0.003	0.015
	POW	−0.001	0.005	0.83	−0.011	0.009
	UNC	−0.004	0.005	0.42	−0.013	0.005
	MOT	0.011	0.007	0.08	−0.002	0.024
	LONG	0.005	0.004	0.20	−0.003	0.013
	INDU	−0.005	0.005	0.38	−0.015	0.006
	AIR	0.019	0.011	0.08	−0.002	0.041
	AGE	0.026	0.013	0.05	0.000	0.052
CE → PI	INDI	7.3 × 10−5	0.007	0.99	−0.014	0.014
	POW	−0.005	0.006	0.40	−0.017	0.007
	UNC	0.016	0.007	0.03	0.002	0.031
	MOT	0.004	0.011	0.72	−0.018	0.026
	LONG	0.002	0.007	0.73	−0.012	0.017
	INDU	0.003	0.006	0.67	−0.009	0.014
	AIR	0.000	0.017	0.98	−0.032	0.033
	AGE	0.015	0.019	0.43	−0.022	0.051
AT → PI	INDI	0.005	0.008	0.57	−0.011	0.020
	POW	−0.003	0.011	0.80	−0.025	0.019
	UNC	0.005	0.007	0.48	−0.009	0.018
	MOT	−0.007	0.015	0.64	−0.037	0.023
	LONG	0.007	0.007	0.33	−0.007	0.021
	INDU	−0.008	0.010	0.41	−0.027	0.011
	AIR	0.003	0.015	0.86	−0.026	0.032
	AGE	0.021	0.026	0.42	−0.030	0.071

INDI: individualism/collectivism, POW: power distance, UNC: uncertainty avoidance, MOT: motivation towards achievement and success, LONG: long-term/short-term orientation, INDU: indulgence, AIR: Government AI readiness, AGE: consumer age.

. After applying the Benjamini–Hochberg FDR correction, none of the effects were found to be significant (

Table 12. Selected results of moderation analysis with the Benjamini–Hochberg correction.

Relationship	Moderators	Raw p-Value	BH Critical Value	Result
HL → PI	Product type (Experience vs. Search)	0.007	0.005	NS *
	Product type (Hedonic vs. Utilitarian)	0.014	0.009	NS
SC → PI	Consumer age	0.053	0.005	NS
CE → PI	Uncertainty Avoidance	0.031	0.005	NS

* NS: non-significant.

, full results in Appendix J). Consequently, all the initially significant results discussed below are interpreted as exploratory findings and potential contextual boundary conditions.

5.5.1. Follower Size (High vs. Low)

Based on prior influencer marketing research, we hypothesized that follower size moderates the relationship between antecedents and purchase intention. However, the present results indicate that follower size did not significantly moderate any relationship. In particular, follower size did not moderate the effects of anthropomorphism (r₁ = 0.50; r₀ = 0.56; p = 0.74), source credibility (r₁ = 0.69; r₀ = 0.52; p = 0.18), customer engagement (r₁ = 0.79; r₀ = 0.71; p = 0.46), or attitude (r₁ = 0.79; r₀ = 0.71; p = 0.63) on purchase intention. These findings suggest that, unlike traditional human influencers, the follower size of a VI is not a decisive factor in driving purchase intention.

5.5.2. Information-Based Product Types (Search vs. Experience)

In addition to follower size, this study examined whether product type (search vs. experience) influences the relationship between antecedents and purchase intention. Consistent with our hypothesis, the unadjusted results revealed that experiential products exhibited stronger moderating effects than search products on the relationship between human-likeness and purchase intention (r₁ = 0.75, r₀ = 0.40, unadjusted p = 0.007). However, this effect was not significant after the FDR correction. Viewed as an exploratory finding, this suggests that VIs with human-like characteristics may have a stronger impact on consumers’ purchase intentions when promoting experiential products than when promoting search products.

5.5.3. Motivation-Based Product Types (Hedonic vs. Utilitarian)

As hypothesized, the unadjusted results suggested that human-like cues are more effective for hedonic products than for utilitarian products (r₁ = 0.81, r₀ = 0.46, unadjusted p < 0.05). However, this effect did not hold after FDR correction. Therefore, viewed as an exploratory finding, this result suggests that human-like VIs may have a larger impact on purchase intentions for hedonic products than for utilitarian ones, implying that the effectiveness of human-likeness may be tied to a product’s hedonic value (associated with pleasure and enjoyment).

5.5.4. Consumer Age

We hypothesized that older consumers, being more analytical and skeptical, would rely more heavily on credibility cues than younger consumers would. The unadjusted results provided initial support for this hypothesis (β = 0.026, SE = 0.013, unadjusted p = 0.05). However, this effect did not hold after FDR correction. Viewed as a potential boundary condition, this finding suggests that the impact of source credibility on purchase intention may be stronger for older consumers than for younger ones.

5.5.5. Cultural Factors

Although prior research suggests that cultural differences broadly shape consumer responses, the unadjusted results revealed that only uncertainty avoidance had a significant effect on the relationship between customer engagement and purchase intention (β = 0.016, SE = 0.007, unadjusted p = 0.03). However, this effect was not significant after the FDR correction. Viewed as an exploratory finding, this suggests that customer engagement may function as a psychological reassurance mechanism in cultures with high uncertainty avoidance, highlighting the importance of cultural characteristics when implementing VI strategies in international markets.

5.5.6. AI Readiness

Contrary to our expectations, the results show that AI readiness does not moderate the relationship between the antecedents and purchase intention. In particular, AI readiness did not moderate the effects of anthropomorphism (β = 0.006, SE = 0.016, p = 0.70), source credibility (β = 0.019, SE = 0.011, p = 0.08), customer engagement (β = 0.000, SE = 0.017, p = 0.98), and attitude (β = 0.003, SE = 0.015, p = 0.86) on purchase intention. These findings suggest that a consumer’s decision to purchase from a VI may depend more on individual factors than on their country’s AI development.

6. General Discussion

6.1. Theoretical Contribution

First, addressing the gaps identified in the literature comparison (Table 1), this study proposes a comprehensive and VI-focused framework for understanding how human-likeness impacts purchase intentions. By including 186 effect sizes from 76 studies (N = 64,545), this study assessed three alternative models and one integrated model. The findings reveal that purchase intention is influenced by human-likeness through both direct and indirect effects, which are mediated by source credibility, customer engagement, and attitudes.

This result also clarifies the importance of social evaluation based on the Stereotype Content Model (SCM) [87]. According to the SCM, consumers typically consider others based on two important dimensions: perceived warmth and perceived competence [87]. In this context, human-likeness can enhance perceived warmth by making VIs appear more relatable, whereas source credibility can enhance perceived competence through expertise and trustworthiness. Customer engagement may further strengthen warmth and competence by increasing emotional interaction, thereby enhancing purchase intention.

Second, while follower size had a significant moderating effect on consumer behavior regarding human influencers [48,49], in the context of VIs, this study found no moderating effect on the relationship between the antecedents and purchase intention. This study contributes to the VI literature by indicating the diminished role of social validation cues. These differences between human influencers and VIs arise from the way consumers perceive them. For human endorsers, the number of followers can be a cue to their genuine social proof, as famous idols act as mega influencers in the real world [47]. In contrast, consumers interpret VIs as digital entities [88], thereby diminishing the social significance of the follower count. As mentioned in previous studies, consumers tend to engage with human influencers rather than VIs [89]. In these scenarios, consumers may neglect social proof (i.e., the number of followers) and prioritize the VIs’ believability and psychological appeal.

Third, this study contributes to the literature by identifying an exploratory finding regarding the moderating effect of product type in the VI context. Specifically, human-likeness appears to have a stronger impact on purchase intentions for hedonic and experiential products than for utilitarian and search products. This preliminary result may be explained by the different ways in which consumers evaluate hedonic/experiential and utilitarian/search products. Previous studies have highlighted the dependence of hedonic and experiential products on affective processing and anticipated consumption [54,57]. In this context, the entertainment value provided by human-like VIs is highly effective for persuasion [7]. Human-likeness may enhance consumers’ emotional imagination and perceived social connection, facilitating a mental simulation of the product experience. Hence, human likeness appears to have a greater impact on the purchase intention of products requiring emotional engagement than on those evaluated based on functional performance.

Fourth, prior research on technology adoption highlights that older consumers are inherently resistant to technologies due to lower digital literacy or social isolation issues [90]. This study contributes to the literature by providing preliminary signals that older consumers might still adopt VIs under specific conditions. In particular, this exploratory finding showed that older consumers respond positively to VIs that display credible behaviors. This result extends the study of Benson et al. [91] into the VI context. Accordingly, young consumers are tech-savvy and have higher risk tolerance. In contrast, older consumers make fewer purchases on social media because they are more concerned about privacy, trust, and information control [91]. In other words, if VIs can establish strong source credibility, older consumers may overcome these barriers and formulate purchase intentions.

Fifth, this study extends the cross-cultural AI adoption literature by highlighting uncertainty avoidance (UA) as an exploratory cultural moderator. Specifically, UA may amplify the relationship between customer engagement and purchase intention. Previous studies have demonstrated that individuals from high uncertainty avoidance cultures experience discomfort in situations perceived as unclear or unpredictable [64]. Consequently, they seek greater certainty before making purchase decisions. In this context, customer engagement may act as an uncertainty-reduction mechanism [92]. Engaging with VIs may foster familiarity, predictability, and psychological comfort among users. Therefore, for consumers from high uncertainty avoidance cultures, deeper engagement with VIs could tentatively provide the confidence needed to make purchasing decisions.

Sixth, previous studies have demonstrated that consumers’ purchase behavior in less developed nations is influenced by technological infrastructure [93]. However, the findings of this study challenge these studies by indicating that a country’s AI readiness does not moderate the relationship between these antecedents and purchase intention. This lack of moderation may stem from individual-level psychological processes. Because purchase intention is an individual (micro-level) psychological process, customers may purchase a product based on whether the virtual agent looks real and whether it is trustworthy and engaging [7]. They are largely unconcerned with macro-level factors such as national AI readiness. Consequently, national AI readiness may be less relevant in this context than other factors. Therefore, customers from countries at different stages of national AI development exhibit similar behavioral reactions to VI marketing.

6.2. Practical Implications

6.2.1. The Importance of Human-Likeness

In this study, human-likeness is an important antecedent that creates source credibility and influences customer engagement and attitude, leading to enhanced purchase intention. Marketers, brand managers, and organizations that employ VI marketing strategies can leverage human-like characteristics to increase purchase intentions. Brands should design VIs that emphasize the visual, behavioral, and emotional dimensions. For instance, facial movements, gestures, and other body language that mimic real human communication can also result in a higher purchase intention [94]. Emotional human-likeness should also be focused on, which consists of the infusion of affective expressions, such as joy and excitement, that are aligned with the tone of the conversation and are consistent with the brand message [27]. More importantly, consistent with the uncanny valley effect, making VIs appear too realistic may create discomfort or reduce consumer acceptance levels. Therefore, brands should create human-likeness that appears natural but is not overly realistic. In addition, according to the pratfall effect, small imperfections can make highly competent individuals appear more human and likable. Therefore, brands should consider creating minor flaws in VIs to make them appear more engaging.

6.2.2. Generational Strategies in VI Marketing

In this study, generational differences emerged as a potential boundary condition, offering exploratory guidance to marketing managers. Specifically, brands should focus on building source credibility among older consumers to strengthen their purchase intentions. Campaigns should use advertising content to enhance perceived attractiveness, expertise, and trustworthiness [39]. For example, managers can incorporate cognitive cues that showcase their consciousness in influencer content [95]. In addition, humor and curiosity can foster trustworthy impressions [96]. By combining these elements, brands can attract the attention of older consumers and increase their purchase intentions. Conversely, brands should consider shifting their focus from source credibility to other factors, such as customer engagement, to be more effective for younger consumers. Similarly to human influencers, brands can use storytelling, emotional connections, or first-person pronouns to stimulate consumer engagement [97]. Creating content that aligns with audience interests further increases engagement, especially when the content is entertaining, fun, and thrilling [7].

6.2.3. Aligning VIs with Product Types

The unadjusted results of this study suggest that information- and motivation-based product types may moderate the effectiveness of VIs, demonstrating the potential importance of choosing product types for VI marketing [98]. Therefore, to improve purchase intention, the VI and product type should be consistent with each other. In particular, for hedonic and experiential products such as cosmetics, fashion, travel, and entertainment, marketers should focus on visual, behavioral, and emotional human likeness [24,99]. In contrast, when promoting utilitarian or search products, such as electronics or insurance, brands should focus on technical quality, informative clarity, and rational rather than emotional appeal.

6.2.4. Adapting VI Strategies to Cultural Orientation

Culture has been demonstrated to be important in designing marketing strategies for multinational corporations. This study suggests that uncertainty avoidance tentatively moderates the relationship between customer engagement and purchase intention. As an exploratory insight, in high-uncertainty-avoidance markets (Greece, Japan, and Korea), brands should consider applying customer engagement strategies that reduce perceived risk and reassure consumers.

6.2.5. Emphasizing AI Quality over Follower Quantity and Government AI Readiness

This finding confirms that the number of followers does not affect purchase intention. In addition, the study showed that countries with weak AI readiness can still foster higher consumer purchase intentions. Therefore, regardless of whether the country is AI-ready, brands could target smaller VIs with more attractiveness, credibility, or trustworthiness, who would achieve maximum purchase intention rather than spending on VIs with millions of followers.

7. Conclusions

This meta-analysis collected 186 effect sizes from 76 studies (N = 64,545) to investigate the mechanisms and moderating effects of the factors affecting consumers’ purchase intentions. The findings show that human-likeness is a key antecedent that directly and indirectly affects purchase intention through source credibility, customer engagement, and attitudes. This study also evaluated the moderating effects of follower size, product type, consumer age, culture, and government AI readiness. The results revealed that follower size and national AI readiness demonstrated no moderating effects, challenging traditional social proof assumptions. Furthermore, because the remaining moderators did not survive the FDR correction, they offered valuable exploratory insights. Specifically, human-likeness appears to be more impactful for hedonic and experiential products than for utilitarian and search products, while consumer age and uncertainty avoidance may tentatively shape how credibility and engagement influence purchase decisions. This study provides actionable insights for market researchers and brand strategists, who may use these findings to improve consumer trust, foster emotional attachment, and underpin VI acceptability as representatives of the brand to achieve higher purchase intentions (

Table 13. Practical implications for VI marketing.

Strategies	Context/Target	Guidelines
The design of human-likeness	In the context of digital commerce	Program natural visual, behavioral, and emotional cues (e.g., facial gestures and joy). Avoid overly realistic design. Introduce minor flaws. Prioritize VI quality over quantity.
Generational targeting	Older consumers	Consider focusing on building source credibility, expertise, and trust (e.g., by incorporating cognitive cues).
	Younger consumers	Consider focusing on maximizing customer engagement (e.g., using storytelling, first-person pronouns, and highly entertaining content).
Product alignment	Hedonic & experiential products (e.g., cosmetics, fashion)	Consider emphasizing the visual, behavioral, and emotional human-likeness of VI.
	Utilitarian & search products (e.g., electronics, insurance)	Consider emphasizing technical quality, information clarity, and rational appeal.
Country targeting	High uncertainty avoidance markets	Consider designing engagement strategies that actively reduce perceived risks and reassure consumers.

).

8. Limitations and Future Research Directions

While the above contributions are mentioned, some limitations are related to the scope of this study.

First, similar to other meta-analyses, this study encountered heterogeneity in the operationalization of key constructs. For example, source credibility is measured by expertise, attractiveness, and trustworthiness. Future research should develop standardized and validated measurement frameworks for these constructs across different VI types and cultures.

Second, this study focused on psychological mechanisms such as credibility, engagement, and attitudes. Other relevant constructs such as perceived warmth, perceived competence, parasocial interaction, uncanny valley, or authenticity were not examined due to data availability. Future studies should examine how the uncanny valley, perceived warmth, and competence shape responses to VIs. In addition, future research could explore whether the pratfall effect applies to VIs, particularly whether minor imperfections increase perceived authenticity [100].

Third, many recent studies have focused on purchase intention as a reflection of behavioral responses. Future studies should focus on purchase behavior by examining whether psychological variables, such as trust and authenticity, affect consumer behavior (e.g., brand loyalty and repeat purchases), which would clarify the long-term impact of VIs.

Fourth, because the moderating effects of product type, consumer age, and uncertainty avoidance did not survive rigorous FDR correction and currently serve as exploratory findings, future studies are needed to statistically validate these boundary conditions. Furthermore, researchers should explore additional moderators such as brand familiarity and consumer gender. Regarding cultural factors, a near-significant result provides further discussion of the topic [101]. Specifically, the marginal moderation of the customer engagement–purchase intention relationship by motivation towards achievement and success suggests that consumers from cultures with a high achievement orientation may be more likely to engage with VIs. Therefore, this finding is important for future investigations. Furthermore, previous studies have highlighted the effectiveness of TikTok, YouTube, and X (formerly Twitter) in influencing purchase intentions. In this study, the data pertaining to these platforms were inadequate. Future research could conduct platform-comparative analyses to determine where VIs can most effectively convert trust and engagement into purchase intentions and actual behavior.

Finally, a methodological limitation concerns the constrained statistical power of this study’s data. Specifically, a few studies are insufficient to draw robust conclusions regarding the relationship between customer engagement and attitude, human-likeness and customer engagement. Therefore, the findings for these paths were interpreted cautiously. Future research should accumulate additional data to re-examine and validate the underlying mechanisms of VIs.