How Can AI Virtual Streamers Gain Consumer Trust to Influence Purchase Intention in Live-Streaming E-Commerce?

Abstract

Within the domain of live-streaming e-commerce, the application of artificial intelligence (AI) technologies presents novel growth opportunities. Live-streaming sales by AI streamers have attracted widespread attention as a new form. To promote the application of AI streamers in live-streaming e-commerce, this paper mainly addresses the following question: How do AI streamers in live-streaming e-commerce influence consumer trust and purchase intention? Based on this, guided by the stimulus–organism–response (SOR) theory, we construct a research model to study the impact of AI streamers’ characteristics and scenario fit on purchase intention through trust mediation. Subsequently, 410 valid samples of consumers who watch AI streamers promoting products are collected through an online questionnaire. Structural equation modeling (SEM) and fuzzy-set qualitative comparative analysis (fsQCA) are integrated to investigate both the influence mechanisms and configurational pathways affecting purchase intention. This study reveals that AI streamers’ image characteristics (cuteness, vitality) and scenario fit are positively associated with purchase intention. However, AI streamers’ ability characteristics (professionalism, responsiveness) are not positively associated with purchase intention. In addition, consumer trust partially mediates the relationship between these key factors and purchase intention. Further, consumer innovativeness significantly negatively moderates the effect of AI streamers’ characteristics and scenario fit on consumer trust. Moreover, the influence mechanisms diverge substantially between incremental and breakthrough products.

1. Introduction

With the rapid development of the e-commerce industry, live-streaming e-commerce, as a new business model in the digital retail sector, has become an important shopping channel for consumers [1]. Currently, the global live-streaming e-commerce industry is in a stage of rapid growth. It not only forms a huge market size but also demonstrates strong market vitality. According to a report released by RetailX, the e-commerce revenues in Asia, North America, and Africa amounted to approximately USD 4.2 trillion, USD 1 trillion, and USD 55 billion, respectively, in 2024 (see: https://www.ennews.com/article-30184-1.html (accessed on 8 July 2025)). In March 2025, the U.S. TikTok Shop’s merchandise volume reached USD $1.044 billion, with average daily sales of USD 33.69 million (see: https://www.ingstart.com/blog/17510.html (accessed on 8 July 2025)). As the core link connecting consumers and products, streamers have become the key factor determining the success of live-streaming e-commerce.

AI-driven technological advances reconfigure live-streaming commerce paradigms while generating novel opportunities. AI streamers (or virtual streamers) are new digital agent entities built with artificial intelligence for live-streaming e-commerce. These agents autonomously perform complex streaming tasks comparable to human streamers [2]. Compared to human streamers, the AI-driven sales model offers notable advantages in both product promotion and risk management. On the one hand, merchants leverage AI streamers for 24/7 live broadcasting. Continuous operation elevates store rankings within platform algorithms and expands potential customer reach [3]. On the other hand, AI-driven live-streaming mitigates reputation damage risks from streamers’ personal issues, which reduces customer loss for stores. These advantages accelerate AI streamers’ adoption in live-stream marketing, which demonstrates great market appeal and sales potential. According to industry forecasts, the virtual streaming market is expected to reach USD 1246.8 billion by 2032 (see: https://www.wiseguyreports.com/cn/reports/virtual-live-market (accessed on 8 July 2025)).

In business practice, many well-known brands have incorporated AI streamers into their product marketing models. In April 2024, the JD.com platform initially employed the “Cai Xiao Dongge” AI digital human for live-streaming e-commerce activities. As a result, the order volume in a single live-streaming session increased by 7.6 times. Subsequently, in May 2025, the “Luo Yonghao” digital streamer conducted sales in Baidu’s live-streaming e-commerce channel. This event generated CNY 5.0 million in GMV within four hours (see: https://www.qbitai.com/2025/06/293805.html (accessed on 8 July 2025)). Although AI streamers have performed well in live-streaming sales, their market penetration rate remains below 1% (see: https://www.ebrun.com/report/529959.shtml (accessed on 8 July 2025)). Moreover, some brands have discontinued the use of AI streamers in e-commerce live-streaming due to their failure to meet the expected performance goals. For example, L’Oréal, Philips, and Estée Lauder discontinued AI streamer deployment after initial trials and reverted to human streamers. The conversion of virtual streamer attraction into purchasing behavior presents a key issue that restricts the widespread adoption of AI streamers in live-stream commerce. Consequently, delving deeply into the impact mechanism of AI streamers on purchase intention is critical for optimizing their practical deployment in live-streaming e-commerce.

In the field of live-streaming e-commerce, the rise of AI streamers promoting products is reshaping the traditional “people, goods, and scenarios” logic [3,4,5]. AI streamers in live-streaming e-commerce display diverse characteristics. Scholars classify the diverse characteristics of AI streamers and explore how these characteristics influence consumers, such as anthropomorphism, attractiveness, and credibility [6,7,8]. During product promotions, the image characteristics and interactive communication of AI streamers influence consumers’ purchasing intentions. AI streamers and virtual endorsers share visual similarities that mainly manifest in hyper-realistic and anime-like styles. Different styles vary in cuteness, and compared to other characteristics, cuteness is also a key factor influencing the endorsement effect of virtual endorsers. The professional recommendations that human streamers provide strengthen consumer trust and purchase intention [9]. Vitality and responsiveness have been incorporated by scholars into the category of AI streamers’ characteristics and confirmed to have a positive impact on purchase intention [2]. Therefore, this paper examines four characteristics of AI streamers from image characteristics and ability characteristics. Cuteness measures the AI streamer’s visual appeal and popularity. Vitality assesses how authentic and natural the appearance seems. Professionalism assesses the AI streamer’s expertise and performance during product promotions. Responsiveness evaluates how quickly and efficiently the streamer communicates with users.

Furthermore, scholars have identified core mechanisms underpinning AI streamers’ effectiveness, including empathy, perceived satisfaction, flow experience, and technological affordances [10,11]. These mechanisms derive intrinsically from AI streamers’ social and technical attributes. However, few studies have focused on the significant role of the trust relationship between AI streamers and consumers in influencing purchase intention. In live-streaming e-commerce, consumer trust serves as a foundation for purchase decisions, where it increases purchase intention by elevating product credibility and mitigating risk perceptions [3,12]. Moreover, consumer heterogeneity and product differentiation are key factors influencing the effectiveness of digital marketing. Previous studies have overlooked the impact of consumers’ innovative characteristics and product innovation types. In this paper, we aim to fill this research gap. Specifically, our study seeks to address the following questions:

(1)

How do the characteristics and scenario fit of AI streamers affect purchase intention?

(2)

How do consumers perceive and trust different types of AI streamers and live-streaming scenarios?

(3)

How do consumers’ innovativeness and product types moderate the differences in consumer trust caused by AI streamers’ characteristics and live-streaming scenarios?

To answer the above questions, this study adopts the SOR (stimulus–organism–response) model to investigate how AI streamers’ characteristics and live-streaming scenarios impact purchase intention. This study contributes in four ways: First, existing research on AI streamers centers on streamer characteristics and product attributes [3,4,5]. However, live-streaming e-commerce evidence shows that streamers, products, and scenario fit critically affect purchase intention. Consequently, this study introduces the variable of live-streaming scenarios into the research framework. Second, current research emphasizes social attributes and functional capabilities (e.g., empathy, perceived closeness, satisfaction, and social presence) in AI live-streaming e-commerce [13,14] but neglects consumer trust. However, studies on human streamers confirm trust’s importance [1,3,12]. This study therefore examines how AI streamers’ characteristics and scenarios influence purchase intention through consumer trust. Third, in live-streaming commerce, product differentiation and consumer innovativeness significantly influence consumer trust. Hence, this study explores how product consumer innovativeness moderates the impact of AI streamers on consumer trust. Moreover, the type of product innovation has a significant difference in the impact of AI streamers on consumer trust. Fourth, this study combines SEM with fsQCA to validate the model’s logic and identify the multiple pathways.

The remainder of this paper is structured as follows. Section 2 outlines the theoretical background and literature review. Section 3 establishes the research model and puts forward the hypotheses. Section 4 provides the descriptive statistics, research design and data collection procedures. Section 5 reports the variable measurements and empirical results. Section 6 discusses the results and suggests directions for future research.

2. Theoretical Background and Literature Review

2.1. Stimulus–Organism–Response Framework

The stimulus–organism–response (SOR) framework is one of the most influential frameworks in environmental psychology. It dissects how external environmental factors systematically influence consumers, thereby triggering a series of complex and specific psychological mechanism changes and behavioral responses [15]. Eroglu et al. [16] first applied the SOR theory to the study of online shopping environments. Within this framework, stimuli (S) refer to the existing external environmental factors that elicit consumers’ purchase decisions. Organism (O) reflects consumers’ internal perceptions and experiential states regarding the product. Response (R) denotes the behavioral reactions exhibited by consumers after being exposed to external stimuli. With the rapid development of live-streaming e-commerce, the SOR model has been widely used in the study of consumer purchase behavior in live-streaming scenarios [17,18,19,20].

The rise of AI streamers promoting products is reshaping the logic of “people, goods, and scenarios” in traditional marketing. Firstly, AI streamers conduct live-streaming sales in the form of virtual digital humans and generate differentiated characteristics through algorithms to match the aesthetic preferences of different consumers. Secondly, AI technology breaks through the traditional boundaries of product selection in live-streaming rooms. AI streamers can dynamically adjust product recommendation strategies based on user behavior data analysis. Thirdly, the live-streaming room of AI streamers can quickly switch between different theme scenarios, thereby providing users with an immersive experience of product usage scenarios. Based on the SOR theory, scholars have regarded the characteristics of human streamers, interaction quality, and the live-streaming environment as new stimulus variables and have explored their impact on purchase intention from different perspectives. Yang et al. [21] examine the impact of various factors in e-commerce live-streaming, including streamers’ characteristics, product, domain, trust, and impulsivity on purchase intention. They utilize SEM, confirmatory factor analysis, and hypothesis testing to evaluate the relationships. Liu et al. [22] explore the influence of tour guide streamers’ characteristics (interactivity, authenticity, and entertainment) on purchase intention from the perspective of consumer immersive experience and trust. Kang et al. [23] investigate the effect of interaction quality in live-streaming e-commerce on purchase intention and find that relational strength mediates the relationship between interactivity and purchase intention. Li et al. [5] examine the impact of different live-streaming environment characteristics on purchase intention from the perspective of presence and perceived trust.

By integrating advanced perception and cognitive intelligence technologies, AI streamers can display realistic facial expressions and gestures as well as natural interactive responses [24]. With human-like appearance, voice, expressions, and movement characteristics, AI streamers can enable consumers to have more realistic and in-depth interactions with them in live-streaming [25]. Therefore, the emergence of AI streamers has reshaped the existing ways of connection and social organizational forms. In live-streaming sales, AI streamers become social participants, responsible for providing thinking, decision-making, and emotional expression [26]. Meanwhile, the research on human–computer interaction has shifted from a rationalist perspective to an emotional and cognitive-driven interaction experience perspective [27,28,29]. Trust is a multi-dimensional latent variable that directly affects the practical application of artificial intelligence products. People’s perception of anthropomorphic characteristics, effectiveness, and practicality of non-human entities will further influence their emotional responses and behavioral choices [30]. Personification refers to the psychological process of attributing human entities (e.g., characteristics, motives, intentions, or mental states) to non-human entities, used to describe the similarity between AI and humans [31]. This process not only constitutes a core dimension of AI’s human-like characteristics but also becomes a key variable influencing the formation of trust in human–machine interaction [32]. In AI streamers’ live-streaming, trust becomes the core factor enabling effective interaction between consumers and AI streamers. Scholars have conducted an exploratory study on the trust mechanism in live-streaming e-commerce [33,34,35].

Based on the above research, this paper considers AI streamers’ characteristics, product types, and live-streaming scenarios as external stimuli. Consumers’ trust is regarded as an organism with internal cognition, and their purchase intentions are viewed as a behavioral response under this psychological reaction. Based on the SOR framework, this paper conducts an in-depth analysis of how AI streamers’ characteristics, product types, and live-streaming scenarios influence purchase intention.

2.2. AI Streamer Marketing

The most salient characteristics of e-commerce live-streaming are the visual scenarios and real-time interaction. The live-streaming streamer is the core of interaction and also a key factor influencing consumer purchase intention [14]. In addition, scholars have also studied the impact of different market cultures and market contexts on the impact of streamers on consumer trust and purchase intention [36,37]. Extensive research has focused on the impact of live-streaming by human streamers on consumer purchase intention [38,39,40]. With the development of artificial intelligence technology, AI streamers have received extensive attention from scholars. Scholars have conducted in-depth research on the current development status, functional characteristics, and application prospects of AI streamers [41,42,43]. Moreover, some scholars study the differences between AI streamers with human streamers or KOLs [12,44]. The intelligence level of AI streamers is not high enough, and there are still significant issues in rich and natural emotional expression [45]. Therefore, compared with the live-streaming scenarios of AI streamers, consumers are more willing to choose products recommended by human streamers [40]. To enhance the intelligence level of AI streamers, some scholars have conducted research on aspects such as role generation, content production, and visual presentation of AI streamers [41]. Furthermore, some scholars have proposed driving AI streamers through AI–human collaboration to overcome the shortcomings of AI streamers in live-streaming sales and meet consumers’ demands [46,47,48].

To further promote the application of AI streamers in live-streaming e-commerce, some scholars have studied the impact of streamer characteristics, product types, and live-streaming scenarios on purchase intention [3,4,5]. Gao et al. [2] explore the impact of three key characteristics on social presence and purchase intention. They indicate that there are differences in the influence of human and animal virtual streamers on social presence and presence. Sun and Tang [49] verify the impact of the verisimilitude of AI streamers on customer purchase intention. Zhu et al. [50] investigate the live-streaming product sales of virtual streamers and their impact on brand image and brand loyalty. Furthermore, live-streaming sales by AI streamers possess distinct technical characteristics as well as social attributes. Scholars have focused on exploring the influence mechanism of AI streamers on consumers and purchase intentions by centering on theories such as psychological contract theory, persuasion theory, social presence theory, and social identity theory [10,13,14]. The influence mechanism of AI streamers on consumer behavior is shown in Table 1.

Table 1. Influence mechanisms of AI streamers on consumer behavior in live-streaming e-commerce.

Source	Antecedents	Outcome	Perspective	Moderator	Method
Gao et al. [2]	AI streamers’ characteristics	Purchase intention	Social presence and telepresence	Streamer image	SEM
Wu et al. [3]	AI streamers’ socialness	Experiential value	Social presence	Communication style and situation	SEM
Li et al. [5]	AI streamers’ characteristics	Impulse buying	Emotional responses	Viewer engagement	SEM-ANN
Chen and Li [10]	Negative expectation violations	Discontinuance behavior	Psychological states	N.A.	SEM
Qin and Liu [13]	Consumer perceptions	Purchase intention	Psychological contract	N.A.	SEM
Zhang et al. [14]	Persuasive cues	Parasocial interaction and impulse buying	Arousal	Mindset	SEM
Sun and Tang [49]	Form realism and behavioral realism	Purchase intention	Parasocial interaction	Relationship norm orientation	ANOVA
Zhu et al. [50]	AI streamers’ characteristics and perceived value	Brand loyalty	Brand image	N.A.	SEM-ANN
Our paper	AI streamers’ characteristics and scenarios	Purchase intention	Consumer trust	Consumer innovativeness and product type	SEM-fsQCA

Existing research has mainly focused on the influence of the streamer, product, and scenario on purchase intention in the context of live-streaming sales by human streamers [21,22,23]. In the context of live-streaming sales by AI streamers, scholars have mainly concentrated on comparative studies between AI streamers and human streamers [12,43,45], as well as the application methods of AI streamers in live-streaming sales [41,42,43]. Research on the influence mechanism of AI streamers on purchase intention is relatively scarce. Moreover, most scholars have studied the impact of AI streamers’ technical characteristics and social attributes, live interaction, and product types on purchase intention from the perspective of AI streamers’ characteristics [3,4,5]. However, existing research has overlooked the significant influence of consumer trust on purchase intention in the context of live-streaming sales by AI streamers. Based on the above research, this paper takes AI streamers’ characteristics and the fit between the scenario and the streamer as external stimuli and regards consumer trust as an organism with internal cognition, with purchase intention being the behavioral response under this psychological reaction. Based on the SOR framework, this paper deeply analyzes the influence of AI streamers’ characteristics and the fit between the scenario and the streamer on purchase intention.

3. Research Model and Hypotheses

3.1. Research Model

Interpreting the mechanism through which AI streamers’ characteristics and scenario fit drive consumer purchase intentions, we have developed a research model. Based on the SOR theory and the background of this study, the research model is shown in Figure 1. In the context of live-streaming sales, the individual image and interactive communication of AI streamers can attract consumers and influence their purchasing intentions. We attempt to study AI streamers’ image and ability characteristics (cuteness, vitality, professionalism, responsiveness) by combining and proposing a higher-order construct and then investigate their impact on consumer trust and, finally, on purchase intention. Moreover, we study the impact of scenario fit on consumer trust and purchase intention.

Figure 1. Research model.

Specifically, AI streamers’ characteristics and scenario fit are proposed as two dimensions as “stimulus”, which influence consumer trust (organism). Subsequently, purchase intention is the “response” of consumers in live-streaming e-commerce. By highlighting the dynamic relationship between AI streamers and consumer trust, a theoretical explanation is developed for the mediating role of consumer trust in the pathways. Consumer innovativeness is regarded as a moderating variable. In addition, these relationships also vary among different types of innovative products (incremental and disruptive products). To address potential alternative factors, we control gender, age, education level, and personal income per month, typically incorporated in studies on consumer behaviors.

3.2. AI Streamers’ Image and Ability Characteristics as Second-Order Constructs

The image and ability characteristics of AI streamers are both regarded as second-order reflective formation constructs with first-order reflective measurement constructs (e.g., cuteness, vitality, professionalism, and responsiveness). Higher-order constructs offer two key advantages. Their integration of lower-order constructs reduces hypothesized relationships in the model, which enhances parsimony [51]. Additionally, this approach alleviates multicollinearity concerns and improves result interpretability [52].

According to Jarvis et al. [53], the configuration of a measurement model is reflective or formative. This depends on three criteria: the direction of the causal relationship, the replicability of items, and covariation among the items. The image and ability characteristics of AI streamers are referred to as the second-order constructs, which are measured by four first-order constructs (referred to as the dimensions of the image and ability characteristics of AI streamers, respectively; cuteness, vitality, professionalism, and responsiveness), each with its own set of items. All four dimensions have different conceptual meanings, which are reflected in their definitions and measurement standards. Therefore, all four structures are referred to as reflective first-order constructs. First-order reflective or formative constructs employ their own measurement items. Higher-order constructs use first-order components through either a repeated indicator or a two-stage approach [52].

The proposal to take “the image and ability characteristics of AI streamers” as high-order constructs is mainly based on three reasons. First, many studies on AI streamers in live-streaming e-commerce have widely adopted these characteristics to explore their impact on perceived value, flow experience, technological affordance, and purchase intention [8,10,24]. Second, many previous studies have proposed and studied the concept of “AI streamers’ characteristics” as a multi-dimensional structure based on the interactivity and sociability of AI streamers [3]. Research on AI streamers’ characteristics has advanced significantly in recent years. Current studies now examine multi-dimensional characteristics including anthropomorphism, attractiveness, vitality, professionalism, and responsiveness [6,8,24]. Finally, from a theoretical perspective, consumers’ perception of AI streamers’ characteristics in live-streaming e-commerce mainly focuses on two dimensions: image and ability [54]. Cuteness and vitality constitute the image dimension, while professionalism and responsiveness form the capability dimension. These two dimensions are interrelated and jointly shape consumers’ cognition of AI streamers, thereby influencing their emotional responses and behavioral decisions. Therefore, this paper classifies the characteristics of AI streamers from the dimensions of image and ability and constructs high-level structures.

3.3. Hypotheses Development

3.3.1. The Effect of AI Streamers on Purchase Intention

In the context of live-streaming e-commerce, the streamer conveys information to consumers and guides their purchases through visual presentation and real-time interaction. This role positions the streamer as a central figure in sculpting the emotional and cognitive responses of consumers during live-streaming sessions. These mediated reactions subsequently drive online purchase intention [55]. Interactive and social characteristics of AI streamers shape consumer social presence experiences and trust, which strengthens perceived product value and purchase decisions. Scholars subsequently delineated AI streamers’ characteristics, including friendliness, attractiveness, responsiveness, professionalism, vitality, and intelligence. These characteristics have been empirically shown to satisfy consumers and stimulate purchase intention [56]. The characteristics of AI streamers have two dimensions: image and ability. Based on this, we propose that the image and ability characteristics of AI streamers form a higher-order formative structure, consisting of four factors: cuteness, vitality, professionalism, and responsiveness.

The cuteness of AI streamers refers to their affinity and popularity. Research shows that consumers consider cute devices both more humanized and practical [57]. Specifically, the cute designs in AI streamers improve perceived usefulness and ease of use, which in turn positively influences purchase intention [58]. Furthermore, cute AI streamers evoke stronger consumer closeness and immersion. The high level of presence minimizes psychological distance toward both streamers and products, which facilitates purchase decisions [59]. Crucially, consumers exhibit a preference for visually appealing AI streamers during advertising. These streamers provide post-work psychological restoration while fulfilling emotional needs, thereby strengthening positive brand impressions.

Vitality measures the naturalness and realism of the AI streamer’s image, reflecting its degree of resemblance to a real human [57]. Research indicates that individuals form initial impressions largely based on visual and vocal cues [60], and these first impressions significantly shape overall evaluations [61]. A vibrant AI streamer enhances the perceived sense of realism, which encourages consumers to anticipate deeper relational connections and experience more positive affective responses [59]. Furthermore, higher vitality contributes to more natural and authentic interactions, leading to more favorable consumer attitudes and increased purchase intention toward recommended products.

Professionalism critically determines the accuracy and authority of AI streamers’ product information. Research demonstrates that highly professional streamers not only enhance viewer favorability but also bolster consumer trust, which directly drives purchase intention [62]. The performance of streamers in interactive sections such as product display and question answering significantly influences purchase intention [1]. Through precise product introductions based on scripted expertise, AI streamers reduce perceived trust risks and strengthen purchase intention.

Responsiveness denotes AI streamers’ timeliness and efficacy in addressing consumer inquiries [59]. The capacity to offer real-time replies, maintain coherent dialogues, and provide thorough product information constitutes a key mechanism for building consumer trust, which significantly elevates purchase intention [5]. Timely feedback reduces psychological distance while improving service quality, ultimately supporting user goal attainment [35]. Importantly, enhanced responsiveness reinforces perceived agency, interaction intensity, closeness, and usefulness, thereby strengthening purchasing behavior.

Scenario fit denotes the congruence level between live-streaming settings and product characteristics [63]. Product-customized scenario designs significantly enhance consumer experiences, as this fit demonstrates compatibility, adaptability, or functional capability. High streamer–product matching substantially elevates purchase intention. According to fluency theory, the coherence of resources improves processing fluency, which in turn fosters positive brand attitudes and deeper cognitive engagement. This enhanced fluency strengthens brand–consumer congruence and ultimately shapes purchasing behavior [64]. Furthermore, greater brand–content congruence in live-streaming stimulates positive affect, improves customer retention, and enhances the effectiveness of brand communication.

Trust represents consumers’ confidence in the quality and reliability of products or services offered by merchants [65]. This construct critically shapes online user judgments and behaviors. In traditional e-commerce, product evaluation relies primarily on text descriptions and static images, which restricts physical inspection and creates substantial information asymmetry and valuation difficulty. Live-streaming e-commerce reduces consumption uncertainty through real-time product demonstration, thereby enhancing seller trust [12]. Given the inherent transaction uncertainty in online shopping, trust remains a fundamental determinant of digital purchasing behavior [66]. Consequently, the following hypotheses are developed:

H1a. 

AI streamers’ image characteristics are positively associated with purchase intention.

H1b. 

AI streamers’ ability characteristics are positively associated with influence purchase intention.

H1c. 

AI streamers’ scenario fit is positively associated with purchase intention.

H1d. 

Consumer trust positively is positively associated with purchase intention.

3.3.2. Mediating Role of Consumer Trust

Trust constitutes a fundamental affective foundation for purchase intention formation in live-streaming commerce. As a pivotal mediator in consumer behavior research, its theoretical significance is robustly supported by empirical evidence [67,68,69]. Within AI-streamed contexts, AI streamers’ characteristics serve as primary stimuli for consumer perception. These characteristics enhance emotional states through informational transmission and affective contagion, elevating trust and ultimately boosting purchase intention. Specifically, heightened cuteness in AI streamers generates a more dynamic atmosphere, which enhances their appeal and elicits both pleasure and satisfaction among viewers. [70,71]. Attractive service scenarios potentially trigger halo effects, intensifying product responses and service personnel trust [72]. Streamer vitality reflects behavioral naturalness and personalization. Elevated vitality enables authentic product promotion perception, facilitating brand image comprehension and identification [73]. Technologically enabled professionalism and responsiveness facilitate effective consumer interactions. Highly professional AI streamers deliver superior purchase recommendations through comprehensive product knowledge, feature expertise, and demonstration proficiency. Such competence reduces purchase hesitancy while establishing durable trust [74]. Furthermore, entertaining, professional, and responsive characteristics enhance streamer appeal, which strengthens consumer trust and, in turn, purchase intention.

Live-streaming e-commerce demonstrates a positive relationship between scenario fit and perceived trust [75]. In the context of AI streamers promoting products, AI streamers can provide consumers with more fitting live-streaming scenarios. Product-relevant scenario arrangements enhance product feature comprehension through heightened fit. Conversely, brand-irrelevant settings distract consumer attention and trigger negative reactions by providing extraneous information. Adaptive scenario designs reduce cognitive distance via simulated consumption environments. Compared to static website displays, immersive scenarios facilitate realistic product interaction, enriching consumption experiences. Consequently, consumer trust mediates the impact of AI streamer characteristics and scenario fit on purchase intention. This mediation mechanism strengthens consumer–product connections within live-streaming contexts, elevates trust, and amplifies purchase intention. Based on these findings, the following hypotheses are developed:

H2a. 

Consumer trust has a significant mediating effect between AI streamers’ image characteristics and purchase intention.

H2b. 

Consumer trust has a significant mediating effect between AI streamers’ ability scenario fit and purchase intention.

H2c. 

Consumer trust has a significant mediating effect between AI streamers’ scenario fit and purchase intention.

3.3.3. Moderating Role of Consumer Innovativeness

Consumer innovativeness refers to the willingness to adopt novel concepts, manifesting as early adoption behavior within social groups to try new experiences [76]. In digital marketing, the role of consumer innovativeness is becoming increasingly significant [77]. Innovative consumers display marked tendencies to seek novelty. High-innovativeness consumers maintain elevated expectations toward new technologies. Disruptive experience unmet by AI streamers generates profound disappointment. Consequently, AI streamers that emphasize superficial promotion over precise recommendations lack substantive dialogue. This approach fails to maintain consumer engagement, which subsequently erodes trust. Moreover, technically sophisticated consumers with high innovativeness show greater awareness of algorithmic limitations. AI streamers possess appealing, natural, and highly anthropomorphic designs. They deliver professional product presentations and provide prompt responses that suit live-streaming environments. However, consumers still perceive interactions as unnatural and detect behavioral inconsistencies, which diminishes their trust. Conversely, low-innovativeness consumers exhibit lower technology expectations. These consumers actively accept innovations in AI live-streaming technology despite imperfections, which enhances consumer trust and motivates purchase intention. Consequently, the following hypotheses are developed:

H3a. 

Consumer innovativeness negatively moderates the impact of AI streamers’ image characteristics on consumer trust.

H3b. 

Consumer innovativeness negatively moderates the impact of AI streamers’ ability characteristics on consumer trust.

H3c. 

Consumer innovativeness negatively moderates the impact of AI streamers’ scenario fit on consumer trust.

3.3.4. Moderating Role of Product Type

Brand endorsements have significantly different effects on the promotional activities of different types of products [78,79]. Products offered by e-commerce platforms are categorized into incremental products and disruptive products based on the degree of product innovation [80]. Incremental products refer to continuous improvements and optimizations made on the basis of existing products. Incremental products typically do not rely on new technologies but enhance performance, functionality, or user experience through minor improvements. Since such optimizations build on existing products, they yield higher market acceptance and lower risks. In contrast, disruptive innovations establish entirely new technologies, business models, or design concepts that redefine markets. Due to their novel concepts, user acceptance and market feedback are subject to greater uncertainty. In business practice, disruptive and incremental products differ significantly in their innovation paths, which creates distinct consumer perception trade-offs. Disruptive versions solve intractable problems with unprecedented functions yet impose relearning demands that increase cognitive load. Incremental types build on existing products with lower thresholds that minimize onboarding, though constrained enhancements seldom achieve transformative value. AI live-streaming functions as an innovative marketing method which prioritizes product information transmission above experiential immersion. The purpose of brand endorsements through AI streamers tends to be the promotion of the product’s “novelty” image, with a more pronounced impact on high-innovation disruptive products. Based on previous research [81], this study defines a disruptive product as the basic version and the incremental product as the PLUS version of a brand’s fitness bracelet. Consequently, the following hypotheses are developed:

H4a. 

For incremental products and disruptive products, the effects of AI streamers’ image characteristics on consumer trust and purchase intention differ.

H4b. 

For incremental products and disruptive products, the effects of AI streamers’ ability characteristics on consumer trust and purchase intention differ.

H4c. 

For incremental products and disruptive products, the effects of AI streamers’ scenario fit on consumer trust and purchase intention differ.

4. Methods

4.1. Questionnaire and Measures

This study collected empirical data through a structured questionnaire comprising three sections. The first part assesses subjects’ comprehension of AI streamers and live-streaming scenarios after viewing product promotion videos. The second part measures core variables using validated scales. The third part collects audience demographics including gender, age educational level, and personal income per month.

Some measures were slightly modified in order to suit the live-streaming e-commerce context. We used a seven-point Likert scale ranging from (l) strongly disagree to (7) strongly agree to measure all of these items. The constructs in the questionnaire were specifically adapted from the following sources: the AI streamers’ cuteness (CU) and vitality (VI) were adapted from Gao et al. [59]; professionalism (PR) was derived from Trepte and Scherer [82] and Ohanian [83]; responsiveness (RE) was sourced from Zhang et al. [35]; scenario fit (SF) came from Gorn [84]; consumer trust (CT) came from McKnight et al. [85]; and purchase intention (PI) came from Bansal and Voyer [74] and Hong et al. [86]. All the items for measuring the constructs are shown in Appendix A.

4.2. Control Variables

This study selected gender, age, educational attainment, and personal monthly income as control variables to control for the potential interference of individual characteristics on the research results. In e-commerce research, these demographic variables were considered important control factors as they are widely recommended in the existing literature and may influence an individual’s psychological state and purchase intention [87].

4.3. Sampling

China dominates global live-streaming e-commerce with the largest market size and most active user activity. This study conducted an online survey in China using structured questionnaires. From 2 February 2025 to 10 February 2025, we distributed 426 questionnaires to the target group via the online questionnaire collection platform Credamo. The questionnaire included screening questions to eliminate invalid responses, resulting in 410 valid responses retained for analysis.

Two methods are adopted in this study to determine the sample size. Firstly, the sample size in this study satisfies the PLS-SEM requirements based on the rule of thumb that each indicator requires at least 10 samples [88]. Secondly, a priori methodology represents the optimal approach for determining necessary sample size and study power prior to the research design phase. This approach can effectively reduce the risk of Type I and Type II errors in the hypothesis testing process [89]. Based on existing research [90], we further employ the widely used statistical power analysis software G*Power (3.1.9) for rigorous calculation to determine the minimum required sample size. With an effect size of 0.5, a sample size of 210 is required to achieve statistical power of 0.95. Therefore, we have a sufficient sample size to conduct statistical analysis.

Demographic characteristics of the participants are shown in Table 2. The sample was predominantly females (70.2%, n = 288). Age distribution peaked in the 20–29 cohort (47.6%), with 62.5% holding bachelor’s degrees or higher. The gender and age distribution of the research subjects matches that of China’s current live-streaming e-commerce consumers (see: https://www.iimedia.cn/c1088/106477.html (accessed on 8 July 2025)). Therefore, this demographic confirms the sample’s strong requirement matching and generalizability. Monthly incomes exceeded CNY 5200 for 51.2% of respondents. Regarding live-streaming commerce engagement: 36.1% reported high proficiency, 25.4% viewed streams ≥ 7 times weekly, and 24.1% averaged > 6 viewing hours weekly.

Table 2. Sample profile.

Variable	Category	N	Percentage (%)
Gender	Male	122	29.8
	Female	288	70.2
Age	<20	6	1.5
	20–29	195	47.6
	30–39	178	43.4
	40–49	18	4.4
	>50	13	3.2
Educational level	Junior high school or below	1	0.2
	High school	68	16.6
	Junior college	85	20.7
	Undergraduate college	160	39.0
	Graduate and above	96	23.4
Proficiency	Very unskilled	2	0.5
	Not skilled	6	1.5
	General proficiency	124	30.2
	More skilled	130	31.7
	Very skilled	148	36.1
Frequency	1–2 times per week	104	25.4
	3–4 times per week	102	24.9
	5–6 times per week	100	24.4
	7 or more times per week	104	25.4
Viewing time (Average per week)	Within 2 h	91	22.2
	2–4 Hours	124	30.2
	4–6 Hours	96	23.4
	More than 6 h	99	24.1
Income (CNY per month)	<2500	94	22.9
	2500–4999	106	25.9
	5000–7499	74	18.0
	7500–9999	65	15.9
	≥10,000	71	17.3

4.4. Normality Test

We use a web-based calculator to test the multivariate normality of the data through the test [91]. Multivariate normality is one of the criteria for more accurate model prediction. The outcome of the multivariate normality analysis shows that Mardia’s multivariate skewness (β = 63.438, p < 0.01) and multivariate kurtosis (β = 673.814, p < 0.01) suggest multivariate non-normality. This result further confirms the appropriateness of the PLS-SEM approach, which demonstrates strong adaptability to non-normal data distributions.

4.5. Data Analysis

This study adopted a dual analysis method combining symmetry and asymmetry. The main analysis adopted the symmetrical method. To explore and evaluate the role of the AI streamers on purchase intention, the PLS-SEM method was employed. PLS-SEM is mainly suitable for exploratory theoretical research and can ensure the integrity of all relationships between independent and dependent variables during analysis. PLS-SEM is more adaptable than CB-SEM in the theoretical development stage and has been proven to be an effective alternative to CB-SEM in most social science research [92]. This study is more suitable for using PLS-SEM for the following reasons: (1) PLS-SEM is more suitable for complex models involving more than six variables; (2) PLS-SEM can effectively handle small sample data; and (3) PLS-SEM is suitable for non-normally distributed data. The Smart PLS v4.1.1 software was used for measurement and structural model evaluation. According to the previous literature [92], this study adopted a two-stage analytical approach that consisted of measurement model evaluation followed by structural model assessment.

In addition to the symmetrical analysis method, this paper employs fuzzy-set qualitative comparative analysis (fsQCA) as an asymmetrical analysis method. While PLS-SEM effectively verifies antecedent variables’ net effects on outcomes, it inadequately addresses complex causal interdependencies among multiple antecedents. To address this limitation and better explain the multi-faceted nature of consumer behavior in the e-commerce practice, this study adopts fsQCA as a supplement. This method overcomes this limitation by pinpointing core condition configurations and necessary-but-insufficient conditions. Furthermore, this method offers a new analytical perspective on the complexity of consumer intentions. Consequently, this study integrates PLS-SEM and fsQCA to verify the theoretical model’s underlying triggering logic and elucidate multiple pathways influencing purchase intention in AI-streamed e-commerce.

4.6. Common Method Bias

Given that the latent variables in this study are derived from a single data source, common method bias (CMB) may exist. This bias can be detected by analyzing whether a single factor dominates the variation between variables. In a study, the two main strategies for reducing the risk of CMB are procedural design and statistical control [93]. In terms of procedural design, this study adopted multiple measures: ensuring respondents’ anonymity to minimize social desirability bias, optimizing the questionnaire structure to enhance simplicity and readability, placing demographic questions at the end of the questionnaire to avoid leading effects, and conducting a pretest of the questionnaire before formal data collection. Two statistical control methods were employed to address potential CMB. Firstly, we employ the Harman single-factor test to evaluate the CMB. The results indicate that the total variance explained by a single factor is 46.104% of the total variance, which is below the critical value of 50%. Secondly, this study examines the potential existence of multicollinearity by evaluating the variance inflation factor (VIF). All VIF values remain below the critical threshold of 3.3, which indicates no evidence of multicollinearity [94]. Finally, according to existing research [95], most of our pathways involve regulatory effects, the measurement of which is unlikely to be substantially influenced by CMB. Therefore, CMB is not a significant issue in this study.

5. Results

5.1. PLS-SEM

5.1.1. Assessment of Reflective Constructs

In order to examine the adequacy of the measurement model, reliability and validity analyses are performed. The instrument reliability is estimated by using Cronbach’s α, composite reliability (CR), and average variance extracted (AVE). As shown in Table 3, the values of α and CR surpass the recommended threshold of 0.7. The value of AVE also exceeds the recommended threshold of 0.5. Moreover, as illustrated in Table 4, the factor loadings of all items on the measured constructs are greater than their cross-factor loadings on other constructs. The results indicate that the measurement model has a strong reliability level. To determine the discriminant validity, Fornell and Larcker’s criteria [96] and the Heterotrait–Monotrait (HTMT) ratio criteria are also assessed using the threshold criteria suggested by [97]. The values of the HTMT ratio should be below the recommended threshold of 0.9 [98,99]. According to Table 5, the results demonstrate discriminant validity. According to Table 3 and Table 5, the results of the analysis indicate that that the study established convergent and discriminant validity.

Table 3. Construct validity.

Construct		Items	Factor Loading	Mean	S.E.	Cronbach’s α	AVE	CR
AI streamers’ image characteristics (IC)	Cuteness (CU)	CU1	0.875	5.700	1.056	0.831	0.747	0.899
		CU2	0.853	5.530	1.223
		CU3	0.865	5.480	1.243
	Vitality (VI)	VI1	0.891	5.500	1.193	0.826	0.743	0.896
		VI2	0.840	5.690	1.145
		VI3	0.854	5.440	1.328
AI streamers’ ability characteristics (AC)	Professionalism (PR)	PR1	0.834	5.500	1.066	0.788	0.703	0.876
		PR2	0.833	5.430	1.137
		PR3	0.847	5.360	1.206
	Responsiveness (RE)	RE1	0.869	5.480	1.207	0.828	0.744	0.897
		RE2	0.847	5.370	1.227
		RE3	0.872	5.210	1.294
AI streamers’ scenario fit (SF)		SF1	0.836	5.620	1.049	0.784	0.698	0.874
		SF2	0.822	5.630	1.041
		SF3	0.849	5.690	1.021
Consumer innovativeness (CI)		CI1	0.837	5.680	0.968	0.813	0.728	0.889
		CI2	0.864	5.410	1.300
		CI3	0.858	5.660	1.130
Consumer trust (CT)		CT1	0.807	5.650	1.222	0.783	0.698	0.874
		CT2	0.820	5.650	1.100
		CT3	0.877	5.620	1.051
Purchase intention (PI)		PI1	0.792	5.900	1.037	0.793	0.708	0.879
		PI2	0.847	5.670	0.982
		PI3	0.884	5.660	0.989

Table 4. Cross-loadings analysis with PLS confirmatory factor analysis.

Construct		Item	CU	VI	PR	RE	SF	CI	CT	PI
AI streamers’ image characteristics (IC)	CU	CU1	0.879	0.653	0.502	0.470	0.545	0.514	0.548	0.632
		CU2	0.851	0.601	0.526	0.516	0.492	0.458	0.546	0.569
		CU3	0.863	0.618	0.551	0.522	0.485	0.505	0.571	0.556
	VI	VI1	0.636	0.892	0.551	0.537	0.540	0.462	0.529	0.568
		VI2	0.609	0.839	0.552	0.503	0.529	0.450	0.470	0.506
		VI3	0.622	0.854	0.553	0.513	0.471	0.436	0.491	0.489
AI streamers’ ability characteristics (AC)	PR	PR1	0.491	0.525	0.838	0.557	0.551	0.380	0.480	0.499
		PR2	0.512	0.546	0.831	0.505	0.479	0.355	0.458	0.480
		PR3	0.528	0.539	0.846	0.525	0.505	0.430	0.481	0.507
	RE	RE1	0.525	0.519	0.530	0.867	0.516	0.432	0.532	0.461
		RE2	0.469	0.493	0.539	0.850	0.454	0.394	0.447	0.455
		RE3	0.510	0.543	0.565	0.871	0.522	0.418	0.499	0.443
AI streamers’ scenario fit (SF)		SF1	0.478	0.485	0.519	0.452	0.836	0.403	0.543	0.549
		SF2	0.503	0.524	0.493	0.475	0.822	0.395	0.522	0.526
		SF3	0.492	0.487	0.519	0.518	0.849	0.405	0.567	0.587
Consumer innovativeness (CI)		CI1	0.408	0.363	0.311	0.338	0.366	0.837	0.393	0.420
		CI2	0.560	0.527	0.469	0.459	0.441	0.864	0.423	0.434
		CI3	0.485	0.439	0.400	0.429	0.418	0.858	0.404	0.402
Consumer trust (CT)		CT1	0.467	0.440	0.438	0.398	0.470	0.297	0.807	0.543
		CT2	0.559	0.477	0.478	0.526	0.574	0.428	0.820	0.564
		CT3	0.575	0.524	0.495	0.499	0.581	0.458	0.877	0.625
Purchase intention (PI)		PI1	0.576	0.517	0.503	0.419	0.536	0.405	0.552	0.793
		PI2	0.558	0.488	0.498	0.419	0.563	0.357	0.578	0.846
		PI3	0.578	0.521	0.493	0.485	0.575	0.475	0.618	0.884

Note: CU = cuteness, VI = vitality, PR = professionalism, RE = responsiveness.

Table 5. Discriminant validity.

	CI	CT	CU	PI	PR	RE	SF	VI
Fornell–Larcker Criterion (1981) [96]
CI	0.853
CT	0.477	0.835
CU	0.569	0.642	0.864
PI	0.491	0.693	0.677	0.842
PR	0.463	0.564	0.609	0.591	0.838
RE	0.481	0.572	0.582	0.524	0.631	0.863
SF	0.480	0.652	0.587	0.664	0.610	0.577	0.836
VI	0.521	0.577	0.722	0.605	0.640	0.601	0.596	0.862
Heterotrait–Monotrait Ratio
CI
CT	0.592
CU	0.690	0.793
PI	0.611	0.877	0.835
PR	0.576	0.717	0.753	0.749
RE	0.584	0.705	0.701	0.647	0.781
SF	0.600	0.827	0.728	0.841	0.776	0.715
VI	0.633	0.714	0.871	0.747	0.794	0.726	0.741

Note: CI = consumer innovativeness, CT = consumer trust, CU = cuteness, VI = vitality, PR = professionalism, RE = responsiveness, SF = AI streamers’ scenario fit, PI = purchase intention.

5.1.2. Assessment of the Formative Construct

In this study, the image and ability characteristics of AI streamers are proposed as second-order (reflective-formative) constructs. According to Becker et al. [100] and Hair et al. [101], the two-step approach put forward is employed to examine these higher-order constructs. In the first step, the repeated indicator approach is utilized to attain the latent variable scores. In the second step, the weights and significance levels are calculated based on the latent variable scores obtained from the PLS algorithm. The validity of the formative measurement is evaluated through the collinearity of the indicators (VIF) and the significance of weights. The results are presented in Table 6. According to Hair et al. [93], all VIF values of the measured items are below the threshold of 3.3, indicating that there was no serious problem of collinearity. We use the 5000 resample bootstrap technique to assess the significance of the weight. According to the results in Table 6, all weights demonstrate statistical significance at the p < 0.001 level. This confirms that each formative indicator contributes substantially to the higher-order reflective–formative construct.

Table 6. Assessment of higher-order construct.

Higher-Order Construct	Formative Indicators	Outer Weights	VIF	T-Value	p-Value
AI streamers’ image characteristics	Cuteness	0.549	2.088	68.672	0.000
	Vitality	0.529	2.088	67.362	0.000
AI streamers’ ability characteristics	Professionalism	0.553	1.661	64.727	0.000
	Responsiveness	0.555	1.661	55.936	0.000

5.1.3. Structural Model and Hypothesis Testing

The hypotheses are tested using PLS-SEM. Following the recommendations [101], we adopt the standardized root mean square residual (SRMR) to test the goodness-of-fit of the structural model. In the estimation model, the value of SRMR should be less than 0.08 to be regarded as a good approximate fit. The estimated model’s SRMR value is SRMR = 0.076, which means that it is considered a good fit. Based on Table 7, the results allow for the testing of the main effect hypothesis.

Table 7. Results of the direct effects.

Hypothesis	β	S.E.	T-Value	95% Bca-CIs	p-Value	Remarks
H1a: IC → PI	0.315	0.063	5.024	[0.192; 0.441]	0.000	Supported
H1b: AC → PI	0.055	0.061	0.905	[−0.060; 0.178]	0.365	Not Supported
H1c: SF → PI	0.241	0.056	4.301	[0.130; 0.352]	0.000	Supported
H1d: CT → PI	0.305	0.058	5.264	[0.193; 0.414]	0.000	Supported

Note: IC = AI streamers’ image characteristics, AC = AI streamers’ ability characteristics, SF = AI streamers’ scenario fit, PI = purchase intention, CT = consumer trust.

Next, to assess the quality of the structural model, the coefficient of determination ($R^2$), effect size ($f^2$), and predictive significance ($Q^2$) were reported. Overall, the model has considerable explanatory power, as consumer trust explains 54.3 percent of the variance in personification quality, service quality, and AI streamers’ scenario fit. Finally, by explaining 62.0 percent of the variance in purchase intention, personification quality, service quality, AI streamers’ scenario fit, and consumer trust, considerable explanatory potential in the model is indicated.

In terms of effect size, image characteristics ($f^2$ = 0.105) are found to be the most significant purchase intention predictor, with a medium effect size. On the other hand, consumer trust ($f^2$ = 0.104) and scenario fit ($f^2$ = 0.069), both with a medium effect size, are also important purchase intention predictors. The removal of the ability characteristic ($f^2$ = 0.004) had no significant impact on purchase intention, indicating that it lacks explanatory power in the model. Finally, using $Q^2$, predictive relevance was evaluated. The $Q^2$ value is greater than 0 for consumer trust (0.534) and purchase intention (0.560), demonstrating that the model is predictively valid. The findings indicate that $Q^2$_predict values are higher than $Q^2$ values, showing more consistency in the predictive potential of the model.

As shown in Table 7, the test results confirm that both of the proposed direct effect hypotheses received empirical support. AI streamers’ image characteristics are positively associated with purchase intention (β = 0.315, t = 5.024, p < 0.001), supporting H1a. Similarly, AI streamers’ scenario fit significantly affects purchase intention (β = 0.241, t = 4.301, p < 0.001), supporting H1c. Consumer trust is positively associated with purchase intention (β = 0.305, t = 5.264, p < 0.001), supporting H1d. Conversely, AI streamers’ ability characteristics are not positively associated on purchase intention (β = 0.055, t = 0.061, p > 0.05). H1b is not supported. Therefore, these results indicate that AI streamers’ image characteristics and scenario fit have a significant positive impact on purchase intention, while the ability characteristics do not show a significant influence. Moreover, we conduct additional checks to assess the impact of control variables in the model. There is no statistically significant correlation observed between gender (β = −0.097, p > 0.05), age (β = −0.003, p > 0.05), education level (β = 0.031, p > 0.05), income (β = −0.049, p > 0.05), and purchase intention.

5.1.4. The Mediating Effect of Consumer Trust

Based on Rungtusanatham et al. [102], we adopt the BC–bootstrap method to test the mediating effect, with 5000 repeated samplings. When zero is not included in the 95% confidence interval, the indirect effect is significant, indicating a significant mediating role. Table 8 illustrates both direct and indirect effects, which confirms that consumer trust partially mediates the relationship between AI streamers’ image characteristics, ability characteristics, and scenario fit. According to Table 8, the results show that the total effect of AI streamer’s image characteristics on purchase intention (β = 0.411, t = 5.908, p < 0.001) is significant, while the indirect effect (β = 0.096, t = 3.336, p < 0.01) is also significant, confirming H2a. In addition, the direct effect of consumer trust on purchase intention is significant (β = 0.315, t = 5.024, p < 0.001), which suggests partial complementary mediation. Similarly, the total effect of the AI streamer’s scenario fit on purchase intention (β = 0.342, t = 5.697, p < 0.001) is significant, and the indirect effect (β = 0.101, t = 3.580, p < 0.001) is also significant. The direct effect of consumer trust on purchase intention is significant (β = 0.241, t = 4.303, p < 0.001), which suggests partial complementary mediation. However, the total effect of AI streamer’s ability characteristics on purchase intention (β = 0.111, t = 1.754, p > 0.05) is not significant, while the indirect effect (β = 0.056, t = 2.661, p < 0.01) is significant. Consumer trust plays a fully mediating role in the impact of AI streamers’ ability characteristics on purchase intention. This result further proves that the direct impact of the ability characteristics on purchase intention is not significant (β = 0.111, t = 1.754, p > 0.05). Therefore, H2a, H2b, and H2c are supported. Therefore, these results indicate that consumer trust plays a significant mediating role in the relationship between AI streamers’ image characteristics, ability characteristics, scenario fit, and purchase intention.

Table 8. Mediating effect estimates.

Model Path	Parameter	β	S.E.	T-Value	p-Value	VAF (%)
IC → CT → PI	Indirect effect	0.096	0.029	3.336	0.000	23.358
	Direct effect	0.315	0.063	5.024	0.000
	Total effect	0.411	0.070	5.908	0.000
SF → CT → PI	Indirect effect	0.101	0.028	3.580	0.000	29.532
	Direct effect	0.241	0.056	4.303	0.000
	Total effect	0.342	0.060	5.697	0.000

Note: IC = AI streamers’ image characteristics, CT = consumer trust. SF= AI streamers’ scenario fit, PI = purchase intention.

Furthermore, the strength of some mediating effects can be evaluated through the variance accounted for (VAF). According to the literature [96], a VAF value exceeding 0.2 indicates the presence of a mediating effect, which is classified as partial when the VAF ranges between 0.2 and 0.8 and considered complete when it exceeds 0.8. As shown in Table 8, AI streamers’ characteristics and scenario fit partially mediate the purchase intention, with VAF values of 23.358% and 29.532%, respectively.

5.1.5. The Moderating Effect of Consumer Innovativeness

Using the bias-corrected and accelerated bootstrap method in Smart PLS v4.1.1, this study investigates the moderating effect of consumer innovativeness on the relationships among AI streamers’ characteristics, scenario fit, and consumer trust. Using parameters from the program under two levels of the moderating variable (high vs. low), we perform 5000 bootstrap iterations to generate 95% confidence intervals for composite coefficients, thereby assessing the statistical significance of moderated mediation. Subsequently, we dichotomize consumer innovativeness into high (+1 SD), low (−1 SD), and mean groups. Table 9 presents the results.

Table 9. Moderating effects.

Parameters	Effect Size	S.E.	T-Value	p-Value	Remarks
CI × IC → CT	−0.095	0.033	2.834	0.005	Supported
CI × AC → CT	−0.072	0.028	2.564	0.010	Supported
CI × SF → CT	−0.076	0.027	2.835	0.005	Supported

Note: IC = AI streamers’ image characteristics, AC = AI streamers’ ability characteristics, SF = AI streamers’ scenario fit, CI = consumer innovativeness, CT = consumer trust.

H3a and H3b indicate that consumer innovativeness negatively moderates the relationship between the image and ability characteristics of AI streamers and consumer trust. The interaction term between AI streamers’ image characteristics (β = −0.095, t = 2.834, p < 0.01) and ability characteristics with (β = −0.072, t = 2.564, p < 0.05) consumer innovativeness significantly negatively influences consumer trust. Hence, H3a and H3b are fully supported. Therefore, these results reveal that consumer innovativeness has a significant moderating effect on the relationship between AI streamers’ image characteristics, ability characteristics, scenario fit, and consumer trust. The information of path coefficients is used for plotting the simple slope analysis, indicating the improved relationship (Figure 2). As shown in Figure 2, the effect line between the low-level AI streamers’ image characteristics, ability characteristics, and the scenario fit has a steeper slope than that of high-level streamers. This steeper slope indicates that these variables have a negative impact on purchase intention. Simple slope analysis (Figure 2a) reveals that the image characteristics’ positive effect on trust weakened at high consumer innovativeness (+1 SD: β = 0.134, t = 1.395, p > 0.05) but strengthened at low innovativeness (−1 SD: β = 0.323, t = 2.941, p < 0.01). Similarly, the positive influence of ability characteristics on trust weakened at high consumer innovativeness (+1 SD: β = 0.078, t = 1.068, p > 0.05) but strengthened at low innovativeness (−1 SD: β = 0.223, t = 3.300, p < 0.01). H3c indicates that consumer innovativeness negatively moderates the relationship between AI streamers’ scenario fit and consumer trust. Based on Figure 2b, the interaction term between AI streamers’ scenario fit and consumer innovativeness has a significant negative impact on consumer trust (β = −0.076, t = 2.835, p < 0.01). Hence, H3c is fully supported. Likewise, the information of path coefficients is used for plotting the simple slope analysis (Figure 2c). Simple slope analysis (Figure 2c) reveals that the scenario fit’s positive effect on trust strengthened at high innovativeness (+1SD: β = 0.198, t = 2.503, p < 0.05) but weakened at low innovativeness (−1 SD: β = 0.350, t = 5.685, p < 0.001).

Figure 2. Moderating effect of consumer innovativeness on AI streamers and consumer trust.

5.1.6. Multigroup Analysis

The multigroup analysis method is used to test the path coefficients and hypotheses of the research model under the contexts of incremental products and disruptive products. We use the PLS-MGA method to evaluate the differences in variation in the path coefficient for incremental and disruptive products. Before running MGA, the measurement invariance is first tested based on the three-step MICOM method [103]. According to the literature [104], the results show that all p values are greater than the critical threshold of 0.05. Table 10 presents the MICOM analysis results. These results confirm that the model satisfies configural invariance, metric invariance, and equality of means and variances [105].

Table 10. Results of invariance measurement testing using permutation.

Constructs	Configural Invariance		Mean Invariance		Variance Invariance		Configurational Invariance
	Correlation	p Value	Difference	p Value	Difference	p Value
IC	1.000	0.758	0.024	0.838	0.027	0.918	Yes
AC	1.000	0.304	−0.068	0.494	−0.201	0.344	Yes
SF	1.000	0.697	−0.087	0.364	−0.096	0.727	Yes
CT	1.000	0.504	0.032	0.734	0.041	0.883	Yes
CI	0.999	0.084	0.036	0.717	−0.071	0.725	Yes
PI	0.999	0.582	0.030	0.757	0.113	0.695	Yes

Note: IC = AI streamers’ image characteristics, AC = AI streamers’ ability characteristics, SF = AI streamers’ scenario fit, CI = consumer innovativeness, CT = consumer trust, PI = purchase intention.

The comparative analysis results of paths for different product types are shown in Table 11. In the context of disruptive products, the role of AI streamers’ image characteristics (β = 0.088, t = 2.290, p < 0.05) and scenario fit (β = 0.112, t = 2.947, p < 0.01) in influencing purchase intention through consumer trust is significant. In contrast, in incremental products, the influence of AI streamers’ image characteristics (β = 0.040, t = 1.226, p > 0.05) and scenario fit (β = 0.044, t = 1.480, p > 0.05) on consumer trust and purchase intention is not significant. This divergence requires strategic resource reallocation. Enterprises should deploy AI streamers according to product innovation type. For disruptive products, they should implement systems with expressive anthropomorphic designs and immersive scenarios that build consumer trust through visual appeal and scenario consistency. These synergies lower psychological barriers through character appeal and convert novelty perceptions into purchases. For incremental products, they should redirect saved development resources toward practical product demonstrations. Enterprises should divert AI performance budgets toward disruptive product launches.

Table 11. Product type path comparison analysis.

Path	Path Coefficient (Incremental Product)	T Value	p Value	Path Coefficient (Disruptive Product)	T Value	p Value
IC → PI	0.230	2.234	0.026	0.346	4.121	0.000
IC → CT → PI	0.040	1.226	0.220	0.088	2.290	0.022
CI × IC → CT	−0.109	2.795	0.005	−0.091	1.329	0.184
AC → PI	0.260	3.079	0.002	−0.025	0.302	0.763
AC → CT → PI	0.058	1.829	0.067	0.032	1.196	0.232
CI × AC → CT	−0.115	2.992	0.003	−0.026	0.433	0.665
SF → PI	0.260	3.079	0.002	0.232	3.083	0.002
SF → CT → PI	0.044	1.480	0.139	0.112	2.947	0.003
CI × SF → CT	−0.087	2.529	0.011	−0.088	1.544	0.123
CT → PI	0.263	3.146	0.002	0.324	4.155	0.000

Note: IC = AI streamers’ image characteristics, AC = AI streamers’ ability characteristics, SF = AI streamers’ scenario fit, CI = consumer innovativeness, CT = consumer trust. PI = purchase intention.

In the context of incremental products, consumer innovativeness exerts a significant moderating effect on the influence of the AI streamers’ image characteristics (β = −0.109, t = 2.795, p < 0.01), ability characteristics (β = −0.115, t = 2.992, p < 0.01), and scenario fit (β = −0.087, t = 2.529, p < 0.05) on consumer trust. In contrast, in disruptive products, the moderating effect of the AI streamers’ characteristics (β = −0.091, t = 1.329, p > 0.05), ability characteristics (β = −0.026, t = 0.433, p > 0.05), and scenario fit (β = −0.088, t = 1.544, p > 0.05) on purchase intention is not significant. Enterprises can use existing consumer innovativeness assessments to identify high-potential customer segments. For these segments, companies should minimize AI streamers’ anthropomorphic designs and technical demonstrations. They should instead adopt practical product demonstrations that highlight core functionalities. With disruptive products, firms require no extensive differentiation of consumer innovation levels.

In the context of disruptive products, AI streamers’ image characteristics exert a significant influence on purchase intention (β = 0.346, t = 4.121, p < 0.001), and the same is true for incremental products (β = 0.230, t = 2.234, p < 0.05). Similarly, in the context of disruptive (β = 0.232, t = 3.083, p < 0.01) and incremental products (β = 0.260, t = 3.079, p < 0.01), AI streamers’ scenario fit has a significant impact on purchase intention. For incremental products (β = 0.263, t = 3.146, p < 0.01) and disruptive products (β = 0.324, t = 4.155, p < 0.001), consumer trust has a significant impact on purchase intention. Consumers prioritize positive emotional experiences in live-streaming e-commerce. When AI streamers exhibit strong cuteness and vitality while their promotion scenarios maintain high product alignment, consumers perceive more authentic human-like interactions. This consistency deepens audience understanding of both the streamer and products, which creates an enjoyable viewing experience that builds trust and strengthens purchase intention. Enterprises that market products with varying innovation levels should prioritize AI streamers with enhanced appeal and vitality. These businesses must also develop live-stream scenarios that demonstrate strong alignment with their products.

In the context of incremental products, the role of AI streamers’ ability characteristics (β = 0.260, t = 3.079, p < 0.01) in influencing purchase intention is significant. In contrast, in disruptive products, the influence is not significant (β = −0.025, t = 0.302, p > 0.05). Enterprises should utilize AI hosts that demonstrate professional knowledge and deliver real-time responses for progressive products. Product explanations and immediate Q&A sessions enhance consumer perception and purchase intention. For disruptive products, they should adjust communication strategies by reducing emphasis on AI expertise and responsiveness and reallocate resources from AI streamer development to alternative feature construction.

This study reveals that disruptive and incremental products exhibit differentiated influence mechanisms in the relationship between consumer innovativeness and AI streamers on consumer trust. Moreover, the scene dimension of AI streamers may need to further introduce latent variables, such as the informativeness of the scene, which may have a significant impact on consumer trust and purchase intention in the context of AI live-streaming.

5.2. fsQCA

5.2.1. Variable Calibration

This study examines the joint effects of influencing factors on consumer purchase intention in AI streamers’ e-commerce sales by using fsQCA. We designate AI streamers’ characteristics (e.g., cuteness, vitality, professionalism, and responsiveness), scenario fit, consumer innovativeness, and consumer trust as conditions, with purchase intention as the outcome variable. Following the previous literature [106], we establish three calibration anchors: 5% (full non-membership), 50% (crossover point), and 95% (full membership). All calibrated values of 0.5 received 0.001 adjustment. Table 12 presents the results.

Table 12. Calibration.

Construct		Fully In	Crossover Point	Fully Out
AI streamers’ image characteristics	Cuteness	6.500	5.833	3.408
	Vitality	6.333	5.667	3.667
AI streamers’ ability characteristics	Professionalism	6.667	5.667	4.333
	Responsiveness	6.667	5.667	3.667
AI streamers’ scenario fit		6.667	6.667	5.667
Consumer innovativeness		6.667	6.667	5.667
Consumer trust		6.667	6.500	5.833
Purchase intention		6.667	6.333	5.667

5.2.2. Necessity Analysis

Before conducting the conditional configuration analysis, this study performs a necessary condition test to assess whether a single variable constitutes a necessary condition for the outcome variable. According to Table 13, the consistency values of each individual antecedent condition did not exceed 0.9 [106]. The results demonstrate that no indicator constitutes a necessary condition for consumer trust in AI streamers. We therefore conduct an antecedent variable configuration analysis.

Table 13. Necessity analysis of single conditions.

Condition	Purchase Intention		~Purchase Intention
	Consistency	Coverage	Consistency	Coverage
AI streamers’ image characteristics	0.776	0.845	0.654	0.491
~AI streamers’ image characteristics	0.533	0.690	0.793	0.710
AI streamers’ ability characteristics	0.739	0.829	0.629	0.488
~AI streamers’ ability characteristics	0.543	0.680	0.780	0.674
AI streamers’ scenario fit	0.780	0.830	0.662	0.486
~AI streamers’ scenario fit	0.517	0.689	0.768	0.707
Consumer innovativeness	0.782	0.818	0.669	0.483
~Consumer innovativeness	0.506	0.689	0.748	0.703
Consumer trust	0.802	0.819	0.696	0.490
~Consumer trust	0.500	0.704	0.743	0.722

5.2.3. Configuration Analysis

This study configures fsQCA thresholds at case frequency = 3, consistency = 0.8, and PRI consistency = 0.75 and then performs path standardization. Boolean operations yield three solution types: complex, parsimonious, and intermediate. We adopt the intermediate solution because it can distinguish the core conditions (strong causal association) and marginal conditions (weak causal association) in each solution configuration. The intermediate solution analysis (shown in Table 14) reveals both high-purchase-intention and low-purchase-intention configurations. Our study focuses on high-purchase-intention configurations because they have greater theoretical and practical significance. The results demonstrate that the overall solution consistency for high purchase intention is 0.864 (>0.75), while the overall consistency for low purchase intention is 0.494. These results indicate that the overall consistency possesses a satisfactory explanatory power. The overall explanatory power for high purchase intention is 0.803, while the overall coverage for low purchase intention is 0.718. This indicates the extent to which the configuration interpretation is valid for the cases.

Table 14. The combination configuration of the impact of AI streamers on purchase intention.

Antecedent Condition	Purchase Intention								~Purchase Intention
	S1	S2	S3	S4	S5	S6	S7	S8	S1	S2	S3	S4	S5
AI streamers’ image characteristics	⬤	⬤	⬤			⬤				●		●	⊗
AI streamers’ ability characteristics	⊗	⬤	⬤	⬤	⬤		⬤		●			⊗	●
AI streamers’ scenario fit	⬤			⬤	⬤			⬤			●	●	●
Consumer innovativeness		●		●		⬤	⬤	⬤	●
Consumer trust			●		●	⬤	⬤	⬤		●	●
Consistency	0.939	0.914	0.923	0.917	0.916	0.922	0.923	0.923	0.512	0.525	0.521	0.740	0.766
Raw coverage	0.376	0.608	0.610	0.596	0.607	0.626	0.599	0.615	0.545	0.585	0.581	0.429	0.438
Unique coverage	0.024	0.013	0.008	0.010	0.014	0.020	0.006	0.019	0.027	0.038	0.027	0.019	0.004
Overall solution consistency	0.864								0.494
Overall solution coverage	0.803								0.718

Note: ● and ⊗ circles depict the presence and absence of a condition, respectively. Additionally, large circle and the small circles depict core conditions and the peripheral conditions, respectively. Blank space depicts a “don’t care” situation.

In live-stream commerce, AI streamers’ image characteristics, ability characteristics, and scenario fit serve as core factors that influence consumer purchase intention, playing a key role in driving high purchase decisions. Firstly, AI streamers’ image characteristics help enhance high purchase intention. Configurations 1, 2, 3, and 6 all consider AI streamers’ image characteristics as core conditions, which effectively validates their positive role in influencing purchase intention. Similarly, configurations 1, 4, 5, and 8 all consider AI streamers’ scenario fit as a core condition for achieving high purchase intention. Therefore, H1a and H1c are further supported. Secondly, consumer trust directly enables high purchase intention. Configurations 6, 7, and 8 all take consumer trust as the core condition, which confirms its positive effect on purchase intention. H1d is further supported. Thirdly, the absence of AI streamers’ image characteristics and scenario fit impacts purchase intention. However, AI streamers have matching live-streaming scenarios and other characteristics, which also help enhance purchase intention. Finally, the significance of these findings was further confirmed by the analysis of the conditions leading to low purchase intention. The results indicate that the absence of AI streamers’ image characteristics, scenario fit, consumer trust, and consumer innovativeness all make it difficult to effectively enhance purchase intention.

6. Discission and Conclusions

6.1. Key Findings

Based on the SOR framework, this study develops a model to examine the impact of AI streamers on consumer trust and purchase intention. Using SEM and fsQCA methodologies, we analyze mechanisms that link AI streamers’ characteristics and scenario fit to purchase intention. The research findings are as follows.

First, AI streamers’ image characteristics and scenario fit are positively associated with consumer trust, which subsequently enhances purchase intention. These findings align with previous live-streaming studies [22,24,43] and confirm the positive impact of streamer characteristics [3]. AI streamers’ image characteristics and scenario-aligned environments constitute key factors that build consumer trust. These elements collectively drive purchase decisions. Brands strategically design AI personas with emotionally resonant characteristics and contextually matched scenarios to activate consumer pathways from trust to conversion. In traditional live-streaming with humans, trust stems from the streamer’s professional ability and immediate response during the interaction process. However, AI streamers’ ability characteristics have no significant impact on consumer trust and purchase intention. Highly intelligent AI streamers answer consumer questions with professional and rapid responses, which reduces their performance gap with human streamers [44]. Moreover, the uncanny valley theory suggests that excessively human-like behaviors in artificial intelligence may evoke user discomfort or rejection [107]. Consumers also recognize that the demonstrated professionalism and responsiveness originate from algorithmic programming rather than autonomous cognition or genuine emotion. This cognitive limitation restricts the trust-building of consumers towards AI streamers, thereby weakening their purchasing intentions.

Furthermore, high product–environment compatibility directly fosters positive attitudes toward the products [64]. To implement this insight, marketers need to create seamless integrations between promoted products and streaming contexts. This integration can enhance consumer trust and accelerate purchase intention. As a result, scenarios designed to align with products improve consumer experiences [63]. The optimization of consumer experiences relies on proactively aligning scenario design with product attributes. This alignment can transform passive viewers into engaged purchasers through curated environmental congruence.

Second, consumer innovativeness acts as a significant barrier to trust—significantly negating AI streamers’ impact on consumer trust. Specifically, AI streamer characteristics and well-matched scenarios enhance consumer trust, which in turn raises purchase intention. These findings indicate that trust functions as a pivotal psychological gateway through which AI streamers impact consumer decisions. This underscores the necessity of crafting AI streamers that are not only emotionally engaging but also contextually pertinent. Rather than acting as passive sales tools, AI streamers must simulate credible social presence to drive behavioral outcomes.

Third, consumer innovativeness significantly negatively moderates AI streamers’ impact on consumer trust, a finding that contradicts previous research [10,13,14]. This counterintuitive finding reveals a strategic blind spot: innovative consumers’ novelty-seeking tendencies raise their technological expectations and accelerate detection of AI limitations. As a result, standard AI streamer deployments often fail to meet the novelty thresholds of this consumer segment, thereby eroding consumer trust and purchase intention. Moreover, product differentiation dictates AI trust-building efficacy. This pattern aligns with established influences of product type [108]. Where disruptive products require demonstrated technical mastery from AI streamers, incremental products benefit more from scenario-embedded usability proofs.

6.2. Theoretical Contributions

This study makes three key theoretical contributions. First, this study advances research on AI streamers’ purchasing intention mechanisms. Previous research has emphasized the socio-technological attributes of human streamers. In contrast, this study establishes AI streamers’ characteristics and scenario fit as critical drivers. By introducing consumer trust as a mediator, we reveal new pathways to enhance AI streamer marketing and purchase intention. Second, this study demonstrates how product type and consumer innovativeness moderate AI streamers’ impact on purchase intention. Analysis of these moderators shows their differential effects on characteristic–scenario relationships, extending applications in AI live-streaming e-commerce. Finally, we broaden purchase intention antecedents through fsQCA. The conditions include AI streamers’ characteristics, scenario fit, consumer innovativeness, and consumer trust. This analytical approach identifies core configuration paths that affect purchase decisions, validates empirical findings, and uncovers key causal combinations.

6.3. Managerial Implications

Based on the research conclusions, this study provides managerial implications for enterprises, platforms, and the government in AI live-streaming e-commerce across four key dimensions: AI streamers’ image and ability characteristics, scenarios, heterogeneous consumers, and differentiated products.

First, small- and medium-sized enterprises should select AI anchors that specialize in single categories, which prevents resource fragmentation from cross-category operations. Small businesses typically concentrate on one or two core product lines. These firms can develop AI streamers with cute and energetic characteristics that align with their primary products through coordinated live-streaming environments. In contrast, large companies may increase R&D investment to deliver more personalized consumer experiences. These AI avatars should have warm, friendly appearances to enhance approachability and possess credible expertise to provide detailed product information. This combination ensures effective brand communication. Meanwhile, enterprises should not allocate funds to advanced interaction characteristics for AI streamers. These firms should select standardized interaction packages that provide essential functions. Specifically, enterprises can use 3D technology to enhance the body movements of AI streamers. Character modeling and motion capture enable natural gestures and postures, which improves information delivery. Additionally, voiceprint recognition and speech synthesis enhance AI streamers’ fluency, clarity, and expressiveness during live broadcasts.

Second, enterprises should design product-aligned scenarios to enhance consumer immersion and trust, as this strategy can stimulate purchases. Strategic room setups (e.g., backgrounds, product displays, props, lighting, sound, logos, color schemes, and visual elements) can create brand-coherent streaming environments. Large enterprises should deploy immersive technologies to elevate sales outcomes through scenario development. For premium product categories like home appliances and electronics, VR simulations enable AI hosts to demonstrate products in virtual living spaces. This system allows consumers to inspect internal mechanisms and access real-time performance metrics, which achieves seamless product–scenario integration. For fast-moving consumer goods such as cosmetics and apparel, AR virtual try-on functions permit users to visualize effects in digital makeup rooms or fitting rooms after they submit biometric data. Small- and medium-sized enterprises should choose high cost–performance scene solutions that are in line with their core products. For instance, those selling fast-moving consumer goods can adopt the platform’s AR cosmetic trial template. These enterprises only need to upload the product color data to achieve interactive functions. For those selling high-value products, they can utilize the virtual exhibition hall template, replacing the background with actual product photos and integrating the standardized function descriptions of AI hosts. This approach not only maintains the consistency between the product and the scene but also controls the operational costs.

Third, consumer innovativeness significantly and negatively moderates the effects of AI streamers on consumer trust, which requires differentiated marketing strategies. Moreover, this study reveals significant differences in how AI streamers’ characteristics and scenario fit affect consumer trust across breakthrough versus incremental innovations. Enterprises should strategically align innovative product types with target consumer segments. Enterprises should collect purchase behavior data through e-commerce platforms, social media, and offline stores to segment consumers into high-innovativeness and low-innovativeness groups. For high-innovation user segments, stream presentations should minimize technical details while emphasizing after-sales guarantees and exclusive benefits. These sessions should feature breakthrough innovations in high-relevance usage scenarios that maximize trust conversion. For low-innovation segments, they should strengthen product–scenario alignment through incremental innovations as core offerings. AI streamers’ appealing visual characteristics further enhance user trust within these segments. Small- and medium-sized enterprises can utilize platform data to categorize consumer segments and develop accurate user profiles as they control operational expenditures. For high-innovation user groups, streaming scripts should emphasize after-sales support and eliminate complex technical terminology. For low-innovation segments, they should focus promotional efforts on scenarios that align with product characteristics. Regarding product strategy, they should prioritize incremental innovations that professional-looking AI streamers present. When introducing breakthrough innovations, they should ensure strong scenario compatibility to prevent consumer trust erosion.

Finally, algorithm-driven interactions between AI streamers and consumers may erode user trust when individuals detect personal data collection or encounter misinformation. Platforms must implement mandatory identity disclosure mechanisms that identify AI-generated content through non-skippable popups at broadcast onset. These measures prevent cognitive bias that stems from information asymmetry. Governments should define legal boundaries for AI data collection and usage to protect privacy rights. Platforms also require dedicated units that address ethical complaints about AI live-stream commerce, which enables targeted governance of issues like misleading visuals or fabricated scenarios. Regulatory frameworks should further mandate comprehensive filing systems for AI streamer data, including image design source documentation and scenario modeling authenticity reports. These requirements prevent virtual image infringement and false product–scenario alignment.

6.4. Research Limitations and Future Studies

There are certain limitations associated with this study that suggest promising areas for future research. Firstly, while this article examines the impact of AI streamers on purchase intention across three dimensions, it does not concentrate on specific industries or brands. Future research could focus on the impact of AI streamers within particular industries or brands on consumer behavior. Secondly, this study employs questionnaire surveys as its primary methodology. Technical constraints introduce objective variables that include lighting variations, emotional inconsistencies in background music, and fluctuations in vocal delivery naturalness. These production factors may influence audience perceptions despite our directional prompts about AI streamer attributes. To control for such interferences, future research could add a “production quality interference group” (such as low lighting quality, chaotic background music) in each set of stimulus materials and eliminate the impact of production environment quality on the evaluation of AI streamers through comparative analysis. Furthermore, adding a survey item that explicitly asks about production quality influence would help separate these technical factors from streamer evaluations. By controlling potential influencing variables, it can better approximate real-life live-streaming scenarios. Thirdly, this study restricted control variables to demographic characteristics. It excluded influential factors like platform trust, brand familiarity, and prior virtual host experience. These omitted factors could moderate how consumers form purchase intentions. Future investigations should examine how these variables alter AI streamer influence mechanisms. Moreover, data collection for this study was concentrated within a relatively short period. While this approach helped to increase the questionnaire return rate and ensure the timeliness of responses, it might have affected the reliability of the sample. Future research could explore the design of more refined knowledge test items or experience assessment scales and combine them with in-depth interview formats to more effectively evaluate the depth of knowledge and understanding levels of respondents. Finally, this study’s empirical grounding rests exclusively on Chinese live-streaming e-commerce users. Although the Chinese market dominates the live-streaming e-commerce industry, there are limitations in terms of generalizability across different demographics. Future research could expand the sample population, such as comparative studies that can be conducted among different regions.