In the realm of AIVAs, comprehending user intentions to utilize technology is critical for both its development and broader adoption. In this research, usage intention is defined as the probability that users will adopt a technology based on their favorable evaluation of it. Studies in the field of AIVAs have demonstrated that aesthetic design profoundly impacts decision-making related to behavioral intentions. Notably, research indicates that AIVAs with a ‘cute’ appearance positively influence users’ purchase intention (Yu et al., 2022). Moreover, an appealing design is crucial in enhancing user expectations and increasing tolerance toward service failures (Lv et al., 2021). Thus, a thorough investigation of aesthetic design plays a significant role in improving AIVAs’ market penetration and achieving a distinctive advantage in product design.

Cuteness is generally defined as being appealing in an endearing or adorable manner (Hellén and Sääksjärvi, 2013). Based on previous research, cuteness has been divided into two branches, one is Kindchenschema cuteness (e.g., baby schema), and another is whimsical cuteness. Kindchenschema cuteness is characterized by the design of cartoon-based, more or less human-like graphics, such as big heads, animal ears, and round eyes. Whimsical cuteness refers to an object that people feel is funny (Nenkov and Scott, 2014). Humor, one of the cuteness characteristics, can influence consumers’ evaluations in human-robot interaction (Zhang et al., 2021). In the context of AI applications, the concept of cuteness is predominantly associated with graphical and phonetic features (Lv et al., 2022). Although these two dimensions of cuteness distinctly influence consumer behavior, this study narrows its focus to the design aspects of product appearance, specifically excluding whimsical cuteness. Consequently, our literature review concentrates on exploring the concept of Kindchenschema cuteness, underscoring its relevance and implications within the realm of product design.

Performance expectancy is one of the most important elements in the UTAUT model (Venkatesh et al., 2003). It plays a decisive role in whether users adopt the technology. In the context of AI, Performance expectancy refer to consumers’ expectations of the degree to which reliable and consistent service will be provided (Gursoy et al., 2019). It represents a critical determinant in the adoption of AI devices, encapsulating the belief that these technologies can perform tasks as well as or better than humans. Users typically base their decision to invest time and resources in a technology on these expectations. In the scenario of human-computer interaction, users expect the voice assistant to accurately understand and respond to their voice commands, which includes precise recognition of natural language and timely feedback (Mohd Rahim et al., 2022). Additionally, the breadth and applicability of features are also focal points for users, who expect the AIVAs to go beyond performing basic tasks and excel in advanced functions like smart home management, schedule planning, and information retrieval (Lopatovska et al., 2019). As technology evolves, users’ performance expectancy for AIVAs gradually increases. They are no longer satisfied with basic functionality but expect more intelligent and personalized services. These expectations drive continuous advancements in AIVA technology and provide direction for developers to optimize products and services. Therefore, deeply understanding and fulfilling users’ performance expectancy is crucial to ensuring the success of AIVAs in stimulating usage intention.

Short et al. (1976) conceptualized social presence as the perceptual phenomenon of feeling connected with others through communication media. Social presence refers to how salient the other person is in interaction as well as how salient interpersonal relationships are (Parker et al., 1978). Initially, this theory was used to evaluate the effects of social context on media consumption (Osei-Frimpong and McLean, 2018). In human-computer interaction scenarios, AIVAs could engender a sense of social presence, simulating the feeling of interacting with another human being. This simulation is not merely about the functionality of these devices but also their ability to emit social cues that resonate with users on a personal level, thus fostering a deeper sense of connection and engagement (Kim et al., 2020). Users show a greater willingness to engage with AI technologies that manifest a strong social presence (Yu et al., 2022). This inclination is primarily attributable to the enhanced relational dynamics that a well-designed AIVA can facilitate. By incorporating elements of human-like interaction styles—such as conversational nuances, empathetic responses, and even elements of humor and personality—AIVAs are able to transcend their traditional roles as merely task-oriented assistants and evolve into social companions.

Customer value is a pivotal determinant in consumer behavior, encapsulating consumers’ perceptions, preferences, and evaluations of product attributes, performance, and outcomes (Kim et al., 2011). As defined by Woodruff (1997), customer value encompasses the assessment of product features and the consequential benefits or detriments that align with or impede the attainment of consumer objectives. This multifaceted construct includes dimensions such as social, functional, and emotional value (Lee et al., 2019; Su et al., 2019), in addition to monetary, symbolic, and normative value (Lee et al., 2019), along with expected, perceived, and received value (Ma et al., 2022). Moreover, hedonic and functional values (Boakye et al., 2018), alongside conditional and epistemic values (Hur et al., 2012), further delineate its breadth. Within the realm of human-computer interaction, emphasis has often been placed on hedonic and functional value due to their pronounced effects on consumer engagement and decision-making with AI technologies (Frank et al., 2021; Yin and Qiu, 2021; Yuan et al., 2022; Li et al., 2022b). In the context of human-computer interaction, hedonic value, defined as the emotional gratification consumers derive from engaging with AIVAs, propels interest and curiosity, leading to an enhanced intention to use (Lee et al., 2018; De Kervenoael et al., 2020). Functional value is conceptualized as the utility and effectiveness derived from the use of AIVAs, encompassing aspects such as task efficiency, information accuracy, and overall usefulness in facilitating daily activities (Hu, 2021).

In the field of human-computer interaction, the influence of a product’s aesthetic appeal, particularly cuteness, on user behavior is well-documented. Research has demonstrated that a cute appearance not only enhances tolerance for service failures (Lv et al., 2021) but also significantly encourages customer adoption of AI devices (Lv et al., 2022). Such aesthetic attributes have been shown to profoundly affect consumers’ cognitive experiences (Huang et al., 2021) and increase engagement with AIVAs (Yu et al., 2022). The critical role of AIVAs’ appearance in shaping customer behavior is underscored by findings that designers strategically employ cuteness to drive user adoption (Caudwell et al., 2019; Caudwell and Lacey, 2020). The prevailing consensus among researchers is that the cuteness of a product can mitigate the perceived negatives associated with AI devices, leading to a more forgiving user attitude toward their limitations (Hui and Wenan, 2022). This phenomenon extends to chatbots, where increased cuteness has been linked to improved customer satisfaction, despite potential shortcomings (Zhu et al., 2023). The general positive consumer perception toward cute products (Lu et al., 2021), and the specific role cuteness plays in enhancing the marketability of virtual assistants (Caudwell and Lacey, 2020), further illustrate the potent influence of cuteness on consumer behavior. The ability of cute design elements, both visual and auditory, to stimulate purchase intentions (Shin and Mattila, 2021) reinforces the value of cuteness in product design and marketing strategies. Given this body of evidence indicating the broad and significant impact of cuteness on enhancing user interaction, adoption, and satisfaction with AI applications, we propose the following hypothesis:

Previous studies have elucidated the significant mediating roles of hedonic and functional values in shaping purchase intentions on online platforms, particularly under the influence of AI technology (Yin and Qiu, 2021). These values have been shown to directly impact consumers’ intentions toward AI products (Frank et al., 2021), with the hedonic value of AI contactless services notably affecting perceived service quality (Li et al., 2022b). Despite this, the nuanced interplay between cuteness—aesthetic appeal—and usage intention, mediated by these values, remains underexplored. Thus, this study seeks to bridge this gap by examining how the aesthetic attribute of cuteness in AIVAs influences consumer behavior through the mediation of hedonic and functional values. Cuteness, characterized by features that evoke endearment and affection, holds the potential to alleviate consumers’ negative emotions, thereby augmenting the hedonic value associated with products (Chou et al., 2022). This hedonic value, representing the emotional gratification consumers derive from engaging with AIVAs, propels interest and curiosity, leading to an enhanced intention to use (Lee et al., 2018; De Kervenoael et al., 2020). Notably, the presence of hedonic motivation has been found to influence intentions to use AI-enabled devices significantly (Lin et al., 2020), with prior experiences moderating the extent of this effect (Hu, 2021). In summary, we believe that hedonic value is the emotional and aesthetic satisfaction generated by contact with AIVAs, and is a key mediation in the relationship between cuteness and usage intentions. This mediation is based on the idea that cuteness enhances hedonic value, thereby positively influencing the intention to adopt AIVAs. Therefore, we hypothesize:

The proposition that cuteness can augment functional value stems from the premise that aesthetically pleasing designs can influence user perceptions of a product’s utility. Specifically, cute products often evoke positive emotional responses, which in turn can lead to a more favorable assessment of the product’s functional attributes. This phenomenon is supported by the Emotional Design Theory (Norman, 2005), which suggests that emotionally engaging designs can enhance user experience and perception of product utility. Furthermore, research by Chou et al. (2022) indicates that consumers purchasing cute products with functional value experience less guilt and more satisfaction, suggesting that cuteness enhances the consumption experience by satisfying both hedonic and functional needs. Moreover, empirical studies have demonstrated the importance of functional value in technology adoption and usage intention. For instance, de Kervenoael et al. (2020) highlighted that the perceived usefulness of technology significantly predicts consumers’ intention to adopt it. Similarly, Hu (2021) found that consumers with extensive experience with service robots placed greater emphasis on functional value in their continued interaction and engagement decisions. These findings underscore the critical role of functional benefits in shaping consumer behavior toward technological products. Given the aforementioned arguments, it is posited that the cuteness of AIVAs can positively influence their perceived functional value, thereby impacting usage intention. This mediation effect is predicated on the notion that cuteness, by enhancing positive emotional engagement, can lead consumers to attribute higher utility to AIVAs, ultimately influencing their decision to use these technologies. Therefore, we refine our hypothesis as follows:

Research in human-computer interaction demonstrates the significant influence of social presence on user engagement and intention to use technology (McLean et al., 2021; Hu et al., 2023; Zhu et al., 2023). Specifically, the aesthetic design feature of cuteness in products can elicit social interactions, leading to stronger emotional connections with technology, as evidenced by consumers treating AI devices as companions (Caudwell and Lacey, 2020; Yu et al., 2022). Furthermore, a high level of social presence in robots has been shown to increase consumer acceptance, with cuteness enhancing this effect by triggering sympathy and caring emotions, thereby influencing adoption intention (Kim et al., 2022; Alsaad, 2023; Jin and Youn, 2023). Moreover, social presence is instrumental in elevating the perceived hedonic value of AIVAs, because consumers will perceive the product as a family member (Caudwell and Lacey, 2020), enhancing functional value (Choi et al., 2020; Lin et al., 2020; Moriuchi, 2021). This sequence suggests that cuteness enhances social presence, which in turn amplifies the hedonic value, ultimately fostering a stronger intention to use AIVAs. In the context of AIVAs, cuteness is hypothesized to sequentially enhance users’ social presence and hedonic value perceptions, thereby positively influencing their intentions to use these devices. Specifically, cuteness, by invoking a stronger sense of social connectivity and emotional engagement, elevates the perceived hedonic value (emotional and aesthetic satisfaction) associated with the AIVAs. This enhanced hedonic value, in turn, fosters a greater inclination toward usage. This sequential mediation underscores the pivotal roles of social presence and hedonic value in translating the aesthetic appeal of cuteness into practical user usage. Thus, the following hypothesis is proposed:

This study proposes a nuanced understanding of performance expectancy in the context of AIVAs by examining the influence of cuteness on these expectations. Contrary to the traditional positive correlation between performance expectancy and technology adoption (Andrews et al., 2021; Chuah et al., 2021), this study explores the premise that high levels of cuteness might initially lower performance expectancy due to the Kindchenschema effect, where cuteness induces a more lenient or forgiving attitude toward the capabilities of the object (Lv et al., 2022). However, this initial reduction in performance expectancy does not completely preclude an increase in usage intentions. When users interact with AIVAs and begin to appreciate their capabilities, their performance expectancy may not be high if the AIVAs are perceived as cute. Studies have shown that consumers can tolerate mistakes if the AIVAs are cute (Lv et al., 2021, 2022). This initial perception of cuteness can lower performance expectations but does not necessarily discourage usage. As users continue to interact with their devices, the practical utility of AIVAs can modify these expectations, influencing their decision to adopt the technology. Because the functional value, or the actual utility generated by using AIVAs, acts as a bridge in this relationship, somewhat increasing consumers’ usage intention (Matsuo et al., 2021). When performance expectations align with the actual functionality of AIVAs, the perceived functional value increases, positively impacting usage intent. From these observations, we can deduce the following hypothesis:

Perceived risk plays a pivotal role in shaping consumer behavior toward new technologies, including AIVAs. In the context of this study, perceived risk is defined as the consumer’s assessment of the potential negative outcomes associated with using AIVAs, which may include concerns about privacy, security, or functionality failures (Yu et al., 2022). While the design feature of cuteness in AIVAs can enhance social presence and hedonic value, leading to increased usage intentions, this positive trajectory is contingent upon the level of perceived risk associated with the technology. Research indicates that cuteness, as a design attribute, can elicit positive emotional responses and foster a sense of social connection with technology (Lv et al., 2022; Yu et al., 2022). However, when perceived risk is high, concerns over potential negative outcomes may overshadow the positive perceptions elicited by cuteness, leading to a reduction in the perceived social presence of AIVAs. This diminished sense of social presence, coupled with decreased hedonic value due to escalated risk perceptions, adversely affects consumers’ intentions to use AIVAs. Conversely, in scenarios where perceived risk is low, the endearing qualities of cuteness can fully manifest, enhancing both social presence and hedonic value, and thereby, strengthening usage intentions (Lu et al., 2019; Choi et al., 2020; Lin et al., 2020). Thus, following hypothesis is proposed:

Furthermore, while cuteness can enhance the appeal of AIVAs, elevating both performance expectancy and functional value due to its positive emotional and aesthetic impact, the presence of perceived risk could alter this dynamic. In scenarios of elevated perceived risk, concerns about potential errors, privacy breaches, or the reliability of AIVAs may lead to heightened performance concerns. Such concerns can overshadow the positive perceptions associated with cuteness (Li et al., 2022a), thereby adversely affecting performance expectancy and the perceived functional value of AIVAs (Hussain et al., 2021). This attenuation of performance expectancy and functional value, in turn, dampens the propensity to use these devices. Conversely, when perceived risk is minimized, consumers may display a greater tolerance for potential service failures associated with AIVAs, allowing the positive attributes of cuteness to enhance social presence, performance expectancy, and functional value, ultimately encouraging usage intentions (Trivedi, 2019). Thus, following hypothesis is proposed:

The research model is shown in Figure 1.

First of all, experimental materials were designed. Participants were given experimental assessments of cuteness. In order to exclude participants’ bias toward a particular brand preference and a particular product range, we set up 3 AIVA products without brand identity. The only difference between these three products is in terms of appearance design, and there is no difference in function. The first material (No.1) is a common speaker. The second one (No.2) is a speaker with a pair of eyes. The third (No.3) is a cat-like, which has a pair of cat ears and cute round eyes. We created movies of these three products, which will serve as experimental materials for future research, after determining, via repeated testing, that customers see three distinct products.

Then, the specific process of the experiment consists of three stages: In the first stage, participants will first answer a question about whether they have used an AIVA. If they answer “No,” the system will automatically reject the sample. In the second phase, participants who answered “Yes” were randomly assigned a video about the AIVA. After watching the video, participants were asked to take a questionnaire. In the third stage, participants will answer all structured questions, including questions related to demographic variables, according to their feelings after watching the video. At this stage, the system will automatically extract samples with missing values, samples that fail to pass the screening questions, and samples that answered too fast.

This study developed a high-quality measurement tool, namely the questionnaire. The 23 items in the questionnaire were drawn from prior research, such as Cuteness (C) (Lv et al., 2021), Social Presence (SP) (Mishra et al., 2021), Performance Expectancy (PE) (Gursoy et al., 2019), Hedonic Value (HV) (Ashfaq et al., 2021), Functional Value (FV) (McLean and Osei-Frimpong, 2019), Usage Intention (UI) (Hsiao and Chen, 2018), and Perceived Risk (PR) (Trivedi, 2019). In all questionnaires, a Likert scale of one to five was used [strongly disagree (1) to strongly agree (5)]. Considering the various uncertainties of the participants, which may affect the experimental results, we were assisted by the electronic questionnaire system (Credamo.com). The system can be configured to automatically reject invalid entries, recognize questions, handle missing values, and permit only one entry per ID.

To ensure the validity of the items’ structure and content, multiple specialists were engaged. The initial translation into Chinese was conducted by the principal author, a proficient multilingual Chinese researcher, employing the translation/back-translation technique (Jones et al., 2001). Subsequently, various language research assistants undertook the task of retranslating the Chinese material to verify the translation’s accuracy. Before formally starting the survey, this study referred to Lu et al. (2021)‘s method to conduct a pre-test on the questionnaire. In the formal analysis, we utilized SmartPLS 4 to conduct PLS-SEM (Partial Least Squares Structural Equation Modeling) on the data. This method is especially well-suited for complex models and is robust against small sample sizes and non-normally distributed data.

Before the experiment commenced, we posted the details online and solicited volunteers familiar with AIVA products. Consequently, the sample group is confined to individuals who are interested in AIVA products, inclined to take risks, and receptive to change. Sample collection was conducted through random sampling to mitigate biases and ensure the representativeness of the samples. Concerning privacy and ethics, all participants’ data was obtained with their explicit consent for research and analysis purposes only. Additionally, all authors are Chinese and possess sufficient expertise to assess the rationality of the results. We distributed the questionnaire through social media platforms such as WeChat, Weibo, Moments, and offered certain rewards for completing it. Eventually, 373 samples were procured through random sampling. Following data processing, 284 valid samples were retained. Data processing procedures involved removing null samples, discarding samples answered within an unreasonable timeframe, eliminating samples that did not pass the screening questions, and excluding samples not considered within the sampling scope. Table 1 provides a summary of the sample data.

The data set was calculated by SPSS and Smart-PLS. Initially, we utilized SPSS 27 to conduct an ANOVA analysis to determine how participants perceived the cuteness of 3 types of AIVAs. As demonstrated in Table 2, the p value <0.05, suggested a statistically significant result, which means that participants can distinguish between the 3 types of AIVAs’ cuteness. This is further underscored by the successful differentiation among the three experimental stimuli.

Before analyzing the results, we evaluated the reliability and validity of the measurement tools to ensure that our experimental results could be reproduced. To evaluate whether the model is reliable, first of all, the Outer Loading (Outer Loading >0.7) is adopted to evaluate the reliability of each item. Then, we adopted AVE (AVE > 0.5) to evaluate the reliability between items. Finally, Cronbach’s α (Cronbach’s α > 0.7) and Composite Reliabilities (CR) were adopted to evaluate the overall reliability of each construct. As can be seen from Table 3, our model is reliable, which means that the model has repeatable applicability.

To evaluate whether the model effectively measured the structure, we first adopted the Fornell-Larcker criterion to evaluate the discriminant validity of each construct (Hair, 2017). The logic of the Fornell-Larcker approach is based on the idea that a construct has a greater variance with its associated indicators than with any other construct. As can be seen from Table 4, our model is valid, which means that the model has the applicability of accurate measurement.

Next, we evaluate the structural model. As shown in Figure 2, the R² value of all constructs is greater than 0.19 (Newsted et al., 1998), social presence (0.345), performance expectancy (0.209), hedonic value (0.419), functional value (0.522), usage intention (0.589), indicating a good fit of the overall model.

Stone-Geisser Q² (Stone-Geisser Q² > 0) was calculated using Blindfolding methods to assess the ability of the model to accurately predict the future (Hair et al., 2016). The results were social presence (0.299), performance expectancy (0.155), hedonic value (0.330), functional value (0.380), and usage intention (0.446).

To verify all the assumptions of the model, we resampled the data 3,000 times using bootstrapping. Finally, Table 5 shows the final analysis results. The path coefficient, standard deviation, and p-value of the model were obtained. From the results of the total effect in Table 5, Cuteness - > Usage intention = 0.634 (0.052) *** [0.424, 0.725]. Thus, H1 was supported.

Having measured the total effect, we next measured the 2 serial-mediation effects. Table 5 shows that there are 6 significant mediation paths: Cuteness - > Hedonic value - > Usage intention 0.099 (0.028) *** [0.050, 0.158]; Cuteness - > Functional value - > Usage intention 0.036 (0.015) ** [0.011, 0.070]. Thus, H2a and H2b were supported. Cuteness - > Social presence- > Hedonic value - > Usage intention 0.030 (0.014) ** [0.008, 0.061]; Cuteness - > Performance expectancy - > Functional value - > Usage intention 0.033 (0.012) ** [0.012, 0.059]. Thus, H3a and H3b were supported. For the specific data see Table 5.

Finally, we evaluated the moderating effect. The results show that Perceived risk adjustment on the serial-mediation model has significant statistical significance. Perceived risk significantly moderated the effect of cuteness on social presence [0.169 (0.063) ***]; Perceived risk significantly moderated the effect of cuteness on performance expectancy [0.170 (0.066) ***], which were shown in Table 5. The negative moderating effect (perceived risk) on the serial mediation model is statistically significant. Thus, H4a and H4b were supported. As shown in Figures 3, 4.

As delineated in Figure 3, under conditions of elevated perceived risk, the favorable influence of cuteness on usage intention is considerably diminished via the sequential mediation of social presence and hedonic value. Conversely, when perceived risk is diminished, the positive effect of cuteness on usage intention is markedly amplified through the same sequence of mediators—social presence and hedonic value.

As delineated in Figure 4, under conditions of elevated perceived risk, the favorable influence of cuteness on usage intention is considerably diminished via the sequential mediation of performance expectancy and functional value. Conversely, when perceived risk is diminished, the positive effect of cuteness on usage intention is markedly amplified through the same sequence of mediators—performance expectancy and functional value.

Firstly, in the fields related to human-robot interaction and AI devices, the research related to product appearance mostly focuses on anthropomorphism, while further research on the perception of cuteness is needed. Therefore, this research made up for the deficiency. This study contributes to the literature regarding cuteness and usage intention in VA industry by revealing the impact of cuteness on usage intention.

Secondly, this study contributes to the literature regarding customer value by examining the mediating effect of hedonic value and functional value. Customer value is an inevitable argument in the research of consumer decision-making, and its mediating role has been discussed in the context of AI voice assistant research. However, the mediating role of customer value in the relationship between cuteness and usage intention is rarely discussed. Therefore, this study filled this research gap. In addition, social presence and performance expectancy are important factors affecting usage intention. However, it is not clear that these two factors are affected by cuteness and their impact on perceived customer value. Therefore, we discussed a serial-mediation model. Furthermore, we used samples from the Asian circle to get the opposite result from previous studies on cuteness and performance expectancy.

Thirdly, this study contributes to the literature of perceived risk by revealing the moderating effect of perceived risk. Consumers tend to be wary of objects that are too cute when the perceived risk is high. If a customer perceives a product to be high risk, especially if it appears suspiciously cute, he or she will be less likely to trust it. Customers will pay more attention to the high risk, so they will not devote as much time to evaluation of the product’s appearance, which means that they will not be able to derive the perception of social presence and performance expectancy from the product. Besides, consumers’ perceived risk of service failure is also an important factor affecting the relationship between cuteness and usage intention, which has been rarely discussed in the past in the context of the use of AIVAs. This study showed that perceived risk plays a moderating role in this scenario.

Firstly, the findings of this study offer vital insights and recommendations for AIVAs’ developers and designers. By uncovering consumers’ preference for cute aesthetics, the research lays a theoretical groundwork for optimizing the appearance design of AIVAs. For instance, it suggests that AI featuring endearing elements like cat ears tends to garner more favor from consumers, thus guiding developers in tailoring products to align with consumer tastes.

Secondly, practitioners should heed the significant role of perceived risk in product perception. In product design, companies must not only prioritize aesthetics but also consider consumers’ perceptions of service quality and functionality. Consequently, striking a delicate balance between cuteness and service quality becomes imperative. Moreover, emphasizing product quality, functionality, technological advancements, and innovation in marketing and advertising endeavors can bolster consumer trust and satisfaction.