Towards a Sustainable Intelligent Transformation in E-Commerce: An Empirical Study of User Expectations and Perceptions of Virtual Anchors

Abstract

E-commerce live streaming is increasingly constrained by the “anchor dilemma” of talent shortages and reputational volatility. Virtual anchors are viewed as a critical nexus for intelligent and sustainable e-commerce transformation, offering scalable and low-carbon potential. Yet, their user experience and perception remain underexplored. Methodologically, this study adopts a mixed empirical design combining literature review, expert interviews, and a structured questionnaire survey (N = 309), followed by reliability testing, paired-sample t-tests, and Importance–Performance Analysis (IPA) to assess user expectations and perceptions. The integrated analysis resulted in a framework of fourteen evaluative attributes, within which spectacle and cross-platformity emerged as distinguishable dimensions observed in participants’ assessments. The results show that expectations (M = 4.41) significantly exceed perceptions (M = 3.74), with all 14 importance–performance gaps reaching significance. Interactivity, professionalism, and technological maturity emerged as priority areas for improvement, while spectacle and novelty were confirmed as key advantages, and credibility and emotional bonding outperformed expectations. Based on these findings, a phased strategy is proposed: short-term optimization of interaction and knowledge support, mid-term development of human–AI collaboration and platform adaptability, and long-term establishment of governance and commercialization ecosystems. The study enriches virtual anchor research and highlights that enhancing core competencies is essential to transform novelty into enduring sales and brand equity, providing a practical pathway for e-commerce’s intelligent and sustainable transformation.

1. Introduction

Sustainability has become one of the most critical issues of our time [1], and many corporations now claim to operate sustainably [2]. Similarly, sustainability has also become an indispensable issue in the field of e-commerce. Online commerce offers consumers greater product variety and more competitive prices, attracting an ever-growing number of users. As consumers’ environmental awareness continues to rise, they increasingly expect companies to adopt green and sustainable e-commerce practices [1,3]. Therefore, discussions on the development of e-commerce cannot be separated from the issue of sustainability [4,5,6,7].

With the rapid advancement of artificial intelligence (AI), virtual reality (VR), and other frontier digital technologies, the intelligent transformation of e-commerce platforms has accelerated substantially. The interaction between consumers and platforms is increasingly mediated by AI-generated content (AIGC), digital employees, and automated interaction systems, reshaping both the organizational structure and operational logic of e-commerce live streaming. Within this shift, virtual anchors—created through the deep integration of technological capabilities and content production—have become an emerging intelligent media role and are increasingly positioned as alternatives to traditional human anchors in live-streaming commerce [8]. Compared with human anchors, virtual anchors offer advantages such as higher controllability, stronger personalization, and lower operational costs [9]. They also present potential sustainability benefits in digital labor substitution [10], as they are not constrained by time, physical resources, or human fatigue, making them a promising component in the broader pursuit of sustainable e-commerce development.

Virtual anchors generally refer to AI-driven digital avatars capable of performing human-like behaviors through virtual reality rendering, speech synthesis, facial expression generation, and deep learning algorithms [11,12]. Their application in e-commerce live streaming has expanded rapidly, attracting significant academic and industry attention. Existing studies on virtual anchors mainly cluster around three research directions. First, technical research emphasizes implementation mechanisms such as fidelity [13], speech synthesis [14], facial expression generation [15], and motion capture [16], focusing on enhancing human-like attributes [17] and interaction fluency [18], both of which influence immersion and user acceptance [19]. Second, consumer-oriented studies investigate how virtual anchors’ characteristics [10,20] affect user experience [21], engagement [22], purchase intention [23], and brand loyalty [24], often drawing on frameworks such as Stimulus–Organism–Response (S-O-R) [20]. Third, emerging literature has begun to explore their macro-level value in sustainability transitions [25], suggesting that virtual anchors can reduce labor dependence [26], lower the resource consumption of live-stream production [27], and foster greener content ecosystems [28]. However, several gaps remain. Although existing studies have examined various features of virtual anchors, their multi-attribute structure has not been systematically conceptualized. Current research tends to converge on explaining purchase behavior, leading to limited diversity in analytical perspectives. Moreover, the gap between consumers’ expectations of virtual-anchor attributes and their actual perceptions has received insufficient attention. The lack of empirical measurement of expectation–perception discrepancies limits both theoretical understanding and the identification of precise optimization pathways for virtual-anchor development.

Building on these gaps, this study differs from prior research in two important ways. First, instead of focusing on isolated attributes or single behavioral outcomes, we construct a more comprehensive multi-attribute evaluation framework that captures how users simultaneously assess different dimensions of virtual anchors. Second, rather than examining user responses in absolute terms, we adopt an expectation–perception perspective to empirically measure the discrepancies between what consumers anticipate and what they actually experience. This approach provides a more diagnostic understanding of where virtual anchors underperform or exceed expectations, thereby offering clearer guidance for their optimization and sustainable development in e-commerce live streaming. To address the intelligent transformation toward sustainable e-commerce, this study focuses on e-commerce live streaming and examines virtual anchors from a user-centered “expectation–perception” perspective. Employing the Importance–Performance Analysis (IPA) method, the study quantitatively identifies key attributes of virtual anchors and diagnoses the gaps between user expectations and perceptions, thereby proposing optimization strategies for sustainable e-commerce transformation. The specific research questions are as follows:

RQ1: In the context of e-commerce live streaming, how can a comprehensive attribute framework for virtual anchors be constructed, and do significant gaps exist between users’ expectations and their actual perceptions of these attributes?

RQ2: Based on the IPA framework, into which of the four quadrants, namely “Keep up the good work,” “Maintain current status,” “Low-priority development,” and “Concentrate here,” do the current attributes of virtual anchors in e-commerce live streaming fall?

RQ3: How do the performance levels of different attributes influence the strategic value of virtual anchors in sustainable e-commerce, and what potential impacts do they have on platform operations, consumer engagement, and the promotion of green transformation?

This study is expected to make both theoretical and practical contributions. Theoretically, by developing a multi-attribute evaluation framework and applying an expectation–perception lens, the study provides a more diagnostic understanding of how users cognitively assess virtual anchors across multiple dimensions. This approach enriches existing research that has predominantly examined single factors or behavioral outcomes and offers a structured basis for conceptualizing virtual anchors as multi-dimensional digital entities within e-commerce ecosystems. Practically, the findings can inform more precise optimization strategies for virtual-anchor design and deployment by identifying where performance deficits and strengths most significantly diverge from user expectations. These insights also provide actionable guidance for platforms and enterprises seeking to advance the sustainable and intelligent transformation of e-commerce through more efficient, scalable, and user-aligned virtual-anchor systems.

2. Literature Review

2.1. Applications of Virtual Anchors Across Different Contexts

Virtual anchors, as a form of digital labor, have gradually been applied across diverse scenarios, including news broadcasting, short video content, social entertainment, and e-commerce live streaming [23]. From a technological perspective, the development of virtual anchors relies on the integration and advancement of underlying AI capabilities such as computer graphics, natural language processing, and deep learning. Li and Allbeck [29] emphasized that integrating commonsense knowledge and environmental awareness into the behavioral system of virtual characters is essential for enhancing their interactive capabilities. Years later, they [30] further argued that the key to constructing virtual humans lies not only in the quality of visual modeling but also in establishing a perception–expression mechanism driven by natural language, enabling virtual agents to respond in real time with more natural and human-like interactions [31].

At the application level, research and practice on virtual anchors or virtual humans have expanded across multiple domains, giving rise to diverse functional roles and research agendas. First, in the fields of news and broadcasting, AI anchors have achieved 24 h automated reporting, multilingual transmission, and rapid data-driven news summarization [32]. However, these developments have also sparked academic and ethical debates concerning media ethics [33], bias dissemination [34], and audience trust [35]. Second, research in companionship and health support indicates that virtual human anchors can play positive roles in educational tutoring [36], companionship, and emotional assistance [37]. Empirical studies have reported their potential value in alleviating loneliness and improving quality of life, while also emphasizing the importance of ensuring accessibility and safeguarding user privacy [38]. Third, within the ecosystem of social entertainment and virtual influencers, virtual characters, owing to their highly customizable visual persona [39], reproducible content production capacity, and cross-media storytelling potential, have been widely employed in brand communication and content marketing [40,41]. Fourth, in the field of AI customer service [42], virtual agents are increasingly used to provide users with more accurate product and service consultations [43], thereby enabling large-scale automation and personalized recommendations. The accumulated experience across these diverse application scenarios has directly facilitated the rapid implementation of virtual anchors in e-commerce contexts. In the e-commerce context, virtual anchors, as a form of “novel content intermediaries,” are reshaping the way platforms and users are connected. Unlike traditional human anchors, virtual anchors possess high plasticity and controllability, and are not constrained by time, space, or physical endurance. They are capable of continuous broadcasting and simultaneous operation across multiple platforms [26,44]. These capabilities may position them as a crucial balancing point for e-commerce platforms seeking to standardize content, maximize operational efficiency, and control costs.

2.2. Construction of Virtual Anchor Attribute Indicators

Identifying appropriate evaluative indicators for virtual-anchor attributes constitutes a central methodological task in IPA-based studies [45]. Although the characteristics of human anchors have been extensively examined in conventional e-commerce, research on virtual-anchor attributes remains limited, resulting in the absence of a systematic and comprehensive evaluative framework. Wu et al. suggest that indicator identification can be advanced through self-report and expert-informed approaches [46].

Accordingly, this study first conducted a comprehensive review of prior research on virtual anchors, AI anchors, virtual streamers, and virtual digital humans, using well-established human-anchor attributes as a conceptual reference. To further enhance content validity and ensure that the proposed indicators accurately reflect industry-relevant perceptions, we conducted two rounds of unstructured interviews. One interview was carried out with a scholar specializing in digital communication, and the other with an experienced senior live-streaming practitioner. As an exploratory technique, unstructured interviews did not follow a predefined question list or procedural sequence; instead, the interviewer guided the conversation through open-ended prompts, allowing participants to freely articulate their observations, evaluative criteria, and experiential insights [47]. Each interview lasted approximately 15–20 min and was audio-recorded with the participants’ consent. Detailed field notes were subsequently compiled based on the recordings. The research team then conducted an inductive analysis of the narrative materials, extracting relevant content as to supplement the analysis of this study, and then translating them accordingly. These emergent themes and dimensions were further compared and integrated with insights from the literature review. The combined evidence from these exploratory interviews and the existing research enabled us to refine, extend, and consolidate dispersed attribute indicators into a coherent multi-attribute framework. The final set of indicators, which forms the empirical basis for the expectation–perception analysis, is presented in

Table 1. Studies on Virtual Anchor Attributes.

Scholars	Research Content	Source
Wang, Y. X [48]	The study verified that the novelty, credibility, human-likeness, and agency of virtual anchors positively influence audience acceptance of AI anchors through the mediating role of attitude.	literature
Tian, X [49]	The human-like appearance and behavior of virtual anchors, along with real-time interactivity and diverse modes of interaction, have the potential to influence consumers’ emotions. Their effect on engagement behavior is particularly pronounced when they can effectively convey information and demonstrate professional competence.	literature
Franke, C [50]	An increasing number of brands are employing virtual characters as their endorsers. Does this trend arise because virtual characters are more appealing than human ones, or because they can provide greater advertising novelty?	literature
Liu, M. S [51]	The study categorized the characteristics of news AI anchors into four dimensions: human-likeness, professionalism, likability, and intelligence, in order to examine their effects on users’ trust in the anchors.	literature
Yu, Y [52]	The study demonstrated that virtual anchors’ appearance attractiveness, interactivity, and entertainment features positively influence audiences’ purchase intentions.	literature
Jinhui, K [53]	The study identified five primary factors influencing the interaction effectiveness of AI virtual anchors: personalized interaction experience, entertainment value, real-time responsiveness, efficient content delivery, and realism of the anchor’s image. These factors affect customers’ sense of social presence, which in turn impacts their purchase intentions.	literature
Zhong, D [54]	The study examined the relationship between consumers’ perceived human-likeness and perceived intelligence of virtual anchors and their trust, and further validated that trust subsequently influences consumers’ purchase intentions.	literature
Dr. Wang	From a special effects perspective, virtual anchors really show off the huge potential of technology. In traditional effects production, we’re often limited by time, budget, and tech, so some supernatural effects are just impossible. But with virtual anchors, advanced digital tech and effects make it easy to pull off amazing supernatural abilities like flying, transforming, or teleporting. Using this tech not only gives virtual anchors way more room to perform, but also delivers an incredible visual punch and immersive experience for the audience.	Interview
Zhang (anchor)	You know, as a host, I often feel the challenges that come with different platforms. Audiences on each platform have their own tastes and preferences. Sometimes, platform contracts even stop us from freely doing activities on other platforms, which makes things even harder. But virtual anchors could be a whole new advantage, even an opportunity, and they might help us achieve much greater success in e-commerce live streaming.	Interview

.

A summary of Table 1 reveals that the dimensional categorization of virtual anchor attributes in each study was established according to specific research needs. To date, there has been no systematic investigation or synthesis of virtual anchor attributes. Building upon the findings of the aforementioned studies and insights from discussions with industry experts and scholars, this study classifies virtual anchor attributes into two main categories: Human-like attributes and technical attributes [55,56]. The Human-like attributes encompass Cross-platformity, Anthropomorphic appearance, Interactivity, Credibility, Behavioral agency, Emotional Bonding, and Entertainment; the technical attributes include Novelty, Intelligence, Spectacle, Expressiveness, and Professionalism. Together, these constitute two dimensions with a total of twelve indicators, serving as the evaluation framework for virtual anchor attributes. The specific meaning of each attribute in the context of this study is presented in

Table 2. Description of Virtual Anchor Attributes.

Dimension	Attribute	Attribute Description
Human-like attributes	Cross-platformity	The ability of virtual anchors to maintain a consistent identity and performance across multiple platforms.
	Anthropomorphic appearance	The degree to which virtual anchors simulate human characteristics in visual and behavioral design.
	Interactivity	The ability of virtual anchors to engage in real-time dialogue, feedback, and contextual responses with users.
	Credibility	The extent to which users perceive the information or recommendations provided by virtual anchors as trustworthy and reliable.
	Behavioral agency	The degree to which virtual anchors demonstrate autonomy and situational adaptability in their actions.
	Emotional bonding	The emotional attachment and sense of companionship formed by users through prolonged interaction with virtual anchors.
	Entertainment	The ability of virtual anchors to provide pleasure, amusement, and immersive experiences during live streaming.
Technical attributes	Novelty	The innovativeness and originality of virtual anchors in design, performance, or content, reflecting their technological appeal.
	Intelligence	The ability of virtual anchors to understand, reason, and respond appropriately based on AI technologies.
	Spectacle	The display of visually striking and technologically enhanced effects that transcend reality, providing users with a sense of “supernatural power.”
	Expressiveness	The richness and accuracy of verbal and non-verbal expressions demonstrated by virtual anchors.
	Professionalism	The depth of professional knowledge and the accuracy of information exhibited by virtual anchors.

.

2.3. Expectation–Perception Framework and Importance–Performance Analysis (IPA)

Before introducing the IPA approach, it is important to theoretically contextualize the expectation–perception logic underlying this study. Prior research in technology adoption and mediated interaction has consistently shown that users form ex ante expectations toward digital systems and subsequently evaluate their experience. For example, the Technology Acceptance Model (TAM) posits that individuals develop cognitive expectations about system usefulness and ease of use, which shape later evaluations and attitudes [57]. Similarly, the Stimulus–Organism–Response (SOR) framework conceptualizes user experience as a process in which environmental or technological stimuli generate internal cognitive–affective states that influence subsequent evaluations and behavioral responses [58]. Social presence theory further emphasizes that users’ expectations about warmth, humanness, and interpersonal cues often diverge from their lived experiences of digitally mediated agents [59]. Across these frameworks, expectation–experience discrepancies are treated not as incidental variations but as a meaningful psychological mechanism through which users make sense of digital technologies.

Although the present study does not employ TAM [57], SOR [58], or social presence theory [59] as full structural models, their underlying principles provide conceptual support for adopting an expectation–perception lens. What these theories collectively suggest is that user judgments about digital agents—including virtual anchors—depend not only on the attributes themselves but also on how perceived performance aligns with prior expectations. This theoretical foundation justifies the analytical premise of this study and complements the diagnostic focus of the Importance–Performance Analysis (IPA), which operationalizes expectation–perception differences at the attribute level. On this basis, IPA offers a suitable and theory-informed approach for examining the multidimensional evaluation structure of virtual anchors.

The Importance–Performance Analysis (IPA) was originally proposed by Martilla and James [60] in 1977 and has been further developed and applied in subsequent studies. This method aims to identify gaps between user expectations (importance) and perceptions (performance) regarding certain attributes or factors [60,61], in order to inform the development of optimization strategies and improvement measures. Based on the classification results of the IPA model, business operators can allocate limited resources and formulate reasonable product improvement strategies accordingly [62]. The IPA method has been widely applied and validated in various fields, including healthcare [63,64], culture [65], tourism [66,67], education [68,69], online communities [70,71], and e-commerce [72,73]. In this study, IPA was selected because it allows for an in-depth analysis of users’ expectations and perceptions of virtual anchor attributes. This approach provides an objective understanding of the relative importance of each attribute and its actual performance, thereby offering a scientific basis and practical guidance for optimizing virtual anchors in e-commerce live streaming.

Based on the two dimensions of expectation (importance) and perception (performance), four quadrants are defined. Attributes located in the first quadrant exhibit both high expectation and high perception, representing the strengths of virtual anchors, which should be maintained in subsequent design and development. The second quadrant includes attributes with low expectation but high perception; for these, developers should sustain their strong performance. Attributes in the third quadrant show both low expectation and low perception, representing minor weaknesses, which can be developed if sufficient resources are available. The fourth quadrant comprises attributes with high expectation but low perception, representing major weaknesses of virtual anchors that should be prioritized for improvement.

3. Research Design

3.1. Research Framework and Procedure

This study takes e-commerce live streaming as the application context, aiming to systematically assess users’ expectations and perceptions of virtual anchor attributes and to explore optimization pathways for sustainable transformation in e-commerce. The overall research framework is illustrated in Figure 1, comprising four main steps: the formulation of research questions, the construction of virtual anchor attribute indicators, data collection and analysis, and the generation of optimization strategies.

After defining the research questions, the study first conducted a literature review and expert interviews to systematically identify the core attributes of virtual anchors, and constructed an evaluation indicator system based on the two dimensions of human-like attributes and technical attributes. In this process, existing research on virtual humans and e-commerce anchors was referenced, and the practical experiences of industry experts and practitioners were incorporated to ensure the scientific rigor and applicability of the indicators. Second, user data were collected through a questionnaire survey, with all items measured using a five-point Likert scale. The collected data were then processed and visualized using the Importance–Performance Analysis (IPA) method. This approach compares the mean importance (I) and mean performance (P) of each attribute across two dimensions, dividing them into four quadrants to reveal the priority of different attributes. Finally, based on the quadrant results from the IPA, this study proposes optimization strategies for virtual anchors in e-commerce live streaming, with a focus on enhancing user experience while promoting intelligent and sustainable development in e-commerce. This stepwise design ensures both theoretical rigor and practical applicability.

3.2. Indicator Measurement and Questionnaire Design

This study systematically reviewed the attributes of virtual anchors, and based on expert interview findings, ultimately established two dimensions and twelve attribute indicators. The two dimensions are human-like attributes and technical attributes. Human-like attributes include cross-platformity, anthropomorphic appearance, interactivity, credibility, behavioral agency, emotional bonding, and entertainment; technical attributes include novelty, intelligence, spectacle, expressiveness, and professionalism. All constructs and items were adapted from validated scales in prior studies and refined to fit the research context.

The questionnaire consisted of two parts. The first part collected respondents’ basic demographic information, including gender, age, and education level (See

Table A1. Demographic Information.

1-Gender
1. Male	2. Female
2-Age
1. 19 years or below	2. 20–29 years
3. 30–39 years	4. 40 years or above
3-Level of Education
1. Junior high school degree	2. High school degree
3. Junior college degree	4. Bachelor’s degree
5. Master’s degree or above

), to depict the sample structure and facilitate subsequent heterogeneity analysis. The second part was designed based on the two dimensions and twelve attribute indicators, with questions assessing users’ expectations (importance) and perceptions (performance) of virtual anchors in e-commerce live streaming (See

Table A2. Questions of the Questionnaire.

Very Unimportant/ Very Dissatisfied	Unimportant/ Dissatisfied	Neutral	Important/ Satisfied		Very Important/ Very Satisfied
1	2	3	4		5
To What Extent
(1) How important do you consider the following attributes of virtual anchors in e-commerce live streaming? Please rate each attribute on a 5-point Likert scale according to its importance in your expectations. (5 = Very important, 4 = Important, 3 = Neutral, 2 = Unimportant, 1 = Very unimportant)
No.		1	2	3	4	5
1	Human-like attributes
2	Cross-platformity
3	Anthropomorphic appearance
4	Interactivity
5	Credibility
6	Behavioral agency
7	Emotional bonding
8	Entertainment
9	Technical attributes
10	Novelty
11	Intelligence
12	Spectacle
13	Expressiveness
14	Professionalism
(2) How satisfied are you with the performance of the following characteristics of virtual anchors in e-commerce live streaming? Please rate each item on a 5-point scale according to your actual experience. (5 = Very satisfied, 4 = Satisfied, 3 = Neutral, 2 = Dissatisfied, 1 = Very dissatisfied)
No.	Questionnaire Content	1	2	3	4	5
1	Human-like attributes
2	Cross-platformity
3	Anthropomorphic appearance
4	Interactivity
5	Credibility
6	Behavioral agency
7	Emotional bonding
8	Entertainment
9	Technical attributes
10	Novelty
11	Intelligence
12	Spectacle
13	Expressiveness
14	Professionalism

). A five-point Likert scale was employed, where “1” indicated “very unimportant/very dissatisfied” and “5” indicated “very important/very satisfied.” By measuring both dimensions, the questionnaire effectively captures the gap between users’ cognitive expectations and experiential perceptions, providing a solid foundation for subsequent IPA. To ensure the reliability and validity of the scale, a small-scale pilot test was conducted prior to the formal survey to examine the clarity and comprehensibility of the items, and appropriate revisions were made based on the feedback. The specific questionnaire items and assigned values are provided in Appendix A. After data collection, the internal consistency of the scale was assessed using Cronbach’s α coefficient to ensure reliability. In addition, the Kaiser–Meyer–Olkin (KMO) measure and Bartlett’s test of sphericity were conducted to verify the suitability of the data for factor analysis, further ensuring the robustness of the research findings.

3.3. Research Sample and Data Collection

The survey participants were selected based on the criterion of having previously watched virtual anchor live streams. This requirement ensured that respondents possessed a certain level of practical experience and perception of virtual anchors, thereby enhancing the validity and representativeness of the study results. Before completing the questionnaire, respondents were shown live-streaming scenarios of virtual anchors, with examples randomly drawn from different mainstream e-commerce platforms, to more comprehensively reflect the performance of virtual anchors across diverse platform environments. Data were collected through offline surveys to ensure accuracy. The survey was conducted from 1 to 15 August 2025, with 350 questionnaires distributed and 342 returned. After screening, 309 valid responses were retained, resulting in an effective response rate of 90.35%. The sample included university students, white-collar workers, and community residents, exhibiting diversity in gender, age, and educational background, which helped ensure the generalizability and reliability of the research findings.

3.4. Data Analysis Method

To ensure the scientific rigor and robustness of the research findings, data analysis was conducted in three sequential steps: reliability and validity testing, paired-sample t-tests, and Importance–Performance Analysis (IPA) matrix modeling. First, Cronbach’s α coefficient and the KMO–Bartlett test were employed to verify the internal consistency and construct validity of the questionnaire scales, ensuring data quality and the suitability of the measurement instruments. Second, based on the confirmed reliability of the scales, the mean values of respondents’ importance (I) and performance (P) ratings for virtual anchor attributes were calculated. Paired-sample t-tests were then conducted to examine the significance of differences between expectations and perceptions, thereby identifying gaps between users’ psychological expectations and actual experiences. Finally, an IPA matrix was constructed by plotting I values on the horizontal axis and p values on the vertical axis, using overall mean values as reference points to define four quadrants. The comprehensive application of these methods not only ensures statistical rigor but also provides empirical support for the optimization and sustainable deployment of virtual anchors in e-commerce live streaming contexts.

4. Results and Discussion

4.1. Demographic Analysis

A total of 309 valid questionnaires were collected, and the demographic characteristics of the sample are presented in

Table 3. Sample Demographics (N = 309).

Features	Classification	Frequency	Percent
Gender	Male	132	42.7%
	Female	177	57.3%
Age	19 years or below	53	17.2%
	20–29 years	101	32.7%
	30–39 years	87	28.2%
	40 years or above	68	21.9%
Education	Junior high school degree	17	5.5%
	High school degree	41	13.3%
	Junior college degree	69	22.3%
	Bachelor’s degree	109	35.3%
	Master’s degree or above	73	23.6%

. Overall, the sample shows reasonable representativeness in terms of gender, age, and educational background, reflecting the current acceptance of virtual anchors among e-commerce live streaming users. Specifically, male respondents accounted for 42.7% of the total, with a gender ratio of approximately 2:3, which is consistent with previous findings that female users tend to dominate online shopping and live streaming participation. In terms of age distribution, respondents aged 20–39 constituted 60.9% of the sample, representing the core group of this study. This cohort typically possesses strong purchasing power and a higher level of acceptance toward emerging technologies. Regarding education, 58.9% of respondents held a bachelor’s degree or above, indicating a generally high educational level within the sample. This aligns with the hybrid nature of virtual anchors as both technological and content-driven entities: users with higher education levels often exhibit greater technological sensitivity and esthetic standards, leading to more critical expectations of both the technical and human-like attributes of virtual anchors. In summary, the sample demonstrates a well-balanced demographic composition, capturing the key characteristics of mainstream e-commerce live streaming users and providing a solid foundation for subsequent analyses of user expectations and perceptions of virtual anchors.

4.2. Reliability and Validity Tests

The collected data were analyzed using SPSS 21.0 to assess the reliability and validity of both the Importance (I) and Performance (P) dimensions. According to the results of Cronbach’s α coefficient, the α values for Importance (I) and Performance (P) were 0.928 and 0.825, respectively, both exceeding the threshold of 0.8, indicating excellent internal consistency. In addition, the KMO test values for the Importance (I) and Performance (P) indicators were 0.951 and 0.876, respectively, with Sig. = 0.000, indicating satisfactory validity for subsequent IPA. The results indicate that the data are suitable for conducting the IPA matrix analysis. Detailed information is presented in

Table 4. Reliability Analysis.

	Items	Total (n = 309)	Cronbach’s A
I value (Importance)	14	309	0.928
p value (Performance)	14	309	0.825

and

Table 5. Validity Analysis.

Construct	KMO Measure	Bartlett’s Test of Sphericity
Importance (I)	0.951	χ2 = 2265.513, df = 91, p < 0.001
Performance (P)	0.876	χ2 = 904.584, df = 91, p < 0.001

.

4.3. Common Method Bias Test

Common method bias (CMB) refers to the artificial covariance between predictor and criterion variables that arises from using the same data source or rater, the same measurement context, shared item context, or item characteristics [74]. Such artificially generated covariance among latent variables may lead to misleading research findings and conclusions. Therefore, it is necessary to test and control for this bias to ensure the reliability of the research findings. Following the recommendations of Zhou and Long [75], this study employed Harman’s single-factor test to assess common method bias. In the exploratory factor analysis, all items of the study variables were loaded onto a single factor. The results of the unrotated factor analysis showed that the first factor accounted for 34.716% of the variance, which is below the critical threshold of 40% [76], indicating that no significant common method bias was present in this study.

4.4. Analysis of I–P Values and Paired-Sample T-Test

To further examine whether there were significant differences between user expectations and actual perceptions of virtual anchor attributes in e-commerce live streaming, this study calculated the I–P (Importance–Performance) values for fourteen indicators and conducted paired-sample t-tests at a 95% confidence level. As shown in

Table 6. Results of Paired Sample t-Test Analysis.

Paired ID	Items	Mean	Standard Deviation	ΔM	t	p
1	Human-likeness (I)	4.29	1.05	0.50	7.190	0.000 **
	Human-likeness (P)	3.80	1.12
2	Cross-platformity (I)	4.50	1.01	0.76	10.427	0.000 **
	Cross-platformity (P)	3.74	1.25
3	Anthropomorphic appearance (I)	4.47	0.88	0.76	10.398	0.000 **
	Anthropomorphic appearance (P)	3.72	1.25
4	Interactivity (I)	4.41	0.93	0.88	11.204	0.000 **
	Interactivity (P)	3.53	1.27
5	Credibility (I)	4.31	1.07	0.55	7.542	0.000 **
	Credibility (P)	3.76	1.19
6	Behavioral agency (I)	4.34	1.04	0.80	10.297	0.000 **
	Behavioral agency (P)	3.54	1.21
7	Emotional bonding (I)	4.32	1.08	0.41	6.228	0.000 **
	Emotional bonding (P)	3.91	0.98
8	Entertainment (I)	4.21	0.93	0.69	9.155	0.000 **
	Entertainment (P)	3.53	1.19
9	Technical attributes (I)	4.45	0.96	0.73	10.014	0.000 **
	Technical attributes (P)	3.72	1.17
10	Novelty (I)	4.46	1.00	0.45	5.831	0.000 **
	Novelty (P)	4.01	1.03
11	Intelligence (I)	4.37	0.98	0.77	9.958	0.000 **
	Intelligence (P)	3.59	1.13
12	Spectacle (I)	4.45	0.98	0.71	9.005	0.000 **
	Spectacle (P)	3.75	1.23
13	Expressiveness (I)	4.60	0.74	1.07	13.885	0.000 **
	Expressiveness (P)	3.52	1.28
14	Professionalism (I)	4.53	0.69	0.82	12.017	0.000 **
	Professionalism (P)	3.72	1.16

** p < 0.01.

, the mean differences in all fourteen I–P pairs were greater than zero, indicating that the perceived mean values were consistently lower than the expected ones. Moreover, all differences reached a statistically significant level (p < 0.01), revealing a clear gap between user expectations and perceptions. These findings suggest that the current performance of virtual anchors still falls short of user expectations, implying substantial room for improvement in both technological capability and user experience.

Among the attributes, expressiveness (ΔM = 1.07) and interactivity (ΔM = 0.88) showed the largest I–P mean gaps, indicating that respondents perceived these two dimensions as requiring the most improvement. Existing research has shown that interactivity [77] and expressiveness [78] are core determinants of consumers’ sense of immersion and purchase intention in live-streaming commerce. Users’ dissatisfaction with these attributes suggests that current virtual anchor technologies have not yet fully replicated the linguistic fluency and emotional reciprocity of human anchors, thereby constraining users’ sense of engagement and trust in live-streaming environments. In contrast, emotional bonding (ΔM = 0.41) and novelty (ΔM = 0.45) exhibited relatively smaller I–P gaps, implying that virtual anchors have largely met user expectations in creating visual freshness and establishing an initial emotional atmosphere. As technology-driven innovations, their inherent novelty naturally captures user attention. However, this “technological freshness” may be short-lived; the long-term sustainability of user engagement will depend on how effectively virtual anchors can evolve from visual appeal toward deeper emotional interaction and companionship—key drivers for maintaining user loyalty and promoting a greener, digitally enabled labor model [10,79].

4.5. IPA Matrix Distribution and Discussion

Based on the IPA results, users’ expectations of virtual anchor attributes in e-commerce live streaming were plotted on the X-axis (Importance), and their perceptions were plotted on the Y-axis (Performance). Using the overall mean values (I = 4.41, P = 3.74) as the dividing thresholds, a two-dimensional four-quadrant coordinate chart was constructed to categorize the 14 attributes (Figure 2), with detailed numerical results presented in

Table 7. I–P Values of Virtual Anchor Attributes.

ID	Dimension	Importance	Performance
1	Human-like attributes	4.29	3.80
2	Cross-platformity	4.50	3.74
3	Anthropomorphic appearance	4.47	3.72
4	Interactivity	4.41	3.53
5	Credibility	4.31	3.76
6	Behavioral agency	4.34	3.54
7	Emotional bonding	4.32	3.91
8	Entertainment	4.21	3.53
9	Technical attributes	4.45	3.72
10	Novelty	4.60	4.01
11	Intelligence	4.37	3.59
12	Spectacle	4.45	3.75
13	Expressiveness	4.60	4.01
14	Professionalism	4.53	3.72

. This distribution clearly illustrates the strengths and weaknesses [60] of virtual anchors within e-commerce live-streaming contexts and provides an empirical foundation for strategic planning toward the intelligent and sustainable transformation of e-commerce.

4.5.1. Strength Maintenance Zone (Quadrant I)

The attributes of virtual anchors that fall into this quadrant include spectacle, novelty, and expressiveness. This indicates that the spectacle of “supernatural power” and the innovative appearance of virtual anchors are crucial components that have gained strong consumer recognition and satisfaction. This aligns with prior work showing that novelty and visually enhanced presentation increase early-stage attention and engagement in technologically mediated environments [48]. Our finding that expressiveness is highly evaluated also corresponds with evidence that clear, controllable expressive cues enhance message credibility and communication fluency [80]. The relatively stable and consistent expression style of virtual anchors resonates with literature noting that human anchors’ emotional variability may reduce credibility [81,82], whereas standardized digital delivery can strengthen perceived reliability [83]. However, existing studies emphasize that novelty-driven appeal diminishes without continuous content and interaction updates [84]. Thus, the empirical pattern observed here reinforces prior conclusions: these attributes are strategic strengths but require ongoing innovation to sustain their effect.

4.5.2. Sustainment Zone (Quadrant II)

In this study, the virtual anchors’ human-likeness, credibility, and emotional bonding fall into the “low-expectation but high-performance” quadrant. This aligns with the Expectation Disconfirmation Model [85], which holds that users respond positively when actual performance exceeds initial expectations. Existing research similarly notes that unexpectedly natural or authentic cues in digital agents enhance perceived trustworthiness [86,87]. That virtual anchors perform well on credibility is consistent with studies indicating that controlled scripting and design processes improve message consistency and reduce reputational risks relative to human anchors [88,89]. Emotional bonding observed in our results also accords with findings that parasocial interaction can emerge even with non-human agents when communication cues are coherent and human-like [90]. Overall, the empirical pattern corresponds closely with established theories and prior evidence, suggesting that these attributes should be maintained as stable perceptual advantages.

4.5.3. Low-Priority Improvement Zone (Quadrant III)

This quadrant includes three attributes of virtual anchors: entertainment, behavioral agency, and intelligence. All three exhibit low expectation and low perception, indicating that for the current sample group, these are not urgent priorities for improvement. This is consistent with the characteristics of our sample: consumers aged 20–40 who typically engage in goal-oriented, transactional live streaming rather than hedonic consumption [91]. Prior research similarly notes that in utilitarian shopping contexts, consumers prioritize informational usefulness [57] and interactivity that facilitates decision-making [92], whereas entertainment plays a secondary role [93]. This helps explain the low expectations observed for entertainment in our results. The low perceptions of intelligence and behavioral agency align with technological constraints identified in existing studies. These functions rely on advanced capabilities such as natural language processing, emotion recognition, and context-sensitive reasoning [94], each of which still faces limitations in fluency and affective accuracy [95]. Literature further suggests that partially anthropomorphized agents may generate distrust when behavioral cues lack authenticity or coherence [96]. Current technological trajectories indicate that such limitations may persist in the near term [97], To the extent that these attributes correspond with the modest user evaluations observed in our results, they should not be entirely disregarded. Although categorized as low-priority features, existing research notes that they may function synergistically with high-priority attributes such as interactivity and professionalism. A gradual and evidence-based enhancement strategy, implemented through small-scale pilot trials, may therefore help determine the most efficient ways to strengthen these dimensions without requiring substantial resource investment.

4.5.4. Key Improvement Zone (Quadrant IV)

Quadrant IV encompasses attributes such as technical attributes, interactivity, professionalism, and anthropomorphic appearance. These features exhibit a typical pattern of “high expectation but low perceived performance.” This pattern is consistent with studies noting that these dimensions are central drivers of user acceptance [48], purchasing decisions [53], and trust formation [54]. The low perception scores align with research indicating that current virtual-anchor systems still struggle with multi-turn dialog coherence, real-time emotional responses, and domain-specific expertise in high-pressure live-streaming scenarios [34,95,98]. These constraints reduce informational quality and weaken credibility. The mixed evaluations of anthropomorphic appearance correspond with findings that human-like cues generate positive effects only when behavioral signals match user expectations [99]. Inconsistencies between appearance and behavior can elicit discomfort and lower perceived authenticity [97,100]. Previous work also suggests that novelty-driven esthetics cannot compensate for deficiencies in interaction quality or professional depth. The Quadrant IV pattern reflects a clear gap between user expectations shaped by prior experiences with human anchors and the current technological and operational limits of virtual-anchor systems. Addressing this misalignment requires improvements in interaction design, knowledge structuring, and context-specific professionalism, rather than reliance on appearance-based differentiation.

4.5.5. Cross-Platformity: Boundary Attribute and Strategic Potential

Cross-platformity is positioned at the boundary of Quadrants I and IV, reflecting high user expectations but only moderate perceived performance. This aligns with interview insights indicating that, unlike human anchors, virtual anchors are not constrained by exclusive platform contracts [101], which theoretically enables broader dissemination and operational flexibility. Prior studies have similarly identified cross-platform presence as a strategic resource that enhances creator visibility and reduces platform dependency. However, the moderate perception level suggests that such potential is not yet fully achieved. Research shows that variations in platform architectures can lead to inconsistent interaction quality and user experience across environments [102], creating fragmented performance outcomes. At the same time, creators navigating heterogeneous platform ecosystems must accommodate divergent rules and incentive structures, requiring strategic trade-offs between reach and monetization [103,104]. This challenge is intensified by the increasing demands of “visibility labor” [105] and the resource burdens of content reproduction across multiple platforms. Accordingly, the expectation–perception gap for cross-platformity reflects not only technical constraints but also the joint shaping of operational choices by cultural, economic, and institutional factors [106]. This suggests that the strategic value of cross-platformity can be realized only when platforms, creators, and technical teams work collectively to reduce cross-platform frictions and enhance content adaptability.

5. Implications and Recommendations

5.1. Theoretical Implications

First, the expectation–perception discrepancies identified in this study enrich the theoretical understanding of virtual-anchor adoption by revealing which attributes users value the most for sustaining trust-based digital interactions—namely interactivity, professionalism, and technical reliability. These findings extend prior technology-acceptance and digital-labor literature by demonstrating that user acceptance of AI-mediated content is shaped not only by performance expectancy but also by the alignment between technical behavior and social interaction norms.

Second, the results highlight that sustainability in AI-driven e-commerce is not merely a matter of labor substitution but a function of interaction quality, cognitive trust, and systematically maintained consistency. This provides theoretical evidence that “digital labor sustainability” depends on reducing operational volatility, informational errors, and resource-intensive human interventions.

Third, cross-platformity emerges as a boundary attribute linking micro-level user cognition with macro-level platform governance. This expands existing discussions on platform ecosystems by showing how cross-platform performance influences user-perceived stability and the long-term viability of virtual anchors.

5.2. Practical Recommendations

In the short term, promoting the sustainable transformation of e-commerce requires prioritizing improvements in the high-expectation and low-performance attributes, particularly interactivity, professionalism, and technical reliability. These dimensions exert an immediate influence on user trust formation and purchase conversion. Platforms may strengthen operational performance by developing verified product knowledge repositories supported by retrieval-augmented generation modules, thereby reducing misinformation and enhancing response accuracy. Establishing human–AI hybrid service protocols further ensures that professional staff can intervene when automated systems encounter complex inquiries. Complementary monitoring tools that track key indicators such as response latency, correction frequency, and user satisfaction allow platforms to assess the effectiveness of optimization measures. Together, these actions improve efficiency, reduce customer-service burdens, and contribute directly to operational sustainability.

In the medium term, the focus should shift to reinforcing relational sustainability by advancing human–AI collaborative mechanisms and enhancing the behavioral coherence of virtual anchors across different platforms. Improvements in affective interaction quality, including more refined emotion-recognition cues and optimized dialog-management strategies, can strengthen user engagement. Enhancing performance-level elements such as lip synchronization and facial-expression accuracy reduces the cognitive dissonance associated with the uncanny valley effect and promotes more seamless interaction experiences. To enable consistent performance across multiple platforms, virtual-anchor assets, including persona configurations, knowledge components, and core dialog scripts, should be modularized to reduce duplication costs and facilitate adaptation. In parallel, the development of compatibility standards at the platform or infrastructure level can minimize experience discontinuities. These measures strengthen long-term user trust and reduce the cumulative resource demands associated with virtual anchor operation.

In the long term, emphasis should be placed on constructing an integrated ecosystem that positions virtual anchors as enduring digital assets rather than temporary marketing instruments. Achieving this objective requires sustained investment in domain-specific knowledge graphs and explainable dialog strategies that support the development of professional, traceable, and transferable competencies. At the institutional level, establishing cross-platform governance mechanisms and unified content-regulation frameworks is essential for ensuring identity portability and reducing operational fragmentation. The advancement of ethical and data-governance systems, including transparent disclosure of virtual-identity status and rigorous protection of user interaction data, is also necessary to prevent reputational and regulatory risks. By embedding virtual anchors within a broader operational ecology that spans customer service, brand building, and after-sales support, platforms can mitigate repetitive human labor, enhance system resilience, and advance toward a low-carbon, scalable, and sustainable e-commerce environment.

6. Conclusions

This study focuses on virtual anchors in the context of e-commerce live streaming and integrates literature review, expert interviews, questionnaire surveys, and Importance–Performance Analysis (IPA) to systematically measure and compare users’ expectations (Importance) and perceptions (Performance) (sample size: N = 309). The results reveal that users’ expectations are significantly higher than their perceptions (overall mean I = 4.41, P = 3.74; all 14 I–P differences are positive and statistically significant), indicating that virtual anchors still have substantial room for improvement in both technological implementation and user experience.

This study enhances the theoretical understanding of virtual humans and the live-streaming economy by empirically identifying cross-platformity and spectacle as meaningful evaluative dimensions within the virtual-anchor attribute system. Their inclusion helps refine and broaden existing frameworks in a user-centered manner. Furthermore, the study proposes a three-stage implementation pathway to address the “anchor dilemma [17]” that may hinder the sustainable and intelligent transformation of e-commerce. The findings reveal that only by achieving breakthroughs in core competencies can virtual anchors transform short-term novelty-driven attention into sustained sales performance and brand equity, thereby fully realizing their strategic value in reducing human labor dependence, improving operational efficiency, and optimizing the e-commerce ecosystem. Conversely, without progress in these key areas, the long-term sustainability of virtual anchors will remain limited. Therefore, by focusing on short-term optimization of critical interaction points, mid-term construction of human–AI collaboration and platform adaptability, and long-term establishment of governance and commercialization ecosystems, virtual anchors can evolve into engaging and reliable digital assets that continuously contribute to the intelligent and sustainable development of e-commerce.

This study has several limitations that merit careful consideration. First, the use of cross-sectional and self-reported survey data restricts the ability to infer behavioral outcomes, and potential memory or subjective biases may influence the accuracy of user evaluations. The expectation–perception gaps identified here, therefore, should not be interpreted as predictors of actual purchasing behavior, dwell time, or conversion rates. Second, although the sample includes respondents from multiple occupational groups, its demographic and geographical coverage remains limited. User evaluations of virtual anchors may vary across regions, platforms, age brackets, and product categories, and the present study was not able to systematically investigate such heterogeneity. Third, the use of Importance–Performance Analysis provides descriptive diagnostic insights but does not test causal mechanisms or subgroup stability; future research could incorporate structural equation modeling, longitudinal designs, or multi-group analysis to explore the determinants and consequences of attribute performance more rigorously. Fourth, the study focuses on the attributes of virtual anchors themselves and does not incorporate platform algorithms, business models, or governance environments, all of which may meaningfully shape user perceptions and the sustainability implications of virtual-anchor deployment. Finally, the rapid evolution of generative AI, multimodal interaction technologies, and virtual human production pipelines implies that the attributes and user expectations examined in this study are time-sensitive. Continuous updates and replicated studies will be required to track how technological advances reshape user evaluations, operational practices, and the broader sustainability implications of virtual anchors.