Weaving Cognition, Emotion, Service, and Society: Unpacking Chinese Consumers’ Behavioral Intention of GenAI-Supported Clothing Customization Services

Abstract

The progression of generative artificial intelligence (GenAI) has lowered the technical barriers for consumers to independently customize clothing. Its profit potential has driven fashion brands to adopt GenAI-based interactive customization services. However, the factors influencing consumers’ behavioral intention toward GenAI-supported clothing customization services (GAICCS) remain underexplored. In this study, an extended framework is constructed based on the Technology Acceptance Model (TAM), incorporating trust, perceived enjoyment, service quality, and social influence to examine their effects on behavioral intention across cognitive, affective, service, and social dimensions. A total of 692 valid responses were collected from Chinese consumers via convenience sampling. A multi-stage SEM-ANN analysis was conducted to test the model. The results show that perceived usefulness (β = 0.237, p < 0.001), perceived ease of use (β = 0.332, p < 0.001), social influence (β = 0.181, p < 0.001), and service quality (β = 0.093, p < 0.05) significantly enhance behavioral intention, with perceived ease of use being the most influential. Perceived ease of use and perceived usefulness also act as key mediators. Unlike previous studies, satisfaction showed no significant effect. This study underscores the importance of functional performance, social influence, and service quality in driving GAICCS adoption, providing theoretical insights and multi-level practical guidance for promoting GenAI in fashion contexts.

1. Introduction

As text-to-image and image-to-image functions in generative artificial intelligence (GenAI) technologies have emerged, a new form of clothing customization service—GenAI-based clothing customization services—has also taken shape. These services enable ordinary consumers to autonomously generate differentiated apparel products by inputting prompts and reference images, thereby lowering the threshold for non-professionals to participate in fashion design [1]. Unlike traditional clothing customization services—where customers could only participate in early-stage processes (e.g., providing inspiration) and faced long production cycles and high costs—GenAI-supported clothing customization services (GAICCS) have built foundational information infrastructures, integrated clothing design generation algorithms, and created a complex interactive system that enables users to input their preferences and collaborate with AI. More importantly, through GAICCS—an intelligent interface system—consumers are empowered to bypass designers and take control of the customization process themselves [1], thereby enhancing their identification with the product and the brand culture. This advantage has also prompted some strategically forward-looking traditional fashion brands, sportswear companies, and digital fashion startups to experiment with such customization services in response to the growing trend of intelligent fashion development. For example, H&M’s Creator Studio has introduced a GenAI-powered personalization service, through which consumers can input descriptive text and select design styles, color themes, and other parameters to receive a personalized clothing production plan [2]. The digital fashion brand Tribute has launched an AI-powered sweater generation platform, while the Chinese sportswear brand Xtep has also introduced a consumer-facing T-shirt customization service. Despite active coverage of GAICCS by industry media [3], no market statistics have been released to date, and user adoption data has yet to be publicly disclosed. Moreover, the inherent uncertainty of AI technologies may give rise to consumer distrust or even negative attitudes [4]. This highlights the necessity of conducting empirical research on consumer attitudes and behavioral intentions regarding GAICCS.

Existing studies have mainly concentrated on the influencing mechanisms of consumer acceptance toward online clothing customization [5] and digital apparel customization [6] and have not yet extended into the context of GenAI technologies. In the context of AI in fashion research, greater attention has been given to specific AI technologies and services, such as AI-powered avatars [7], AI chatbots for apparel shopping [8], and AI-curated fashion services [9]. However, these AI technology experiences in the fashion domain are typically closed-source, with limited user involvement during the usage process. In contrast, consumers take the lead in GAICCS, actively interacting with the system and making design-related decisions. This shift in the user’s role and level of participation raises new questions. In addition to evaluating perceived usefulness and perceived ease of use [10], it is also important to consider whether emotional, experiential, and social factors may shape consumers’ evaluations and attitudes. This is particularly relevant in the context of an integrated service system where users directly interact with the interface and receive design feedback. Although qualitative research has investigated consumers’ willingness toward self-directed clothing customization in GenAI-supported environments [11], there is still a lack of systematic quantitative studies that explore the combined effects of individual perceptions, service-related factors, and social dimensions on consumers’ behavioral intention toward GAICCS. Currently, most studies concerning AI applications in fashion have primarily utilized samples drawn from Western populations such as the United States [8,12,13]. These studies have emphasized factors such as expressive perception, optimism, innovativeness, and performance risk. Behavioral mechanism investigations in Western contexts typically reflect how individualistic cultural values shape concerns related to AI, such as identity threats, trust issues arising from the degree of autonomous decision-making, and the extent to which individuality and uniqueness can be expressed and realized [14]. By contrast, Chinese consumers demonstrate distinctive psychological and behavioral responses to AI technologies, shaped by their sociocultural environment. These include a collectivist culture and sensitivity to group norms [15], curiosity and interest in AI innovations [16], and a preference for actively receiving information during AI-assisted consumption, which generates a sense of empowerment [17]. These regional cultural differences—and the resulting distinctions in consumer psychology and behavior—may significantly limit the applicability of findings derived from studies based on Western populations when generalized to Chinese consumers. Therefore, applying and extending theoretically validated models from Western cultural contexts within China’s unique sociocultural environment can contribute to their cross-cultural validation, refinement, and further development. Meanwhile, China currently holds the world’s largest and most promising clothing market [18], making Chinese consumers a crucial group for investigating emerging fashion technologies and services. Therefore, investigating consumers’ behavioral intention toward GAICCS in the Chinese context holds both cross-cultural theoretical value and practical significance.

To address the identified theoretical and empirical gaps, this study focuses on Chinese consumers and adopts the Technology Acceptance Model (TAM) as its theoretical foundation. Recent research on the adoption of AI-powered fashion technologies has largely focused on evaluating their perceived usefulness and ease of use [9,19]. These factors are closely aligned with the core logic of the TAM, which emphasizes that users’ perceptions of a technology service’s functional effectiveness are key determinants of their behavioral intention. Given this theoretical consistency, the TAM remains a relevant and flexible theoretical foundation for constructing models to understand consumer behavior in AI-enabled fashion contexts [8]. Meanwhile, compared with the Unified Theory of Acceptance and Use of Technology (UTAUT) and the Theory of Planned Behavior (TPB), the TAM offers a more user-centric perspective driven by perceived practical efficiency and capability of novel technologies. Given the voluntary and exploratory nature of GAICCS usage, the TAM provides a more concise and flexible structure that allows for the integration of context-specific external variables tailored to the GAICCS context. However, variables such as effort expectancy in the UTAUT or subjective norms and perceived behavioral control in the TPB may not be applicable in the GAICCS context, and could even constrain the explanatory power of the model if rigidly imposed. Therefore, the TAM serves as a more relevant and appropriate theoretical foundation for investigating consumers’ behavioral intention to use GAICCS as active users. More importantly, GAICCS not only offers clothing customization services, but also functions as a human–AI collaborative system that supports user input and provides iterative design suggestions through interaction. As such, consumers’ behavioral intention as users may be shaped by a multidimensional and complex set of mechanisms. Therefore, this study leverages the structural flexibility of TAM and extends it to the context of GAICCS, enabling an exploration of consumers’ responses and affective perceptions across four dimensions: individual cognition of system effectiveness (perceived usefulness and perceived ease of use), emotional perceptions (perceived enjoyment and trust), service evaluations (service quality and satisfaction), and social evaluations (social influence). The four external variables—trust, perceived enjoyment, service quality, and social influence—were not arbitrarily selected; rather, they were identified through a literature review in relevant fields and further refined based on insights from semi-structured interviews with industry experts and consumers. These are combined with the core TAM constructs—perceived usefulness and perceived ease of use—as well as satisfaction. The integrated model not only captures individuals’ emotional perceptions and their evaluations of system effectiveness during interactions with the service, but also reflects externally driven feedback shaped by social evaluation.

Based on this, the study aims to explore the following two key questions in the context of the rapid integration of GenAI into fashion customization services: (1) Through what mechanisms do the core TAM constructs (perceived usefulness, perceived ease of use), satisfaction, and external variables (social influence, perceived enjoyment, trust, and service quality) influence Chinese consumers’ behavioral intention to adopt GAICCS? (2) Among these variables, which ones play significant mediating roles in shaping behavioral intention? These questions arise from both theoretical and practical gaps. First, existing TAM-based studies have rarely examined the context of GenAI-supported interactive fashion services. Second, although some fashion brands have begun experimenting with GAICCS, there remains a lack of understanding regarding how consumers interact with such technology-service systems under the influence of cognitive, emotional, and social evaluation dimensions—particularly in terms of the internal mechanisms among these variables. Therefore, investigating these two questions allows for a holistic examination of the mechanisms through which both external and internal factors influence the behavioral intention of GAICCS, as well as the interrelationships among these factors. Specifically, the GenAI-supported clothing customization service (GAICCS) in this study refers to an end-to-end intelligent customization platform launched by fashion brands based on GenAI technology, which functions similarly to an “operating system” by providing consumers with an integrated environment for feature invocation and personalized creative design.

This study enriches the discussion on consumers’ proactive adoption of GAICCS by offering a multidimensional perspective on behavioral intention. It also extends the application of the TAM framework to AI-enabled services and human–computer interaction, providing empirical insights into the functional design, service architecture optimization, and marketing strategies of such socio-technical systems in real-world usage contexts.

The subsequent sections outline the key theoretical underpinnings and summarize related literature. We then proceed to the empirical phase, in which key variables are identified and the extended TAM is developed based on literature synthesis and semi-structured interviews. Subsequently, the model and associated hypotheses are tested using SEM-ANN, followed by a presentation of results and discussion.

2. Literature Review

2.1. Application of GenAI in Clothing Customization

As human–AI interactive services based on GenAI technology—such as intelligent voice assistants offering personalized styling advice—become increasingly integrated into fashion shopping scenarios, consumers have gradually recognized that AI technologies can rival human capabilities in terms of efficiency and service quality [20]. On the other hand, consumer demand for personalized clothing has been steadily increasing [21]. These two factors together have driven the emergence of GenAI-supported clothing customization services (GAICCS), which feature rapid generation speed, intuitive outcomes [20], and support for unlimited revisions. Although GAICCS is considered a potentially disruptive innovation in the field of clothing customization, current users still exhibit partial distrust toward AI’s involvement in decision-making [17] and concerns about the quality of AI-generated outputs [11]. Therefore, there remains substantial room for improvement in enhancing user recognition and adoption of this socio-technical system. This concern has also driven the continuous iteration of AI-based customization technologies, including performing localized style transfer on clothing images [21] and the intelligent transfer of selected style elements [22]. These technological advancements have elevated the functional capabilities of GenAI-based clothing customization platforms, while also endowing GAICCS with the intelligent and interactive characteristics of a human–AI co-creative system. At the same time, these technological advancements have also influenced consumers’ cognitive evaluations of GAICCS functionalities, including perceived usefulness and perceived ease of use.

These core features also highlight the fundamental difference between GAICCS and traditional clothing customization services, which lies in the mode of information feedback—specifically, the former involves interaction between humans and AI, while the latter relies on communication between humans and human designers [5]. As illustrated in Figure 1, we highlight that traditional clothing customization services rely on an indirect feedback loop—where consumer needs must be interpreted and executed by human designers—often resulting in mismatches between consumer expectations and the final design. In contrast, direct interaction with GAICCS enables consumers to obtain design outcomes that more closely align with their original vision. Certainly, this fundamental difference also reflects a paradigm shift in clothing customization service systems. The interaction model enabled by GAICCS is characterized by real-time feedback, instant modifications, and data-driven creative outputs. As a result, it necessitates the introduction of new evaluation mechanisms for the technical system, such as information quality and responsiveness. However, these distinctive system-level features do not necessarily guarantee consumer adoption of GAICCS, as clothing customization is not an essential component of clothing purchasing. Although consumers may initially explore GAICCS out of curiosity toward this novel AI technology, this does not guarantee continued use. This points directly to the critical issue of consumers’ behavioral intention toward GAICCS.

However, there is currently no consumer research specifically addressing the acceptance and behavioral intention toward GAICCS. Although previous studies have explored user interaction intentions with AI technologies in fashion, such as AI avatars, AI chatbots, and AI-curated fashion services [7,8,9], these systems are primarily characterized by one-way service-side dominance, where users have limited autonomy. In contrast, GAICCS adopts a user-led co-creation model, fundamentally redefining the consumer’s role. This distinction not only enhances the application potential of GenAI in the fashion industry but also necessitates an updated theoretical framework to capture users’ cognitive construction and emotional engagement during the interaction process. It is worth noting that existing studies have explored consumers’ attitudes and willingness to accept AI-generated customized fashion products, providing a theoretical reference for investigating behavioral intention toward GenAI-supported clothing customization. For example, Sohn et al. [23] found that when AI technology is applied to customized design, consumers show a significantly higher willingness to pay for GenAI fashion products compared to non-GenAI ones, mainly because of the enhanced hedonic value derived from such products. Furthermore, Lee and Kim [24] demonstrated that when consumers are able to take the lead in clothing customization, their active participation enhances perceived authenticity. These findings suggest that in the fashion domain, beyond individual cognitive factors, external elements representing emotional experiences—such as personal interaction experience, psychological (emotional) evaluations, and perceived emotional value—can significantly influence individuals’ evaluations and behavioral intentions toward using GenAI technologies for product customization. These influencing dimensions have also been addressed in studies examining consumer behavior toward AI-based fashion services in China and other Asian regions [19], including Chinese consumers’ attitudes toward AI-curated fashion services [9]. The existence of these studies indicates that the application of AI in the fashion domain has become a research topic with practical relevance in East Asian regions, including China. At the same time, these studies also offer valuable insights, suggesting that social evaluations and regional cultural influences may affect consumers’ willingness to adopt GAICCS.

Previous studies have also emphasized that consumers’ evaluations of AI services often involve complex chains of mechanisms rather than isolated effects between variables. For example, Shin and Yang [9] found that perceived usefulness, perceived ease of use, and perceived enjoyment significantly influence attitudes, which in turn indirectly affect the purchase intention of fashion products recommended by AI services. In addition, perceived usefulness also plays a partial mediating role between perceived ease of use and attitude. These findings indicate that multiple factors may exert their influence through mediating mechanisms rather than solely through direct effects. Therefore, looking further, under the complex context of GAICCS—characterized by its data-driven operations, real-time feedback, and human–AI co-creation features—new challenges emerge. It is thus necessary to examine the multidimensional factors that influence behavioral intention, as well as their underlying interaction mechanisms. It remains a critical yet underexplored issue within an integrated framework whether and how multidimensional factors—such as individual cognition, psychological perception, service evaluation, and social evaluation—affect consumers’ interactive experiences and behavioral intentions, particularly in the context of the Chinese market.

2.2. Technology Acceptance Model

Currently, various theoretical models have been applied in behavioral science studies on AI technologies and services, such as the Technology Acceptance Model (TAM) [8,19], Unified Theory of Acceptance and Use of Technology (UTAUT) [25], Theory of Planned Behavior (TPB) [26], and Uses and Gratifications (U&G) [27]. Among these, TAM has become a mainstream model for explaining user acceptance of AI-interactive services in fashion and shopping domains [9,13,19] (see Table 1). Compared with the above models that contain more fixed variables, TAM has a simple and extensible structure [19]. Previous studies have incorporated external variables such as performance risk, technological innovativeness, and perceived trust into TAM to explain user behavioral mechanisms in the context of AI-based technology services [9,13,28]. This demonstrates that TAM provides a solid yet adaptable structure for accommodating diverse factors influencing consumers’ multidimensional perceptions. Moreover, in AI fashion and AI-enabled shopping contexts, TAM has demonstrated strong explanatory power across different cultural settings, including Western countries [8,13], and China [7,9].

Given the consumer-led design outcomes and real-time interaction features of GAICCS, a flexible theoretical foundation is required to integrate additional multidimensional variables related to individual cognition, service evaluation, and social evaluation. In contrast, the core variable effort expectancy in UTAUT is primarily used to assess the level of effort users perceive is required to use a new technology. Similarly, TPB emphasizes the measurement of subjective norms and perceived behavioral control. However, the evaluative value of these variables becomes less critical in the context of GAICCS, which is characterized by voluntary use and driven by hedonic and exploratory motivations. As a result, models built upon UTAUT or TPB may be constrained by the inclusion of fixed constructs that are not well-suited to this context. Although Innovation Diffusion Theory (IDT) focuses on explaining the diffusion of innovative technologies, its parallel-construct modeling approach [29] lacks the capacity to capture mediating mechanisms when assessing consumers’ behavioral intention toward GAICCS as a highly interactive service system. As such, IDT is often integrated with models like TAM [30]—which enable mediation analysis—to construct a more comprehensive framework. At the same time, although GAICCS differs from general AI fashion services by offering a more consumer-driven form of human–AI interaction, its adoption process still aligns with the core principles of technology acceptance—such as usefulness and ease of use [31]. This process necessitates an initial consideration of the actual usage effectiveness of GAICCS. Therefore, this study adopts TAM as the base model to validate its explanatory power in the context of GenAI in fashion for the following reasons: First, its core variables—perceived usefulness and perceived ease of use—align with consumers’ initial judgments about the effectiveness of using GAICCS. Second, TAM provides a more open framework that allows for the integration of external variables related to emotional and service evaluations. In addition, TAM supports the analysis of mediating mechanisms, making it suitable for exploring the layered psychological mechanisms of consumers in the GAICCS context.

However, as emphasized in the Introduction, the emergence of GAICCS highlights that studying user behavioral intention toward such interactive service systems with design feedback loops requires not only capturing users’ evaluations of technological performance, but also integrating their emotional responses and assessments of service quality, including perceived social influence. The emergence of GAICCS challenges the explanatory power of the relatively rigid and simplistic TAM framework. This is because the original model fails to capture users’ sustained usage behaviors and the impact of complex socio-cultural contexts [32]. Current behavioral studies on AI fashion technologies and services have acknowledged this limitation and begun to incorporate external factors. Adawiyah et al. [28], based on TAM, incorporated innovativeness, and perceived trust, and confirmed that these variables positively impact users’ intentions to adopt AI and AR technology for tailored recommendations. In addition, attitude serves as a mediating variable in the relationship between innovativeness and usage intention. Liang et al. [13] emphasized the influence of performance risk and technology attitudes in the context of fashion AI devices. Additionally, Ruiz-Viñals et al. [32] not only found a direct effect of perceived usefulness on purchase intention, but also demonstrated that attitude toward AI enhances perceived usefulness, which in turn influences purchase intention. These studies have demonstrated that consumers’ intentions to engage with AI services are largely influenced by their assessments of service performance, as well as by individual traits and perceived value. Meanwhile, potential mediating effects also exist in the pathways through which user perceptions influence behavioral intention. The current literature largely focuses on individual traits and/or psychological perceptions and preferences toward AI services, with an emphasis on internal psychological dimensions, while the integration of external variables remains relatively limited. Few studies have attempted to recontextualize TAM to capture the dimensions of psychological perceptions, service effectiveness, and social evaluations that jointly influence consumers’ behavioral intention in more complex human–AI collaborative platforms like GAICCS. To address this research gap, this study not only extends TAM by integrating external variables from different dimensions that align with the interaction features specific to GAICCS, but also synthesizes them into a higher-level conceptual framework to better capture the multilayered mechanisms influencing consumers’ willingness to interact with GAICCS. In addition, it examines the potential interaction effects among these variables in explaining behavioral intention. These factors were not predetermined but were identified through a combination of prior research and insights from industry experts, which informed the selection of the external variables ultimately included in the research model.

Table 1. Selected recent studies applying the TAM to examine behavioral intention toward AI-interactive services in fashion and shopping domains.

Reference	Context	Predictor Variables Beyond the Core Constructs of TAM	Sample & Source	Main Findings
[19]	AI-powered speech recognition solutions for mobile fashion retail environments	Consumer smart experience (CSE)	N = 836; Sri Lanka and India	PU → AAI, PEOU → AAI, PE → AAI, CSE → PI
[7]	AI-powered avatar	Compatibility (COMP), Customization (CUST), Perceived interactivity (PINT), Perceived relative advantage (PRA), Gamer intention to play AI-powered avatar (GINT), Perceived enjoyment (PE)	N = 500; China	PEOU → ATT, ATT → GINT, PE → GINT, CUST → GINT, PINT→ GINT
[8]	AI chatbots for apparel shopping	Optimism (OPT), Innovativeness (INO), Relative advantage (RA), Discomfort (DIS), Insecurity (INS), Complexity (COM)	N = 353; USA	PEOU → ATT, PU → ATT, OPT → PEOU, RA → PEOU, RA→ PU, COM → PEOU, ATT → PI
[9]	AI-curated fashion services	Technological innovativeness (TI), Fashion clothing involvement (FCI)	N = 382; China	PEOU → ATT, PU → ATT, FCI → PEOU, TI → PU, PE → ATT, PU → PI, ATT → PI
[13]	Fashion AI device	Performance risk (PR), Positive technology attitudes (PTA), Fashion involvement (FI)	N = 313; USA	PU → ATT, PEOU→ ATT, PR → ATT, PTA → PI, ATT → PI
[32]	AI in online fashion purchasing	Perceived quality (PQ)	N = 210; Spain	AAI → PI, AAI → PU, PU → PI, PQ → AAI
[28]	AI and AR technology in customized user recommendations	Perceived trust (PT), Perceived control (PC), Perceived security (PS), Innovativeness (IN)	N = 387; Google Form	PEOU → ATT, PU → PS, PS → PT, IN → ATT, IN → UI, PT → UI, ATT → UI, PC → PEOU
[10]	AI in online shopping	Level of knowledge about AI (KAAI), Level of use of AI (UOIA), Exposure to AI (EAI)	N = 1128; Facebook, Instagram and WhatsApp	EAI → PI, UOAI→ PI, PU → PI, PEOU → PI, KAAI → PI
[33]	AI-mediated retail environment	Attitude toward technology (AAT)	N = 392; Sri Lankan	PU → AAI, PEOU →AAI, AAT → AAI, AAT → PI, AAI → PI, AAI × CI →PI
[34]	AI in online retail	Active use (AU), Trust (TRU)	N = 220; Pakistan	PEOU → ATT, PU → ATT, PU → BI, BI → AU

Note: The arrows (→) indicate a statistically significant effect from the predictor to the outcome variable. BI = Behavioral intention, PI = Purchase intention, ATT = Attitude, AAI = Attitudes to AI, PE = Perceived enjoyment, SN = Subjective norm, CI = Consumer innovativeness.

Table 1. Selected recent studies applying the TAM to examine behavioral intention toward AI-interactive services in fashion and shopping domains.

Reference	Context	Predictor Variables Beyond the Core Constructs of TAM	Sample & Source	Main Findings
[19]	AI-powered speech recognition solutions for mobile fashion retail environments	Consumer smart experience (CSE)	N = 836; Sri Lanka and India	PU → AAI, PEOU → AAI, PE → AAI, CSE → PI
[7]	AI-powered avatar	Compatibility (COMP), Customization (CUST), Perceived interactivity (PINT), Perceived relative advantage (PRA), Gamer intention to play AI-powered avatar (GINT), Perceived enjoyment (PE)	N = 500; China	PEOU → ATT, ATT → GINT, PE → GINT, CUST → GINT, PINT→ GINT
[8]	AI chatbots for apparel shopping	Optimism (OPT), Innovativeness (INO), Relative advantage (RA), Discomfort (DIS), Insecurity (INS), Complexity (COM)	N = 353; USA	PEOU → ATT, PU → ATT, OPT → PEOU, RA → PEOU, RA→ PU, COM → PEOU, ATT → PI
[9]	AI-curated fashion services	Technological innovativeness (TI), Fashion clothing involvement (FCI)	N = 382; China	PEOU → ATT, PU → ATT, FCI → PEOU, TI → PU, PE → ATT, PU → PI, ATT → PI
[13]	Fashion AI device	Performance risk (PR), Positive technology attitudes (PTA), Fashion involvement (FI)	N = 313; USA	PU → ATT, PEOU→ ATT, PR → ATT, PTA → PI, ATT → PI
[32]	AI in online fashion purchasing	Perceived quality (PQ)	N = 210; Spain	AAI → PI, AAI → PU, PU → PI, PQ → AAI
[28]	AI and AR technology in customized user recommendations	Perceived trust (PT), Perceived control (PC), Perceived security (PS), Innovativeness (IN)	N = 387; Google Form	PEOU → ATT, PU → PS, PS → PT, IN → ATT, IN → UI, PT → UI, ATT → UI, PC → PEOU
[10]	AI in online shopping	Level of knowledge about AI (KAAI), Level of use of AI (UOIA), Exposure to AI (EAI)	N = 1128; Facebook, Instagram and WhatsApp	EAI → PI, UOAI→ PI, PU → PI, PEOU → PI, KAAI → PI
[33]	AI-mediated retail environment	Attitude toward technology (AAT)	N = 392; Sri Lankan	PU → AAI, PEOU →AAI, AAT → AAI, AAT → PI, AAI → PI, AAI × CI →PI
[34]	AI in online retail	Active use (AU), Trust (TRU)	N = 220; Pakistan	PEOU → ATT, PU → ATT, PU → BI, BI → AU

Note: The arrows (→) indicate a statistically significant effect from the predictor to the outcome variable. BI = Behavioral intention, PI = Purchase intention, ATT = Attitude, AAI = Attitudes to AI, PE = Perceived enjoyment, SN = Subjective norm, CI = Consumer innovativeness.

3. Empirical Part

Considering the systemic complexity of GAICCS as an interactive clothing customization service, this study adopted a mixed-method research strategy, integrating qualitative insights and quantitative analysis to ensure the contextual and theoretical relevance of the selected variables (see Figure 2). First, a qualitative approach combining a literature review, semi-structured interviews was employed to identify key external variables that may influence consumers’ behavioral intention toward GAICCS. Based on these insights, the questionnaire was designed using validated measurement items from existing literature, followed by a pre-test to evaluate its content adequacy, logical consistency, and construct validity. The second phase involved a quantitative survey, in which the relationships between the variables and consumers’ behavioral intention were examined using SEM-ANN approach.

It should be noted that this study was approved by the Research Review Committee of the School of Fashion Design & Engineering at Zhejiang Sci-Tech University on 2 December, 2024 (Approval No. ZSTUFDE2024120201), and was conducted in accordance with the ethical principles of the 1964 Helsinki Declaration and its later amendments. Written informed consent was obtained from all interview participants prior to participation. For the online questionnaire (including both the pre-test and formal test), an electronic consent procedure was implemented: participants were required to read a consent statement and click “Yes” to proceed. All collected data were anonymized, used exclusively for academic purposes.

3.1. Qualitative Research

3.1.1. Variable Identification from Prior Research

This study conducted a literature search using the keywords “Technology Acceptance Model and AI fashion” on the Web of Science platform. To broaden the search scope, an extended search was also performed using “Technology Acceptance Model and AI retail”. Meanwhile, to ensure the timeliness of the search, the scope was limited to articles published between 2020 and 2025. A total of 102 papers were retrieved. After excluding studies that focused solely on AI technology optimization or offered macro-level commentary unrelated to user experience, and after merging and removing duplicates, a final set of 28 relevant articles was obtained. Through a review of these articles, we found that TAM-based research on AI-based fashion and consumer behavior provides empirical support for the predictive power of various categories of external factors. Given the limited number of relevant studies, this research adopts a conceptual synthesis approach rather than frequency-based coding, extracting external variables from the selected literature with particular attention to those that appear repeatedly. This approach allows for a focused yet flexible model design aligned with theoretical precedents. The results indicate that the following dimensions and variables have been repeatedly validated: (1) emotional perception dimension, such as trust in AI services; (2) uncertainties associated with AI services, such as performance risk and insecurity; (3) perceived benefits of using AI services, such as perceived quality and perceived enjoyment; (4) consumer-specific experiences, such as fashion involvement; and (5) social dimensions, such as social influence and social pressure. This finding also provides a logical and empirical starting point for examining the impact of the above-mentioned dimensional variables on behavioral intention toward GAICCS. The effects of personal perceptions, service quality, and social influence are likely to persist and may even intensify in this context. To enhance contextual specificity, these constructs were further validated and refined through semi-structured interviews in the next stage.

3.1.2. Semi-Structured Interviews

To enrich the theoretical foundations and contextual relevance of the key external variables identified through the literature review, this study conducted semi-structured interviews. The aim was to further explore the underlying motivations behind consumers’ behavioral intention toward GAICCS, while validating, refining, and supplementing the variable identification process specific to the GAICCS context. Semi-structured interviews were conducted between 10 December 2024, and 1 February 2025, through both face-to-face and video formats. The interviewees included 20 experts from Shanghai, Hangzhou, and Wuxi, all of whom have years of professional experience in fashion design, AI-driven fashion technology development, or user experience design and research (see Figure 3 for expert details), and have prior experience using GAICCS. Experts from Shanghai and Hangzhou were invited because these two cities are major fashion innovation hubs in China. Wuxi was selected due to the presence of Jiangnan University, which ranks among the top universities nationwide in the fields of fashion design education and user experience design and research. In addition, a snowball sampling method was employed to recruit 35 general consumers, starting from initial participants with experience using GAICCS. This approach enabled access to diverse user groups with GAICCS experience, which was essential for capturing authentic experiential insights. Among the consumers interviewed, 54.3% were male, most were aged 26–34, and 80% had at least a undergraduate degree. The average interview duration was 18 min. The experts provided their perspectives on the present development and anticipated future directions of GAICCS. Like the general participants, they also discussed their motivations for adoption. The semi-structured interviews employed open-ended questions (see Figure 4).

To ensure a deeper understanding of the issues under investigation, follow-up inquiries were made whenever participants’ answers lacked specificity or clarity [35]. With participants’ consent, all sessions were recorded and transcribed for analysis, resulting in approximately 35,000 words of textual data. Thematic analysis [26] was conducted using NVivo 12, following a two-stage coding procedure. First, open coding was used to generate 37 initial labels from the transcribed texts. Then, axial coding was applied to cluster these labels into higher-level themes based on semantic similarity and conceptual relevance. To ensure the reliability of the analysis, two master’s students who were not involved in this study independently reviewed and coded the transcripts. Coding consistency was further validated through expert review: three experts with extensive research experience in the field of AI and fashion examined the coding structure and thematic categorization, and discrepancies were resolved through discussion. All themes were finalized only after consensus was achieved. Thematic clustering yielded seven overarching themes. These themes, derived from the coding of participant perspectives, are closely aligned with the core variables identified in the literature review. For example, Expert 11 stated, “I feel relaxed and delighted when interacting with GAICCS”, reflecting users’ real-life concerns regarding the hedonic aspects of using the system. Based on frequency of mention and relevance, the top four themes emerged from the thematic analysis: perceived enjoyment (mentioned 39 times), trust (33), service quality (30), and social influence (28).

Ultimately, based on the literature-guided thematic framework and the interview-derived coding results, and drawing on expert recommendations, this study integrated four external variables from multiple dimensions into the TAM framework: emotional perception (perceived enjoyment, trust), service evaluation (service quality), and social evaluation (social influence). This study proposes several hypotheses aimed at thoroughly examining consumer behavioral intentions for GAICCS. Given that this study focuses on consumers’ overall emotional responses after actual experience with GAICCS—rather than their initial attitudes prior to exposure—satisfaction is adopted as the mediating variable in place of the traditional attitude construct in TAM [36]. Specifically, while the variable “attitude” reflects users’ overall evaluative stance toward GAICCS and general affective responses [36], satisfaction in the context of this study refers to consumers’ affective feedback and evaluations formed after interacting with GAICCS, based on the comparison between their expectations and the system’s actual performance. Given that the extended TAM framework constructed in this study not only examines consumers’ perceptions of GAICCS’s usefulness and ease of use, but also incorporates emotional perceptions and service evaluation dimensions, satisfaction logically serves as a mediating variable representing post-experience affective responses [37]. In contrast, attitude, as a general affective perception variable, cannot function as a mediator between other perceptual variables and consumers’ behavioral intention. In previous TAM-based studies on AI technology services, there are also cases where the attitude variable was no longer included as a mediating construct in model development, which provides theoretical support for the approach adopted in this study [10]. Given the highly interactive and experiential nature of GAICCS as a relatively novel service, it is particularly important to evaluate users’ affective perceptions formed after actual usage. Therefore, selecting satisfaction as the affective predictor is more practically meaningful. Moreover, this substitution is consistent with the Expectation-Confirmation Model of Information System Continuance (ECM-ISC), which was proposed by Bhattacherjee [36] based on the integration of ECM and TAM, and is applicable to studies on continuance intention of information systems. The ECM-ISC replaces attitude (an overall affective expression) with satisfaction (a post-experience affective evaluation) as the key predictor of continuance intention. Previous empirical studies on users’ post-experience evaluation of new AI technologies have also adopted extended models based on TAM. For example, Yu et al. [38] investigated users’ intention to use an AI painting application through an integrated model combining TAM and ECM, in which satisfaction was used to replace attitude—the core construct of TAM—and was shown to have stronger predictive validity in scenarios involving user–technology interaction. Therefore, this substitution of the variable is not a theoretical omission, but rather an improvement grounded in contextual and theoretical justification, enhancing the relevance and explanatory power of the integrated research model in the GAICCS context.

3.2. Research Model and Hypothesis Development

3.2.1. Social Influence

Social influence (SI) denotes the extent to which users believe that significant others and their broader social context shape their decisions regarding technology usage [39,40]. Study has shown that individuals often rely on friends’ opinions and sociocultural norms to determine whether a technology is acceptable and useful [39]. According to Wang and Chou [41], when individuals experience positive SI, their sense of identification is heightened, leading to a greater perception of the service as valuable. Moreover, multiple empirical studies indicate that SI play crucial roles in shaping users’ willingness to engage with AI technologies, demonstrating strong positive linkages [40,42]. In the early stages of adopting emerging technologies, users tend to place greater importance on societal opinions [40]. Similarly, as an emerging technology, consumers may also be more concerned with social perceptions when initially adopting GAICCS. Accordingly, the following hypothesis is put forward:

H1a. 

SI positively impacts on consumers’ perceived usefulness of GAICCS.

H1b. 

SI positively impacts on consumers’ behavioral intention toward GAICCS.

3.2.2. Perceived Enjoyment

Perceived enjoyment (PE) denotes the degree of users experience pleasure, satisfaction, and positive emotional responses during the use of a novel technology or service [43]. As a key component of intrinsic motivation, it can enhance the explanatory power of TAM regarding behavioral intention. Bouwman et al. [44] based on an extended TAM, found that users’ PE of mobile innovative services is a key influencing factor, positively contributing to both perceived ease of use and perceived usefulness. The study also revealed that greater enjoyment leads to stronger usage intentions. Felea et al. [45] further confirmed that users who find new technologies enjoyable tend to perceive them as more useful, which enhances their intention to adopt such technologies. In addition, prior studies have identified PE as a key antecedent contributing to the development of user trust in applications and systems [43,46]. In the context of GAICCS, this service system enables consumers to refine and visualize their creative ideas for clothing customization through interactive processes. This capability may enhance their PE. As a result, they may be more willing to invest time in using GAICCS for customization tasks, which also reflects their trust in the system. During the process, consumers are also likely to develop positive emotions toward the actual utility of GAICCS, thereby enhancing their behavioral intention. Therefore, the following hypothesis is formulated:

H2a. 

PE is positively associated with consumers’ trust in GAICCS.

H2b. 

PE is positively associated with consumers’ perceived usefulness of GAICCS.

H2c. 

PE is positively associated with consumers’ perceived ease of use of GAICCS.

H2d. 

PE is positively associated with consumers’ behavioral intention toward GAICCS.

3.2.3. Trust in GAICCS

Trust (TRU) reflects the willingness to accept vulnerability based on positive expectations of the behavior of others [47]. Unlike traditional computing technologies that rely on real-time user input, AI technologies possess a certain degree of autonomy [48], often bypassing significant portions of human decision-making processes. As a result, TRU becomes a critical factor in user–AI interactions [49]. At present, TRU has been incorporated into TAM as a key external driving factor for predicting users’ behavioral intentions toward AI technologies [50]. Related studies—such as those on AI and AR technologies in personalized recommendations [28], and AI chatbots for e-commerce [51]—have all emphasized the critical role of TRU in shaping the intention to adopt AI-based services. In addition, Choung et al. [48] and Wang et al. [34], respectively, confirmed that TRU exerts a notable positive influence on users’ perceptions of both the ease of use and the usefulness of AI technologies.

In the context of GAICCS as a human–AI collaborative system with autonomous generative capabilities, trust plays a critical role in shaping consumer perceptions. Although the user interface of GAICCS is publicly accessible, its underlying mechanism for generating clothing designs—like other GenAI technologies—remains a “black-box” process. As such, consumers’ trust in the service may influence their perceived usefulness and strengthen their behavioral intention. Accordingly, we propose the following hypotheses:

H3a. 

TRU positively impacts on consumers’ perceived usefulness of GAICCS.

H3b. 

TRU positively impacts on consumers’ perceived ease of use of GAICCS.

H3c. 

TRU positively impacts on consumers’ behavioral intention toward GAICCS.

3.2.4. Service Quality

At the current stage, AI is capable of providing services in human–machine interaction that approximate human abilities [52]. Current research on AI technology acceptance has already recognized the positive role of SQ in shaping consumer experience [53], and has confirmed that SQ is positively correlated with both cognitive and affective trust [54], as well as perceived usefulness [55]. Ashfaq et al. [56] validated that the SQ of AI-powered service agents positively influences user satisfaction. The research also indicated that high-quality service can stimulate users’ intention to use [57].

Due to its ability to generate personalized clothing solutions and support real-time modifications, GAICCS represents a complex interactive system. As such, its system quality and information quality [53] jointly determine the reliability and responsiveness of the interaction process. If GAICCS can attract consumers to use it, this is essentially because its high-quality service fosters consumer trust that GAICCS can fulfill their requirements. Hence, the present study puts forward the following hypothesis:

H4a. 

SQ is positively linked to consumers’ trust in GAICCS.

H4b. 

SQ is positively linked to consumers’ perceived usefulness of GAICCS.

H4c. 

SQ is positively linked to consumers’ satisfaction with GAICCS.

H4d. 

SQ is positively linked to consumers’ behavioral intention toward GAICCS.

3.2.5. Perceived Usefulness

Perceived usefulness (PU) describes the degree to which users subjectively believe that using a technology or service will improve their task performance [58]. Within the TAM, PU serves as a significant positive predictor of an individual’s behavioral intention toward new technologies [31]. Research shows that PU, as it directly reflects the utility of AI services, exerts a significant and favorable effect on consumer satisfaction [56,59]. Furthermore, within the realm of AI services, Wang et al. [34] have also empirically demonstrated that the stronger users’ PU, the higher their intention to use the service.

The ability of GAICCS to support clothing customization is a prerequisite for consumers’ willingness to use it. Based on the TAM theoretical framework, we hypothesize that once consumers perceive GAICCS as truly effective, their likelihood of adopting and utilizing the service increases significantly. Accordingly, the following hypothesis is suggested:

H5a. 

PU positively influences consumers’ satisfaction with GAICCS.

H5b. 

PU positively influences consumers’ behavioral intention toward GAICCS.

3.2.6. Perceived Ease of Use

Consumers care not only about the functional efficiency of emerging technologies but also about their ease of use compared to the systems they are intended to replace [58]. Therefore, when using TAM to evaluate consumers’ behavioral intentions toward new technology, perceived ease of use (PEOU) is also a core factor. PEOU represents an individual’s assessment of the amount of cognitive and physical effort needed to use AI technologies [34]. Previous studies on AI technology adoption has established a significant positive correlation between PEOU and PU [9,60]. Amin et al. [61] discovered that consumers tend to feel more satisfied when they perceive a technology to be user-friendly and effortless to operate. Research has also shown that PEOU significantly enhances users’ intention to adopt and utilize the technology [60].

PEOU plays a decisive role in the application prospects of GenAI technology [62]. Therefore, if consumers find that GAICCS can be used effortlessly, they are more inclined to develop a favorable intention toward its use. Accordingly, the present study puts forward the following hypothesis:

H6a. 

PEOU is positively linked to consumers’ perceived usefulness of GAICCS.

H6b. 

PEOU is positively linked to consumers’ satisfaction with GAICCS.

H6c. 

PEOU is positively linked to consumers’ behavioral intention toward GAICCS.

3.2.7. Satisfaction

Satisfaction (STF) reflects an individual’s psychological state after comparing the actual performance of a product or service with their expectations and is a key predictor of consumer behavior [54]. In previous studies on AI technology behavioral intentions, STF has been confirmed to have a positive influence on usage intention [56,59]. In the context of GAICCS, consumer satisfaction as users depends not only on the quality of the customized clothing solutions delivered by the system, but also on the reliability and intelligence of the interaction process. When consumers perceive high-quality information technology services provided by GAICCS during the co-creation process, their satisfaction may increase, thereby enhancing their intention to use the system. Therefore, this study makes the following hypothesis:

H7. 

STF plays a crucial role in positively shaping behavioral intention toward adopting GAICCS.

3.2.8. Behavioral Intention

In this study, behavioral intention (BI) specifically refers to consumers’ intention to use GAICCS—namely, the user’s intention to use this technical service system based on their overall evaluation of its technical performance and interaction experience. It does not include other related but distinct concepts, such as recommendation intention or willingness to pay. In the field of information systems, researchers place strong emphasis on BI, as it is one of the key factors that drive individuals to use a technology or service [63]. The strength of BI directly determines the actual usage of the service. Wang et al. [34] defined consumers’ subjective behavioral tendency to adopt AI-based service technologies as BI toward AI service technologies. In this study, BI serves as the dependent variable, capturing the overall impact of consumers’ evaluations of GAICCS in terms of both functionality and user experience.

Based on the above discussion and analysis, this study proposes 17 hypothesized paths. The specific research model and the hypothesized relationships among the variables are illustrated in Figure 5. It is important to note that although satisfaction has been identified in recent related studies as a direct predictor of behavioral intention, its influence mechanism in the context of GAICCS may exhibit indirect or delayed effects due to the novelty of this technological service and users’ limited familiarity with it. Therefore, while this study incorporates satisfaction as a core variable in the model, its role is theorized to be more nuanced than in previous TAM-based research.

3.3. Quantitative Study

3.3.1. Survey Instrument and Pre-Test

This study utilized a standardized questionnaire to collect data. The questionnaire comprised two main sections. The first section gathered fundamental demographic data, such as gender, age. The second part addressed behavioral intention and examined eight key constructs: social influence (SI), perceived enjoyment (PE), trust in GAICCS (TRU), service quality (SQ), perceived ease of use (PEOU), perceived usefulness (PU), satisfaction (STF), and behavioral intention (BI). These variables were measured using items adapted from previous scholarly works. For instance, the PE scale was based on Felea et al. [45], TRU on Wang and Chen [64], while SQ and STF referred to Ashfaq et al. [56]. SI was informed by Jiang et al. [65], PU and PEOU were drawn from Bunea et al. [10]. The measurement items for BI were adapted from Wang et al. [26]. To ensure contextual validity, the measurement items for the above variables were contextually adapted and optimized based on the interactive functions and information system characteristics of GAICCS, in order to reflect consumers’ specific perceptions during interaction with GAICCS (e.g., “interacting with GAICCS is an enjoyable experience”) and the features of GenAI-driven clothing customization services (e.g., “generating diverse design solutions to meet personalized needs”). The original questionnaire was developed in Chinese, based on adapted English items. To ensure semantic equivalence between the Chinese version used in the survey and the English version presented in the paper, a back-translation procedure was conducted by two bilingual graduate students [66]. The two versions were then compared and reviewed by the research team to eliminate inconsistencies. Each construct included between three and five items. Participants provided their responses on a 7-point Likert scale, where 1 indicated “strongly disagree” and 7 represented “strongly agree”. Full operational definitions and measurement items can be found in

Table 2. Measurement items.

Variable	Abbreviation	Definition	Items
Perceived enjoyment (PE)	PE1	Perceived enjoyment refers to consumers’ anticipated feelings of pleasure and positive emotion prior to actually using GAICCS. It highlights the emotional satisfaction they expect to gain during the interactive co-creation process.	I believe that using GAICCS for clothing customization would be an enjoyable experience for me.
	PE2		I expect that interacting with GAICCS during the clothing customization process would be fun.
	PE3		I think I would feel pleasure when using GAICCS to customize clothing.
	PE4		I anticipate that interacting with GAICCS will bring positive emotional experiences.
Trust in GAICCS (TRU)	TRU1	Trust refers to the state in which consumers are willing to rely on GAICCS during the clothing customization process, based on positive expectations of the system’s capability, stability, and data security.	I trust that GAICCS can accurately understand my needs and provide clothing customization services that meet my expectations.
	TRU2		I can trust GAICCS to complete the clothing customization accurately and reliably.
	TRU3		I believe that using GAICCS to customize clothing is secure and reliable, with no risk of information leakage.
Service quality (SQ)	SQ1	Service quality refers to the level of reliable functional support provided by GAICCS in responding to consumer needs, as well as the effectiveness of the resulting design outcomes.	I find that GAICCS generates diverse design solutions that meet my personalized needs.
	SQ2		I believe GAICCS responds to my needs in a timely manner and operates reliably and stably.
	SQ3		I trust that GAICCS can effectively provide personalized customization support based on my specific needs.
Social influence (SI)	SI1	Social influence refers to the extent to which opinions and behaviors from the social environment, including family members, friends, and fashion influencers, affect an individual’s use of GAICCS. It reflects the normative pressure and expectation alignment that consumers experience when adopting this GenAI-based interactive clothing customization system.	I have noticed that AI-customized clothing created via GAICCS is increasingly visible in social and public contexts.
	SI2		I have observed that many people around me use GAICCS to customize unique clothing for themselves.
	SI3		I have already noticed that fashion influencers and opinion leaders have begun to endorse the use of GAICCS.
	SI4		I have observed that the use of GAICCS is becoming more common and accepted in society.
Perceived usefulness of GAICCS (PU)	PU1	Perceived usefulness describes the degree to which consumers personally perceive that using an interactive clothing customization service system supported by GenAI technology can improve the efficiency and effectiveness of their customization process.	I believe GAICCS provides highly valuable support during the clothing customization process.
	PU2		I think GAICCS is an effective interactive system that supports personalized clothing customization services.
	PU3		I believe GAICCS is an efficient system for providing personalized clothing customization services.
	PU4		Using GAICCS enables me to complete clothing customization tasks more efficiently.
Perceived ease of use of GAICCS (PEOU)	PEOU1	Perceived ease of use reflects the degree to which consumers perceive the process of interacting with the GAICCS system and customizing clothing as smooth and straightforward. It indicates the level of physical and cognitive effort they believe is required to complete the self-directed customization.	I find that the various operational features of GAICCS, such as the ability to modify design plans through image editing, are easy to use.
	PEOU2		When using GAICCS, I feel that customizing clothing does not require much mental effort, even when many options are provided.
	PEOU3		I believe GAICCS reduces the complexity of the clothing customization process by providing clear and supportive design options.
	PEOU4		It is easy for me to understand how to use the various operation options of GAICCS.
Satisfaction (STF)	STF1	Satisfaction refers to consumers’ emotional responses after interacting with GAICCS. These responses are based on a comparison between their prior expectations and the system’s actual performance during the customization process.	I really enjoy the experience of interacting with GAICCS.
	STF2		I feel that the actual experience of using GAICCS met or even exceeded my expectations.
	STF3		I believe that choosing to use GAICCS was a satisfying decision.
	STF4		I am satisfied with the design solutions that GAICCS generated for my customization needs.
Behavioral intention to use GAICCS (BI)	BI1	Behavioral intention to use GAICCS specifically refers to the degree of consumers’ willingness to embrace and use the GenAI-based interactive service system, namely GAICCS.	I intend to use GAICCS to customize and purchase personal clothing in the near future.
	BI2		I am likely to recommend GAICCS to others for clothing customization.
	BI3		I am willing to spend more money on clothing that is customized through GAICCS.
	BI4		I prefer fashion brands that offer GAICCS as a customization option.
	BI5		When making purchase decisions, I would prioritize brands that provide GAICCS.

. The definitions and item formulations of these variables were contextually adapted to the specific features and service content of GAICCS.

Given the novelty of GAICCS, a pre-test was carried out before the formal survey. Before the pre-test, three fashion design graduate students were invited to assess the readability and clarity of the questionnaire items. Based on their feedback, redundant items—such as “GAICCS provides a user-friendly and functionally clear interface to communicate my needs”—were removed. After confirming that the questionnaire was clearly understood, it was distributed to 70 undergraduate and graduate students majoring in fashion design at Zhejiang Sci-Tech University and the China Academy of Art. The pre-test was conducted from February 10 to 2 March 2025. After excluding incomplete responses, 63 valid questionnaires were collected, yielding a 90.0% completion rate. Reliability testing of the pilot data indicated strong internal consistency across constructs, with an overall Cronbach’s α of 0.928 and individual construct values ranging from 0.826 to 0.913. These results suggest that the newly developed scale demonstrates satisfactory reliability and provides strong support for evaluating user responses within an interactive system framework.

3.3.2. Sample and Data Collection

A convenience sampling method was employed in the formal survey to ensure the collection of an adequate sample size within a limited timeframe. The questionnaire was created on the Creadmo platform and distributed online to members of AI technology sections in fashion-related forums, in order to ensure sample diversity. The survey was conducted between 6 March 2025, and 15 May 2025. Before completing the questionnaire, each participant accessed an AI-based clothing customization website (https://www.tribute-brand.com/odds/generator (accessed on 2 August 2025)). Following the website’s operational instructions, they completed a self-directed clothing customization user experience. Subsequently, participants were required to review a case introduction document about GAICCS, which outlined the full process from initial clothing generation to subsequent modification and refinement. After completing the experience and reviewing the GAICCS case, respondents proceeded to fill out the questionnaire. To encourage participation and compensate for the time and effort involved, respondents who completed the survey received a small monetary reward. A total of 736 participants completed the questionnaire, with an average completion time of 12 to 18 min. During data cleaning, logically redundant items were used to assess participants’ attention and verify the consistency of their responses. In addition, questionnaires with a completion time of less than 120 s were excluded, as such short durations were deemed insufficient for valid survey completion. In the end, 692 valid responses were retained, resulting in a valid response rate of 94.02%. The final valid sample size in this study far exceeds ten times the number of paths leading to the most complex dependent variable, meeting the minimum sample size requirement for structural equation modeling (SEM) based on the “10 times rule” [67]. The detailed demographic characteristics of the respondents are presented in Figure 6.

4. Data Analysis

Currently, PLS-SEM has been proven to be a highly valuable statistical tool for empirical research on behavioral intentions related to information systems and intelligent systems [67,68]. Moreover, compared to CB-SEM, PLS-SEM is well-suited for validating extended models based on existing theories and for explaining complex models [69], which aligns with the objective of our research. Therefore, this research adopted PLS-SEM. We used the PLS-SEM algorithm in SmartPLS 4.1.0, setting the maximum number of iterations to 3000 and keeping the initial weights at their default values. To assess the statistical significance of the model outcomes, a non-parametric bootstrapping procedure with 5000 resamples was conducted [65,66].

4.1. Common Method Bias and Multicollinearity Assessment

To verify that the sample does not suffer from potential common method bias (CMB), we conducted Harman’s single-factor test [68]. Using the factor analysis tool in SPSS 26.0.0, we conducted an unrotated factor analysis by employing principal component analysis as the extraction technique. The analysis revealed that the first extracted factor explained 36.08% of the total variance, which is lower than the 50% threshold [66], indicating that no single factor dominates the explanation of variance, and thus CMB is not a concern. In addition, the variance inflation factors (VIFs) for all constructs were below the critical value of 3 [69] (see

Table 3. Assessment of measurement model reliability and validity.

Constructs	Coding	Mean	SD	EFA Loading	Factor Loading	Cronbach’s α	CR	AVE	VIF
Perceived enjoyment (PE)	PE1	4.19	1.469	0.769	0.833	0.865	0.865	0.711	1.930
	PE2	4.26	1.470	0.788	0.840				2.045
	PE3	4.32	1.551	0.788	0.838				2.005
	PE4	4.30	1.542	0.805	0.862				2.244
Social influence (SI)	SI1	4.16	1.553	0.813	0.834	0.857	0.860	0.700	2.035
	SI2	4.20	1.587	0.773	0.834				1.965
	SI3	4.05	1.573	0.729	0.841				1.894
	SI4	4.22	1.595	0.777	0.838				1.990
Service quality (SQ)	SQ1	4.23	1.498	0.780	0.854	0.812	0.812	0.726	1.818
	SQ2	4.22	1.598	0.751	0.855				1.771
	SQ3	4.20	1.523	0.747	0.848				1.747
Trust in GAICCS (TRU)	TRU1	4.17	1.456	0.745	0.826	0.805	0.806	0.720	1.615
	TRU2	4.08	1.621	0.774	0.854				1.779
	TRU3	4.21	1.545	0.811	0.865				1.922
Perceived usefulness of GAICCS (PU)	PU1	4.09	1.544	0.707	0.784	0.844	0.846	0.682	1.668
	PU2	4.10	1.631	0.742	0.855				2.082
	PU3	4.18	1.514	0.730	0.830				1.887
	PU4	4.21	1.609	0.732	0.832				1.933
Perceived ease of use of GAICCS (PEOU)	PEOU1	3.93	1.559	0.751	0.833	0.859	0.860	0.703	1.994
	PEOU2	4.15	1.611	0.729	0.858				2.120
	PEOU3	4.23	1.567	0.751	0.842				2.029
	PEOU4	4.15	1.614	0.715	0.821				1.852
Satisfaction (STF)	STF1	4.14	1.584	0.751	0.803	0.833	0.835	0.666	1.713
	STF2	4.26	1.546	0.749	0.837				1.865
	STF3	4.21	1.618	0.748	0.804				1.740
	STF4	4.20	1.507	0.773	0.820				1.827
Behavioral Intention to use GAICCS (BI)	BI1	4.08	1.450	0.734	0.835	0.908	0.908	0.730	2.244
	BI2	4.13	1.494	0.752	0.860				2.523
	BI3	4.12	1.577	0.749	0.855				2.441
	BI4	4.05	1.609	0.759	0.868				2.674
	BI5	4.05	1.462	0.741	0.854				2.431

Note: SD = Standard deviation; EFA = Exploratory factor analysis; CR = Composite reliability; AVE = Average variance extracted.

), confirming that there is no risk of multicollinearity in the dataset.

4.2. Measurement Model Analysis

Cronbach’s α coefficients were employed to evaluate the internal reliability of each measured construct. As presented in Table 3, each construct demonstrated a Cronbach’s α exceeding the recommended minimum of 0.7 [70], indicating high data reliability and good internal consistency across all measurement scales in this study.

Before performing structural equation modeling, an exploratory factor analysis (EFA) was employed to identify the latent factor structure and evaluate the construct validity of the integrated model. The analysis produced a Kaiser–Meyer–Olkin (KMO) value of 0.945 (>0.7) and a significant Bartlett’s test of sphericity (p = 0.000 < 0.05). The results indicate robust correlations among the constructs, validating the suitability of the dataset for further factor analysis. Varimax rotation was applied in the EFA of the 31 questionnaire items. Eight factors were extracted, cumulatively explaining 70.897% of the total variance, corresponding to the eight dimensions proposed in the hypothesized model. All EFA loadings ranged from 0.707 to 0.813 (see Table 3). In addition, the minimum communality value was 0.627 (>0.4). These results indicate that the questionnaire demonstrates strong structural validity.

Subsequently, confirmatory factor analysis (CFA) was conducted to assess the model’s convergent and discriminant validity. The analysis focused on factor loadings, composite reliability (CR), and average variance extracted (AVE). As indicated in Table 3, each construct demonstrated factor loadings above 0.7, CR scores exceeding 0.7, and AVE scores greater than 0.5, all surpassing the recommended thresholds [68]. These findings indicate that the measures are highly correlated with their respective constructs, thereby confirming the model’s satisfactory convergent validity.

To evaluate discriminant validity, the criteria proposed by Fornell and Larcker [70] were applied. As shown in

Table 4. Discriminant validity assessment based on the Fornell–Larcker criterion.

	PE	BI	PEOU	PU	SI	SQ	STF	TRU
PE	0.843
BI	0.405	0.855
PEOU	0.428	0.610	0.839
PU	0.416	0.585	0.499	0.826
SI	0.355	0.502	0.366	0.481	0.837
SQ	0.383	0.466	0.377	0.490	0.430	0.852
STF	0.354	0.435	0.487	0.377	0.339	0.458	0.816
TRU	0.413	0.422	0.432	0.388	0.356	0.395	0.392	0.849

Note: The numbers in bold along the diagonal represent the square root of each construct’s AVE.

, the square root of each construct’s AVE (highlighted in bold along the diagonal) exceeded the corresponding inter-construct correlation coefficients. Additionally, the heterotrait–monotrait (HTMT) ratios ranged between 0.400 and 0.691, remaining well below the recommended cut-off value of 0.85 [71], further confirming satisfactory discriminant validity among the model’s constructs.

4.3. Structural Model Analysis

During the model evaluation phase, three key indicators—R², Q², and SRMR—were used to comprehensively assess the structural model’s explanatory power and overall fit. R² indicates the proportion of variance in the dependent variables explained by the endogenous constructs. As shown in

Table 5. Structural model assessment and hypothesis testing results.

Construct	Hypotheses	Path Analysis	β	p	Support	R2	Q2	SRMR
BI	H1b	SI → BI	0.181	0.000	Yes	0.526	0.340	0.058
	H2d	PE → BI	0.024	0.564	No
	H3C	TRU → BI	0.056	0.145	No
	H4d	SQ → BI	0.093	0.021	Yes
	H5b	PU → BI	0.237	0.000	Yes
	H6c	PEOU → BI	0.332	0.000	Yes
	H7	STF → BI	0.049	0.230	No
STF	H4c	SQ → STF	0.299	0.000	Yes	0.324	0.240
	H5a	PU → STF	0.058	0.241	No
	H6b	PEOU → STF	0.346	0.000	Yes
PU	H1a	SI → PU	0.227	0.000	Yes	0.415	0.352
	H2b	PE → PU	0.115	0.007	Yes
	H3a	TRU → PU	0.059	0.186	No
	H4b	SQ → PU	0.229	0.000	Yes
	H6a	PEOU → PU	0.255	0.000	Yes
PEOU	H2c	PE → PEOU	0.300	0.000	Yes	0.260	0.208
	H3b	TRU → PEOU	0.308	0.000	Yes
TRU	H2a	PE → TRU	0.307	0.000	Yes	0.234	0.230
	H4a	SQ → TRU	0.278	0.000	Yes

Note: β = standardized path coefficient; p = significance level; R² = explained variance; Q² = predictive relevance; SRMR = model fit index; “→” indicates the hypothesized directional path between constructs.

, satisfaction (STF) (R² = 0.324) and behavioral intention (BI) (R² = 0.526) account for 32.4% and 52.6% of the variance in their respective dependent variables. Both values exceed the 26% threshold recommended by Jiang et al. [65], indicating that the model possesses strong explanatory capability. Secondly, Q² was employed to evaluate the model’s ability to accurately predict outcomes for future observations. All endogenous variables in the research model exhibited Q² values greater than zero, indicating strong predictive relevance both within and beyond the sample [72]. In addition, the SRMR value was 0.058, which is below the recommended threshold of 0.08 [73], indicating a good model fit.

In terms of direct effects on BI, PEOU had the strongest impact (β = 0.332, p < 0.001), followed by PU (β = 0.237, p < 0.001) and SI (β = 0.181, p < 0.001), while SQ had the weakest effect (β = 0.093, p < 0.05). It is important to highlight that PE (β = 0.024, p > 0.05) and trust (TRU) (β = 0.056, p > 0.05) showed no statistically significant influence on BI. Additionally, the results showed that STF (β = 0.049, p > 0.05) did not exert a significant effect on BI, implying that higher consumer satisfaction does not automatically translate into a stronger willingness to adopt GenAI-supported clothing customization services (GAICCS). This unexpected finding prompts further reflection on the structural logic of the proposed model. As GAICCS remains an emerging and non-essential fashion shopping service, consumers are more likely to base their behavioral intentions on immediate evaluations of functional performance (i.e., perceived usefulness and ease of use), while affective judgments formed after the experience tend to have a delayed and less prominent impact. This suggests the necessity of context-specific analysis of the role of satisfaction in the adoption of novel technologies.

In terms of direct effects on STF, PEOU showed a strong positive influence (β = 0.346, p < 0.001), and SQ was also positively associated with STF (β = 0.299, p < 0.001). In contrast, PU had a weaker and statistically insignificant effect on STF (β = 0.058, p > 0.05).

In addition, the study confirmed that PE (β = 0.115, p < 0.01), SQ (β = 0.229, p < 0.001), SI (β = 0.227, p < 0.001), and PEOU (β = 0.255, p < 0.001) were all positively associated with PU, with PEOU exerting the strongest positive effect. However, the analysis revealed that TRU (β = 0.059, p > 0.05) did not significantly influence PU.

The study also found that PE (β = 0.300, p < 0.001) and TRU (β = 0.308, p < 0.001) had significant positive effects on PEOU. Meanwhile, PE (β = 0.307, p < 0.001) and SQ (β = 0.278, p < 0.001) showed significant positive relationships with TRU.

In summary, all hypotheses were supported except for H2d, H3a, H3c, H5a, and H7, as shown in Table 5 and illustrated in Figure 7.

4.4. Mediating Effect Test

This study employed the bootstrapping approach recommended by MacKinnon et al. [74], using 5000 resamples to examine the presence of mediating effects. This method does not require the assumption of normal distribution. The mediating effect is deemed significant if the confidence interval excludes zero. The analysis primarily examined the mediating roles of PU, PEOU, and STF in the relationships between PE, TRU, SI, SQ and BI.

The results showed that the 95% confidence intervals for the paths SI → PU → BI (95% CI [0.030, 0.088]), PE → PU → BI (95% CI [0.007, 0.057]), and SQ → PU → BI (95% CI [0.030, 0.088]) did not include zero, indicating that PU mediates the relationships between SI, PE, SQ, and BI. Similarly, the 95% confidence intervals for the paths PE → PEOU → BI (95% CI [0.064, 0.146]) and TRU → PEOU → BI (95% CI [0.066, 0.149]) also excluded zero, confirming that PEOU significantly mediates the effects of PE and TRU on BI.

Additionally, we examined the mediating role of STF in the relationship between PU and BI, and also the mediating effects of PU and STF in the relationship between PEOU and BI. The findings indicated that PU played a significant mediating role in the link between PEOU and BI (95% CI [0.033, 0.097]). However, the confidence intervals for the other potential indirect paths—PU → STF → BI, PEOU → STF → BI, and PEOU → PU → STF → BI—all included zero, indicating that these mediating effects were not significant. Detailed results are displayed in

Table 6. Results of mediating effect.

Mediating Variable	Path	β	SD	p Value	CI 2.5%	CI 97.5%	Indirect Effect or Not
STF	PU → STF → BI	0.003	0.004	0.489	−0.002	0.016	No
	PEOU → STF → BI	0.017	0.014	0.242	−0.010	0.047	No
	SQ → STF → BI	0.015	0.012	0.240	−0.008	0.042	No
	PE → PU → STF → BI	0.000	0.000	0.511	0.000	0.002	No
	PEOU → PU → STF → BI	0.001	0.001	0.496	0.000	0.004	No
	SI → PU → STF → BI	0.001	0.001	0.496	0.000	0.004	No
	TRU → PU → STF → BI	0.000	0.000	0.645	0.000	0.002	No
	SQ → PU → STF → BI	0.001	0.001	0.517	0.000	0.004	No
PU	PE → PU → BI	0.027	0.013	0.030	0.007	0.057	Yes
	PEOU → PU → BI	0.060	0.016	0.000	0.033	0.097	Yes
	TRU → PU → BI	0.014	0.011	0.194	−0.006	0.037	No
	SQ → PU → BI	0.054	0.014	0.000	0.030	0.088	Yes
	SI → PU → BI	0.054	0.015	0.000	0.030	0.088	Yes
PEOU	PE → PEOU → BI	0.100	0.021	0.000	0.064	0.146	Yes
	TRU → PEOU → BI	0.103	0.021	0.000	0.066	0.149	Yes

Note: β = standardized indirect effect, SD = standard deviation, p value = the significance level. CI 2.5% and CI 97.5% denote the lower and upper bounds of the 95% confidence interval, respectively. “→” indicates the hypothesized directional path between constructs.

.

4.5. Artificial Neutral Network (ANN) Analysis

4.5.1. ANN Modeling and Root Mean Square Error Test

SEM can effectively analyze multivariate linear relationships between variables. However, it cannot handle nonlinear or asymmetric dependencies. In contrast, multilayer perceptrons (MLP) in artificial neural networks can effectively model complex nonlinear relationships. Therefore, the combined use of SEM and ANN for hypothesis validation [75] has been widely applied in studies predicting behavioral intentions toward emerging technologies [68,76,77].

Therefore, this study employed 10-fold cross-validation [76] to evaluate the predictive performance of Model A, Model B, Model C, Model D, and Model E within the ANN framework, and calculated the root mean square error (RMSE) following the guidelines proposed by Ooi and Tan [75]. Specifically, the dataset was randomly divided, with 90% used as the training set and the remaining 10% as the validation set. This process was repeated ten times to reduce overfitting. Figure 8 shows the ANN models. The resulting RMSE values for the training and validation phases were as follows: Model A (0.113 and 0.115), Model B (0.134 and 0.135), Model C (0.118 and 0.115), Model D (0.147 and 0.145), and Model E (0.140 and 0.138), as shown in

Table 7. RMSE value of ANN models.

	Model A (Input: PU, SI, SQ, PEOU; Output: BI)		Model B (Input: SQ, PEOU; Output: STF)		Model C (Input: SI, PE, SQ, PEOU; Output: PU)		Model D (Input: PE, TRU; Output: PEOU)		Model E (Input: PE, SQ; Output: TRU)
Neural Network	Training	Testing	Training	Testing	Training	Testing	Training	Testing	Training	Testing
ANN1	0.113	0.121	0.135	0.114	0.118	0.106	0.148	0.137	0.139	0.137
ANN2	0.113	0.103	0.132	0.139	0.116	0.113	0.146	0.138	0.141	0.125
ANN3	0.111	0.127	0.134	0.123	0.124	0.101	0.146	0.144	0.142	0.133
ANN4	0.116	0.102	0.130	0.147	0.114	0.144	0.147	0.139	0.140	0.133
ANN5	0.112	0.129	0.136	0.131	0.119	0.118	0.145	0.141	0.139	0.143
ANN6	0.113	0.103	0.134	0.120	0.116	0.119	0.145	0.142	0.139	0.136
ANN7	0.114	0.121	0.131	0.142	0.115	0.124	0.151	0.144	0.138	0.149
ANN8	0.113	0.112	0.132	0.141	0.120	0.116	0.154	0.157	0.141	0.142
ANN9	0.109	0.141	0.141	0.157	0.117	0.102	0.142	0.168	0.139	0.138
ANN10	0.116	0.094	0.132	0.139	0.118	0.109	0.148	0.141	0.140	0.139
Mean	0.113	0.115	0.134	0.135	0.118	0.115	0.147	0.145	0.140	0.138
Standard Deviation	0.002	0.015	0.003	0.013	0.003	0.013	0.003	0.010	0.001	0.007

. All five ANN models demonstrated satisfactory model fit and prediction accuracy, providing reliable data support for subsequent analysis.

4.5.2. Sensitivity Analysis in ANN

This study conducted sensitivity analysis on the five ANN models. The analysis revealed the predictive power of each independent variable on the dependent variable and ranked their relative importance. In Model A, PEOU had the highest normalized relative importance (99.70%) in predicting BI, followed by PU (85.07%). SI (53.14%) and SQ (34.88%) ranked third and fourth, respectively. In Model B, the standardized relative importance of the predictors was PEOU (100.00%) and SQ (90.14%). In Model C, the standardized relative importance of the predictors for PU was ranked as follows: SQ (96.05%), PEOU (83.66%), SI (73.90%), and PE (55.45%). In Model D, the standardized relative importance was TRU (97.63%) and PE (88.56%). In addition, in Model E, PE (100.00%) showed a higher relative importance for predicting TRU than SQ (82.40%).

4.5.3. Comparative Analysis of SEM-ANN Results

The predictive power of independent variables in ANN was quantified using normalized relative importance, which is conceptually similar to the path coefficients in PLS-SEM. Therefore, this study compared the ranking of independent variables across the ANN and PLS-SEM analysis stages. As shown in

Table 8. Comparative analysis of SEM-ANN results.

	PLS Path	PLS−SEM: Path Coefficient	Normalized Relative Importance	SEM Ranking	ANN Ranking	Remark
Model A	PU−BI	0.237	85.07%	2	2	Match
	SI−BI	0.181	53.14%	3	3	Match
	SQ−BI	0.093	34.88%	4	4	Match
	PEOU−BI	0.332	99.70%	1	1	Match
Model B	SQ−STF	0.299	90.14%	2	2	Match
	PEOU−STF	0.346	100.00%	1	1	Match
Model C	SI−PU	0.227	73.90%	3	3	Match
	PE−PU	0.115	55.45%	4	4	Match
	SQ−PU	0.229	96.05%	2	1
	PEOU−PU	0.255	83.66%	1	2
Model D	PE−PEOU	0.300	88.56%	2	2	Match
	TRU−PEOU	0.308	97.63%	1	1	Match
Model E	PE−TRU	0.307	100.00%	1	1	Match
	SQ−TRU	0.278	82.40%	2	2	Match

, the ranking of independent variables in ANN Models A, B, D, and E is fully consistent with that in PLS-SEM. In Model C, the ranking results of the “SQ → PU” and “PEOU → PU” paths are not fully consistent. This discrepancy may be due to the ANN capturing the complex and non-compensatory effects of SQ and PEOU on PU—effects that are not reflected in the standardized path coefficients of PLS-SEM. Overall, the ANN results confirm the robustness of the empirical findings from the PLS-SEM stage. The SEM-ANN hybrid method adopted in this study demonstrates strong explanatory power and provides solid empirical support for the reliability of the research model.

5. Discussion and Implications

5.1. Comparison with Prior Studies

This study demonstrates that perceived ease of use positively affects both perceived usefulness and satisfaction, aligning with prior TAM-based research on technology adoption [78], and extending these findings to the domain of interactive GenAI service technologies. In contrast, perceived usefulness does not significantly influence satisfaction, challenging earlier research conclusions [56,59] and providing a new direction for future inquiry.

This study also identified that both perceived ease of use and perceived usefulness exert significant positive influences on behavioral intention. This outcome aligns with the findings reported by Pang et al. [60] within the context of AI technology. However, this study found that perceived ease of use exerts a stronger influence on behavioral intention compared to perceived usefulness. This divergent result stands in sharp contrast to previous findings on the intention to use AI technologies [33], offering room for further validation in future research.

A particularly noteworthy finding is that service quality strongly influences both satisfaction and behavioral intention. This finding corroborates the conclusions reached by Ashfaq et al. [56] and Lai and Chen [57], respectively. Moreover, service quality was shown to positively and significantly influence trust, aligning with the results reported in earlier studies [54]. Future studies on AI technology interaction can further explore the integrative role of service quality within the full behavioral pathway “from user perception to attitude and then to behavioral intention”.

Moreover, this study also confirmed the positive relationship between social influence and behavioral intention, which extends the existing findings that social influence serves as a stable predictive factor within the domain of AI technology [40,42]. Building upon previous studies, this research further explores and confirms the positive effect of social influence on perceived usefulness. This finding expands the understanding of the role of social influence at the perceptual mechanism level and offers an academic contribution for future research on social influence.

However, in contrast to previous findings [56,59,66], this study found that an increase in satisfaction does not strengthen Chinese consumers’ behavioral intention toward GenAI-supported clothing customization services (GAICCS). While prior research has explored digital and AI technologies, variations in research settings or participant groups may account for the differing results, limiting the applicability of earlier conclusions. This result offers new theoretical support and academic contribution to the study of satisfaction pathways in the acceptance of AI-driven fashion interaction services.

Beyond the findings on how predictor variables influence behavioral intention and satisfaction, the study also demonstrated that perceived enjoyment significantly enhances positive effect on perceived usefulness, further supporting previous research on emerging technologies and digital services [44,45]. However, those studies did not simultaneously examine the impact of perceived enjoyment on trust—a gap that the study addresses. For the first time, this study confirms the positive impact of perceived enjoyment on perceived usefulness, perceived ease of use, and trust, highlighting its synergistic role across multiple variables and expanding the understanding of affective mechanisms within the TAM framework. However, this study found that Chinese consumers’ perceived enjoyment does not lead to a significant increase in behavioral intention, which contradicts the conclusions of previous research [46]. This discrepancy warrants further discussion in light of the mediating pathways identified in the analysis.

With respect to trust, this study corroborates the conclusion of Wang et al. [34], which suggests that trust has a positive effect on perceived ease of use. In addition, while most studies [28,48] reported a positive relationship between trust and perceived usefulness or behavioral intention, the present research found no significant influence of trust on either variable. This result offers new insights for future investigations into the function of trust.

By extending the core pathways of the TAM framework, this study also examined and clarified several convergences and divergences in variable relationships, thereby providing new empirical evidence and theoretical directions for understanding service acceptance mechanisms in AI-based interactions. These multi-path dynamics reflect the systematic interactions among cognitive, affective, service evaluation, and social factors. This suggests that future research should adopt more network-oriented and context-sensitive models.

5.2. Diminished Impact of Satisfaction on Behavioral Intention

This study revealed that satisfaction did not have a significant direct influence on behavioral intention, which is contrary to previous research findings [66]. This suggests that when consumers interact with GAICCS, their decisions may be influenced by a more complex combination of factors rather than following a simple “satisfaction–behavioral intention” linear path. A possible explanation is that satisfaction, as a variable reflecting emotional evaluations [56], may exert its influence later than variables related to the evaluation of technical service effectiveness when users are faced with novel service systems. At the current stage, GAICCS remains a relatively novel service with limited promotion. Therefore, Chinese consumers tend to focus more on the immediate feedback of the system’s performance—such as ease of use and usefulness—rather than purely emotional evaluations like satisfaction, which are less directly related to actual utility. This phenomenon has also been noted in previous studies with similar research contexts, such as the application of AI technologies in e-commerce, where attitude—also an affective evaluation—was found to have no significant effect on behavioral intention [34]. Simultaneously, like other AI technologies and platforms, GAICCS also involves a more complex cognitive evaluation system, including social influence. Specifically, when users on social platforms and friends around them widely adopt GAICCS, consumers tend to follow the social trend and make conforming decisions [25]. This situation weakens the role of individual satisfaction in short-term decision-making. At the same time, when the overall service quality of GAICCS is high, Chinese consumers may develop a compensatory behavioral pattern. Instead of making direct judgments based solely on how well the system matches their initial expectations, they tend to base their decisions on the overall positive system experience. The outcomes of this research, demonstrating that both social influence and service quality significantly enhance behavioral intention. This result also indicates that, in GenAI-supported interactive systems, behavioral intention is shaped by the combined effects of multiple dimensions rather than determined by a single psychological perception factor. Therefore, the findings of this study do not suggest that satisfaction is unimportant in model construction, but rather indicate that in the context of novel GenAI-supported service systems, the influence of satisfaction on behavioral intention may be delayed or indirect. These findings suggest that, when dealing with interactive GenAI technology services in their early stages of diffusion, future research may consider treating satisfaction as a pure mediating variable rather than a direct predictor of behavioral intention.

5.3. Explanatory Power of TAM in AI-Based Interaction Context

This study offers robust empirical support for the core tenets of the Technology Acceptance Model (TAM) when applied to GenAI-driven interactive system contexts. The findings suggest that perceived usefulness positively influences behavioral intention, consistent with prior TAM-based studies that highlight emphasizes the central role of perceived usefulness in behavioral decision-making [58]. Modularizing the customization process and generating diverse personalized solutions represent key system-level advantages of GAICCS. Such service systems optimize consumers’ clothing customization experience through a GenAI-driven feedback loop mechanism. Therefore, when Chinese consumers perceive the practical value derived from these positive system function experiences, they are more likely to develop favorable behavioral intentions [56], thereby forming a clear cognitive path from system interaction to behavioral intention.

In addition, this study also verified another key conclusion from the classic TAM, namely that perceived ease of use has a direct impact on both perceived usefulness and behavioral intention [60]. The simple interface structure, combined with a smooth process design, reduces users’ operational burden. When consumers interact seamlessly with such service systems, they naturally perceive GAICCS as practically useful, thereby increasing their behavioral intention. On the other hand, the significant result shows that perceived ease of use indirectly influences behavioral intention through a “perception–evaluation–intention” pathway mediated by perceived usefulness. This also supports the core assumption of the TAM [33]. For the same reason, the low complexity of operating GAICCS reduces the usage threshold for Chinese consumers and establishes a perceptual pathway linking its advantages to practical value [9], thereby making it easier for them to accept and adopt the service.

However, the study also revealed some results that differ from the assumptions of TAM. Specifically, perceived ease of use exerted a greater impact on behavioral intention than perceived usefulness, a result that diverges from the fundamental assumptions of the TAM framework [8,60]. When consumers encounter familiar technologies, perceived ease of use is generally no longer considered a key factor influencing technology adoption intentions [50]. However, GAICCS functions as a co-creation platform between AI and humans, where brands encourage users to actively participate in the clothing design process. This higher level of interactivity may lead users to place greater emphasis on ease of use. Meanwhile, GAICCS remains relatively unfamiliar to Chinese consumers. As an interactive service system, the primary concerns still center on learning costs and operational convenience. In addition, as a value-added service introduced by brands, GAICCS is not a necessary part of the clothing consumption process. Therefore, Chinese consumers are more likely to try it out of curiosity and interest, lacking the intrinsic motivation to explore its functions in depth. These reasons help clarify why perceived usefulness fails to exert a significant effect on satisfaction. This result also indicates that users’ evaluation mechanisms may shift when interacting with such interactive systems.

5.4. Perceived Enjoyment and Trust: Weakened Effects on Behavioral Intention

As a core element of affective perception, perceived enjoyment plays a systematic motivational role in Chinese consumers’ interaction with GAICCS. While previous studies have confirmed that perceived enjoyment significantly influences behavioral intentions [44], the findings of this study highlight a system of interrelated constructs. Specifically, this study confirms that perceived enjoyment has a significant positive impact on perceived usefulness, perceived ease of use, and trust, thereby serving as an affective input node within the behavioral intention system. However, the finding that perceived enjoyment has no significant effect on behavioral intention challenges key conclusions from previous research [46].

Additional analysis reveals that perceived ease of use and perceived usefulness exert a more pronounced influence on behavioral intention. This indicates that Chinese consumers’ perceptions of technological effectiveness and operational convenience are key determinants of their behavioral choices, while mere hedonic experience can only foster positive emotions. The mediation analysis results indicate that perceived enjoyment affects behavioral intention indirectly through perceived ease of use or perceived usefulness. This underscores the need for a multi-layered explanatory structure when examining behavioral intention within GenAI-based clothing customization systems. In other words, consumers’ perceived enjoyment reflects emotional acceptance [50], but their actual decision-making tends to be guided more by a rational evaluation of GAICCS’s effectiveness. Therefore, perceived enjoyment serves more as an emotional catalyst within the user acceptance framework, making a systematic contribution.

This study verified that trust positively influences perceived ease of use, confirming the role of trust as a key factor in enhancing perceptions of usability during the interaction process, aligning with the findings reported in earlier research [49]. However, the effect of trust was limited to the early cognitive stage rather than outcome prediction. It did not exhibit a significant influence on either perceived usefulness or behavioral intention, diverging from earlier findings. Trust in new technologies enhances users’ perception of their effectiveness [64]. However, as a market where new technologies are rapidly developing and expanding [58], most Chinese consumers still lack experience with GAICCS. As a result, they tend to focus first on operational aspects such as convenience and stability during use. Therefore, trust in GAICCS primarily enhances Chinese consumers’ perception of the ease of use of this emerging service system, rather than its usefulness, and does not directly determine their behavioral decision.

5.5. Significant External Drivers: Service Quality and Social Influence

This research demonstrates that service quality plays a crucial role in shaping users’ perceptions when interacting with the system. It significantly enhances their trust, perceived usefulness, satisfaction, and behavioral intention by improving cognitive, emotional, evaluative, and behavioral responses. This indicates that delivering high-quality service through GAICCS can enhance consumers’ trust in the system’s capabilities [54]. As Chinese consumers experience smooth interactions, their confidence in the system’s reliability is further strengthened. At the same time, service quality directly affects users’ evaluation of AI technology’s effectiveness [55]. When GAICCS can accurately respond to Chinese consumers’ personalized needs and efficiently generate high-quality customized clothing solutions, it reflects an interactive mechanism that has adapted to external demands. This increases the likelihood of a positive feedback loop in which users recognize its practical value. As a key antecedent influencing consumers’ consumption experience [53], GAICCS, by delivering higher-quality service and better interactive experiences, can effectively enhance Chinese consumer satisfaction and further stimulate their interest in adopting GAICCS.

This study also confirmed that social influence significantly affects both perceived usefulness and behavioral intention. Based on conformity effects and collective will [25], the demonstration effect of others actively adopting GAICCS, along with a sense of recognition within social circles, can enhance Chinese consumers’ evaluations of the usefulness of GAICCS. Social influence affects consumer decision-making not only through direct normative pressure but also by reinforcing behavioral intention through word-of-mouth effects and recommendation-based dissemination mechanisms [41].

5.6. Theoretical and Managerial Implications

5.6.1. Theoretical Implications

This study makes three key contributions to the fields of consumer behavior and technology acceptance. First, it applies the well-established TAM framework to the underexplored domain of GenAI-supported co-creation services. Previous research has primarily focused on AI technologies as experience-enhancing tools that assist consumers during fashion consumption and engagement, while paying limited attention to voluntary-use technologies such as interactive clothing customization services. By applying TAM to investigate Chinese consumers’ behavioral intention toward GAICCS—a user-driven clothing customization service system—this study validates the explanatory power of TAM in the realm of AI-based interactive technologies. This study provides theoretical insights into the core constructs of TAM in analyzing user acceptance within human–AI collaborative systems, indicating that these variables maintain their relevance even in complex interactive environments. Although two external variables in this study—trust and perceived enjoyment—were not found to have a direct impact on behavioral intention, the proposed framework does not substantially extend TAM. Nevertheless, it confirms that the TAM remains robust in GenAI-supported co-creation service systems and lays a foundation for future theoretical integration involving variables such as trust and perceived enjoyment.

Second, this study incorporates a multi-dimensional set of variables, including individual cognition, emotional perception, service evaluation, and social appraisal. It offers a synergistic analytical framework that has been largely overlooked in previous research. The findings further highlight the significant combined effects of service quality and social influence on Chinese consumers’ behavioral intention toward AI systems, providing important insights into consumer decision-making mechanisms from both technological performance and social-psychological perspectives.

Third, this study uncovers new mechanisms through which social influence, perceived enjoyment, and service quality indirectly affect behavioral intention via perceived usefulness, while perceived enjoyment and trust exert indirect effects through perceived ease of use. These findings enrich the discussion on the function of external variables within the TAM framework. They also reveal the multi-layered perceptual structure of users in the context of adopting new technology service systems, illustrating how emotional perception and social evaluation influence behavioral intention through individual cognitive assessment. This offers new insights into the complex psychological mechanisms underlying Chinese consumers’ use of GAICCS.

Fourth, this study employed a hybrid SEM–ANN approach in data analysis to better capture the complexity of Chinese consumer decision-making in the context of GAICCS. SEM was used to confirm the linear causal relationships among variables, while ANN captured potential nonlinear dependencies and ranked the relative importance of predictive variables. This hybrid method enhanced the robustness of behavioral modeling and demonstrated its value in research on human–AI interaction.

Finally, this research advances the study of user behavior in the domain of AI-based interaction technologies, contributing meaningfully to the fields of human–computer interaction, information systems, and user behavior science. In particular, within the context of fashion brands implementing GAICCS, this study clarifies the preliminary mechanisms shaping the decision-making processes and intentions of young, highly educated Chinese consumers in first- and second-tier cities. It offers concrete theoretical references for both brand managers and GAICCS technology developers in understanding the behavioral intentions of this emerging consumer group.

5.6.2. Managerial Implications

This study offers the following recommendations for fashion brands operating in the Chinese market that seek to optimize the functionality and promotion of GAICCS. First, since perceived ease of use and perceived usefulness both exhibit significant positive influences on behavioral intention, these two factors should be prioritized in GAICCS development strategies. Fashion brands targeting Chinese consumers—particularly those catering to young users in first- and second-tier cities—are advised to increase technological investment in GAICCS, streamline the user interface, and eliminate unnecessary visual elements (e.g., removing infrequently used functional modules and adopting a flat-style interface design). They should also employ intuitive icons to visually convey key functions and embed interactive guidance into each operational step, such as introductory tutorials that appear when hovering over function buttons and simulated practice sessions that break down each process with detailed explanations. At the same time, these fashion brands can simultaneously launch short video tutorials on platforms popular among young Chinese consumers, such as Douyin and Bilibili, or introduce interactive mini-programs on WeChat. These initiatives would enable consumers to become familiar with the GAICCS workflow during fragmented periods of free time. In addition, brands can leverage users’ personal usage data to anticipate consumers’ usage patterns and design intentions. Based on this, they may incorporate quick-access features such as “reuse previous color scheme” or “match with last generated design,” thereby reducing Chinese consumers’ decision-making time. Meanwhile, real-time operational suggestions can be provided to lower the learning cost and enhance consumers’ confidence in utilizing the service. These improvement measures will address issues within the human–computer interaction subsystem and effectively reduce obstacles across the entire information service architecture. At the same time, given that GenAI is still a relatively new concept for Chinese fashion consumers, fashion brands should focus on optimizing generative models (e.g., enhancing style transfer capabilities for clothing styles and patterns, and expanding the model’s ability to learn and recognize diverse fashion styles) and developing intelligent “clothing customization” algorithms that enable fine-grained classification and flexible combination of design elements such as silhouettes and colors. These efforts aim to strengthen the computational intelligence layer of the service system, thereby improving both the efficiency and creative quality of customization and alleviating Chinese consumers’ concerns about the performance effectiveness of GAICCS. Additionally, introducing co-creation mechanisms—such as real-time editing of generated clothing designs and customizable design workflows—may further enhance consumers’ perception of the service’s practical value. This approach establishes a complete system feedback loop that spans from creative input to algorithmic processing, and ultimately to solution generation and adjustment. In addition, fashion brands should actively highlight the performance advantages of GAICCS over traditional customization methods—such as intelligent recommendations, automated design optimization, and trend-based searches—through comparative displays and user case studies.

Secondly, the finding that social influence significantly affects consumers’ behavioral intention highlights the importance of social effects in the promotion and marketing of GAICCS. While enhancing the practical functionality of GAICCS, clothing brands can also organize experiential activities and invite fashion opinion leaders—such as fashion content creators and commentators active on platforms like Weibo, Xiaohongshu, and POIZON, including fashion video bloggers and verified industry professionals—to try out the service, openly gather their feedback, and promptly optimize the system based on their suggestions. Invite such fashion opinion leaders—who hold discourse power in the fashion industry—to share positive reviews of GAICCS on social media and showcase exemplary applications of clothing customization. In addition, fashion brands targeting young, cognitively engaged Chinese consumers may consider commissioning third parties to establish online communities where users can share their experiences with GAICCS and vote for outstanding creative cases. Periodic user-generated content (UGC) challenges—such as “Customize with Me”—can also be launched to create high-engagement topics and promote the viral spread of GAICCS across digital platforms. Endorsements based on favorable user experiences can help build trust and goodwill toward GAICCS among a broader audience. These initiatives can facilitate the development of the social influence subsystem within the GAICCS adoption ecosystem.

Thirdly, this study also found that service quality positively influences both consumer satisfaction and behavioral intention toward GAICCS, indicating that fashion brands targeting the Chinese market should focus the development and optimization of GAICCS on enhancing its information quality and system interaction level. In terms of system optimization, brands should establish an elastic expansion mechanism for cloud servers that store user data, in order to reduce the likelihood of system crashes and generation errors, thereby improving the stability and availability of the service system. Meanwhile, a “feedback and response” feature should be implemented to allow users to report errors, provide satisfaction feedback (using a 1–10 rating scale) on the generated results, and access a “regenerate” option to support design optimization. Additionally, the system should enable users to inquire about operational issues, thereby enhancing consumers’ confidence in using GAICCS. To meet the diverse style preferences and aesthetic expectations of Chinese consumers, fashion brands should continuously expand the underlying database of GAICCS. At the same time, they should leverage the experience of human designers to enhance AI training, improve the customization algorithm’s flexibility in handling diverse consumer demands. By integrating an expert knowledge base—comprising real-world solutions to consumer needs—into the underlying model in advance, the AI system can better understand consumer requirements and ultimately deliver truly personalized solutions for each individual. Moreover, in this intelligent service system, a joint decision-making mechanism involving both AI and human designers is introduced. Specifically, a “AI + designer” hybrid support function can be added to the interface, allowing consumers to initiate an online conversation with a human designer when they find the AI-generated results unsatisfactory. This enables them to receive timely and high-quality design assistance. This helps establish a flexible architecture within the service ecosystem, where the AI system, user operations, and interactive services can work in coordination.

Additionally, in this study, trust did not have a direct effect on behavioral intention. This may be attributed to Chinese consumers’ concerns about the controllability and transparency of GAICCS, stemming from its “black-box” nature. However, this finding also suggests that for fashion brands targeting young and highly educated Chinese consumers, it is necessary to enhance the transparency of the service system. This can be achieved by disclosing the basic generative principles of GAICCS—for example, using visual charts to intuitively explain the design generation rules—providing transparent logs of GenAI decision-making to the greatest extent possible, and clarifying the sources of the data being utilized, whether from fashion brand archives, trend forecasting databases, or users’ personal works. In addition, granting consumers greater control over the customized clothing design outputs can enhance their sense of agency. This can be achieved by opening more operational functions, providing access to model training and database management, and allowing users to adjust model parameters themselves (e.g., number of iterations, control weights, controlnet) for finer control over the final creative outcomes. These optimization strategies may help establish a clearer cognitive pathway from trust development to the formation of usage intention. At the same time, emphasizing the implementation of data security measures as well as the availability and reliability of customization outputs can help Chinese consumers build a sense of secure and dependable trust in GAICCS.

Altogether, these insights provide fashion brands with a multi-level framework for optimizing their service systems. The predictive ranking results based on the artificial neural network assist fashion brands in identifying the relative importance of influencing factors. This enables more informed decisions regarding resource allocation when optimizing GAICCS functions, enhancing user experience, and formulating consumer engagement strategies. They also help reduce consumers’ concerns about GAICCS by offering a comprehensive roadmap for promoting service acceptance and user engagement, as well as establishing differentiated advantages in the rapidly evolving digital marketplace. More importantly, GAICCS can be embedded as a scalable subsystem within the rapidly developing and expanding Chinese digital fashion ecosystem. Although these managerial implications are primarily based on the behavioral patterns and expectations of young, highly educated urban consumers in China, future research should examine the applicability of these implications across broader demographic and geographical contexts.

6. Conclusions and Limitations

This study is the first to comprehensively examine Chinese consumers’ perceptions of GenAI-supported clothing customization services (GAICCS). It investigates the factors influencing their behavioral intentions across four dimensions: individual cognitive evaluations of system functionality (perceived ease of use, perceived usefulness), emotional perceptions (perceived enjoyment, trust), service evaluations (service quality, satisfaction), and social evaluations (social influence). This study addresses two key research questions. First, Research Question 1: Through what mechanisms do core TAM constructs (perceived usefulness, perceived ease of use), satisfaction, and external variables (social influence, perceived enjoyment, trust, and service quality) influence Chinese consumers’ behavioral intention to adopt GAICCS? The findings reveal that perceived ease of use is a critical factor in encouraging consumer adoption of GAICCS, highlighting the behavioral pathway in which Chinese consumers focus more on the ease of using GAICCS to determine their intention to adopt it. Perceived usefulness is also a significant determinant of Chinese consumers’ favorable behavioral intentions toward GAICCS, emphasizing the close connection between the practical value of GAICCS and the willingness to adopt such services. Meanwhile, the significant impact of social influence on behavioral intention expands our understanding of consumer decision-making. Service quality also exhibits a positive association with Chinese consumers’ intention to adopt GAICCS. The limited impact of perceived enjoyment, trust and satisfaction on enhancing behavioral intention also reveals the complexity of consumer behavior. Furthermore, in response to Research Question 2—among these variables, which ones play significant mediating roles in shaping behavioral intention?—the final SEM results establish a structured explanatory pathway for behavioral intention across individual cognitive, emotional, and social dimensions. Specifically, perceived ease of use and perceived usefulness—two core individual cognitive variables—demonstrate significant mediating effects between external variables (social influence, perceived enjoyment, and trust) and behavioral intention. These findings enhance our understanding of user behavioral intention within GenAI-supported clothing customization systems.

Overall, these findings indicate that GAICCS is still in the early adoption stage among Chinese consumers. At this stage, consumers have not yet developed sufficient familiarity and trust toward this interactive service system. As a result, they focus more on the technical effectiveness of GAICCS and external (social) influences, rather than seeking emotional identification or satisfaction. These conclusions not only deepen our understanding of the behavioral motivations behind Chinese consumers’ adoption of GAICCS—particularly among young consumers living in first- and second-tier cities. These conclusions also offer preliminary practical guidance for fashion brands operating in China and targeting this consumer segment in planning their development, optimization, and marketing strategies. Specifically, fashion brands should continue to optimize the simplicity of GAICCS’s user interface and the practicality of its functions, refine the human–AI co-creation mechanisms, expand customization and editing permissions, and emphasize the recommendation influence of fashion opinion leaders. They should also enhance the personalization and diversity of customization outputs, establish real-time feedback channels, and improve the stability and explainability of the GAICCS service system—including increasing the transparency of intelligent generation algorithms. Such efforts can enhance the adaptability of this service system in user-centered environments.

Although this study reveals key factors influencing Chinese consumers’ behavioral intention toward GAICCS, it also has certain limitations, which highlight the need for further in-depth investigation. First, this study employed a convenience sampling method, which allowed the survey to be distributed within a short time frame to individuals interested in the field of AI fashion, thereby facilitating an exploratory investigation of the emerging GAICCS service. However, this approach may also lead to sampling bias. In terms of the results, the survey sample primarily consists of individuals under the age of 40, with a college-level education, and residing in new first-tier and second-tier coastal cities. Although this group may represent both current and potential users of GAICCS, the findings may not be fully generalizable to Chinese consumers of different age groups, educational backgrounds, and those living in cities with varying levels of economic development. Future research should ensure a more balanced sample distribution across various sociodemographic characteristics to uncover differences in behavioral intention toward GAICCS among different consumer groups. These efforts will further enhance the practical implications and strategic recommendations derived from this study, which are currently limited by the demographic bias of the sample.

Moreover, although this study designated satisfaction as a core variable within the research model, the results revealed that satisfaction did not exert a significant direct effect on behavioral intention. Therefore, future research is encouraged to employ longitudinal or experience-based designs to track the evolving influence of consumer satisfaction over time, as users become more familiar with such technologies, thereby offering a clearer understanding of GenAI’s lasting impact on fashion consumption.

This study’s quantitative data were derived from self-reported consumer responses, which could be affected by individual subjective biases. To improve objectivity and validity, future research should consider incorporating actual interaction data with GAICCS.

The current model does not include system experience variables related to users. Future research should also examine factors such as consumers’ prior AI experience and technological knowledge to develop a more holistic understanding of behavioral intention in GAICCS adoption. These insights can help support the effective implementation and advancement of GenAI in the fashion industry.