“Their posts attracted me!”: unveiling the effect of social media influencers’ attributes on relationship quality and purchase intention Open Access

Social media influencers (hereafter SMIs) are people who create personal brands on social media by creating engaging content and attracting large follower communities. Typically, their posts integrate personal attributes or lifestyles with promoted products, enabling them to significantly influence followers' decisions. For this reason, brands actively partner with SMIs for marketing – an approach termed influencer marketing (Syed et al., 2023). Over the past decade, this sector has grown exponentially, from $1.4 billion in 2014 to $24 billion in 2024 (Influencer Marketing Hub, 2025). With projections for 2025 set at $32.55 billion, influencer marketing has evolved into an integrated part of corporate strategies to widen brand reach.

While brands rely on influencers to build consumer relationships, differences in influencer attributes often lead to inconsistent marketing outcomes. For example, a mega influencer has made inappropriate comments during brand endorsements, causing mass unfollowing and harming brand reputation and sales (Qing, 2023). Such a case is not isolated, as global research shows declining trust in influencers (Zniva et al., 2023). The situation raises a critical question: Can brands build high-quality relationships with consumers and drive sustainable marketing performance through systematic screening and optimization of SMIs attributes?

Previous studies have primarily applied influencer credibility and consumer behavioral responses as driving factors of various influencer-follower relationships, including interpersonal, trans-parasocial and parasocial relationships (Pradhan et al., 2023). Comparatively, few studies have explored how SMIs' persona and content attributes impact their influencing effectiveness and follower relationships. Persona refers to the co-constructed public image of an influencer's identity (Leban et al., 2021), while content attributes describe the specific features of shared content (Ki et al., 2020). To our best knowledge, scholars have yet to examine how SMI attributes affect follower relationship quality (hereafter RQ) or the moderating role of swift guanxi between RQ and marketing outcomes. To address these gaps, we aimed to examine RQ as a fundamental mechanism linking SMIs' persona and content attributes to followers' behavioral intentions. The corresponding research questions are as follows:

To answer these questions comprehensively, this study's theoretical framework was constructed using the Stimulus–Organism–Response (SOR) model and relationship quality theory. The SOR model is widely applied in consumer behavior research, providing a robust theoretical foundation for dynamic consumer processes (Gu and Duan, 2024). However, drawing on Bhardwaj et al.’s (2024) findings, no studies have combined the SOR model with relationship quality to analyze consumer behavior – a gap our framework addresses. By integrating and extending these theories into the influencer marketing context, we deepen the understanding of RQ's chain-mediating role and reveal how SMIs' attributes enhance marketing effectiveness through high-quality relationships with followers.

Furthermore, we explore swift guanxi as a moderator that amplifies the transformation of RQ into marketing outcomes, filling a critical research gap. Swift guanxi refers to the rapid interpersonal connections consumers have during online interactions with salespeople or companies (Ou et al., 2014). Unlike traditional digital marketing, the interactive nature of new media has shifted audiences from passive information receivers to active two-way communication participants and requires relationship building (Wang, 2024). Thus, understanding how consumers perceive these relationships and their effect on marketing outcomes is increasingly important. We expect our findings to provide practical guidance for brands to optimize influencer selection, prioritize long-term SMI partnerships and develop strategies to build swift guanxi, promoting sustainable growth in the influencer economy.

The conceptual framework of this study (see Figure 1) is based on the SOR model, developed by Mehrabian and Russell (1974) to explain how environmental stimuli (S) evoke emotional states (O), leading to behavioral responses (R). Previous influencer marketing studies have applied the SOR framework, revealing that influencers’ content attributes and interaction strategies stimulate the influencer–follower relationship and affect followers’ purchase intentions (Aw et al., 2023). Considering extant findings and the current research context, the SOR model is an appropriate theoretical foundation for analyzing how SMIs’ attributes affect follower relationship dynamics and subsequent behaviors.

As shown in Figure 1, the proposed model of this study enriches the SOR model by clarifying how SMIs' attributes, RQ and follower behavioral intentions interact while incorporating swift guanxi's moderating effect on the organism–response path. The inclusion of RQ and the moderating role of swift guanxi is expected to better explain followers’ future intentions, applicable across diverse service settings. In this model, influencers’ persona attributes (inspiration, similarity and enjoyability) and content attributes (visually attractive, interactivity and informativeness) serve as stimuli; RQ is treated as the organism reflecting followers' internal emotional and relational states upon interacting with SMIs; and recommendation and purchase intention represent followers’ behavioral responses. This framework aligns with evidence that high RQ drives consumer behavioral outcomes like eWOM, loyalty and behavioral intention (Ibrahim and Aljarah, 2023; Al Nawas et al., 2021).

RQ is the assessment of the connection between businesses and consumers, which is fundamental in understanding and predicting consumer behavior, particularly in the long term (Franck and Damperat, 2023). RQ refers to “an overall assessment of the strength of a relationship, conceptualized as a composite or multidimensional construct capturing the different but related facets of a relationship” (Palmatier et al., 2006, p. 138). Grounded in the SOR model, this study proposes that RQ (organism) mediates the relationship between SMIs' attributes (stimuli) and consumers' behavioral intentions (responses). Substantial research suggests that brand-consumer RQ consists of three dimensions: trust, satisfaction and commitment (Song et al., 2023; Lo et al., 2017). Since SMIs act as human brands (Ki et al., 2020), we similarly conceptualize RQ as a formative construct encompassing followers’ trust, satisfaction and commitment toward SMIs.

The influence of seller characteristics on RQ has been verified across various contexts, such as supply chain (Qian et al., 2021) and retail banking (Izogo, 2016). In social media marketing, factors enhancing RQ include social media usage, information quantity, information sharing, social interaction and surveillance (Franck and Damperat, 2023; Song et al., 2023). Notwithstanding these findings, empirical research on how SMI characteristics affect RQ remains limited and requires further exploration. As such, we bridge this gap by identifying the specific antecedents of the SMI–follower relationship. In this regard, we put forward that SMIs’ persona and content are critical in forming relational bonds with fans as supported by Ki et al. (2020). Drawing on expert consultations and literature reviews, this study selected three SMI persona attributes (inspiration, similarity and enjoyability) and three SMI content attributes (visual attractiveness, interactivity and informativeness) as key antecedents of RQ.

Persona refers to the co-constructed public image of an influencer's identity (Leban et al., 2021), encompassing the personality traits, goals, attitudes or lifestyle of influencers as perceived by others. Leban et al. (2021) and Ki et al. (2020) further explored SMIs’ personas in marketing strategies and found that certain unique persona attributes of SMIs could attract large followings. Specifically, Ki et al. (2020) identified factors like enjoyability, inspiration and similarity as key persona attributes positively affecting followers' emotional attachment. Nonetheless, empirical evidence on how these persona attributes distinctly influence behavioral intentions via RQ remains scarce to date.

Inspiration is regarded as a “motivational state that compels individuals to bring ideas to fruition” (Oleynick et al., 2014, p. 1) that is triggered by external sources and leads to subsequent action (Beckert and Naderer, 2023). As human brands, SMIs' ability to inspire followers is a key persona-driven attribute enabling them to introduce new ideas, broaden perspectives and motivate novelty exploration (Ki et al., 2020). Inspiring SMIs who embody unique tastes, styles and lifestyles can effectively attract likes, followings and imitation from their audiences (Ki and Kim, 2019). Within influencer marketing, inspiration is a critical channel of interaction between audiences and the information source, making it vital in assessing RQ (Beckert and Naderer, 2023). Therefore, we propose that inspiring SMIs are likely to build high-quality relationships with their followers, as hypothesized below:

Similarity refers to the degree of resemblance in appearance, lifestyle and status (Palmatier et al., 2006). It is a key factor influencing the effectiveness of influencer endorsements (Schouten et al., 2020) and the development of parasocial relationships between SMIs and fans (Ashraf et al., 2023). According to Ki et al. (2020), the more SMIs share similar tastes and preferences with their followers, the more they foster intimacy and fulfill relational needs. We therefore propose that SMIs who have high similarity with their followers are likely to elicit higher RQ. Supported by the literature, the following hypothesis is postulated:

Referring to Liu et al.’s (2020) work, we define enjoyability as the extent to which SMIs convey feelings of pleasure to their followers through posts. Studies have shown that enjoyment enhances consumers' trust, commitment and satisfaction across different settings. For instance, Wong and Haque (2022) demonstrated that website enjoyability strengthens trust; Park and Ko (2022) indicated that augmented reality game enjoyability increases commitment; and Liu et al. (2023) established that travel application enjoyment boosts tourists' satisfaction. Although enjoyability is linked to cultivating relationships in prior research, its impact on RQ through SMIs remains to be empirically validated. As such, we propose the hypothesis below:

SMIs establish their communities by creating and sharing content with their audiences. Influencer marketing relies on social media content, which helps SMIs build relationships with followers and influence consumption decisions using domain-specific expertise (Cheung et al., 2022). The literature indicates that a website's information, visual appeal and interactive features effectively stimulate online customer engagement (Bilro et al., 2018). For SMIs, attractive, informative and interactive content significantly affects consumer attitudes (Ki and Kim, 2019). However, the specific roles of these content attributes in building high-quality SMI–follower relationships remain underexplored, motivating this study's objective.

Visual attractiveness represents consumers' initial and direct perception of an advertisement's design aesthetics, signifying the enjoyment and pleasure derived from visual elements (Yu et al., 2020). Product placement theory suggests that integrating a product into a storyline in a visually attractive manner enhances the endorsement effect (Schouten et al., 2020), as it can capture consumers' attention and motivate their participation in the marketing activities. Correspondingly, in the influencer marketing environment, visually appealing content positively shapes followers' attitudes toward SMIs as opinion leaders (Ki and Kim, 2019). Aw and Chuah (2021) also found that visual attraction to SMIs' posts fosters parasocial connections. Based on this evidence, we postulate the following hypothesis:

Interactivity refers to continuous comments and responses on SMIs' social media, representing two-way communication between SMIs and followers (Jun and Yi, 2020). Studies showed that higher interactivity between brands and consumers leads to better RQ (Song et al., 2023). Similarly, Aw et al. (2023) reported that SMIs' interactivity strengthens parasocial relationships with their follower base. Therefore, it can be assumed that interactive content directly influences the quality of relationships between SMIs and their followers. As such, we hypothesize that.

In this study, informativeness refers to the capability of online sellers to deliver meaningful information that aids consumers in decision-making involving cognitive processing, critical thinking and problem-solving (Al Nawas et al., 2021). Bilro et al. (2018) established informative content as a primary catalyst of user engagement in online experiences. Similarly, Ki and Kim (2019), applying the influence framework of Scheer and Stern (1992), explained how visually attractive, prestigious, expert, interactive and informative content helps influencers establish their authority as opinion leaders, resulting in positive performance outcomes. Building on this, Ki et al. (2020) adopted Human Brand Theory to posit that SMIs build emotional bonds with their followers by offering ideas, relatedness and competence through inspirational, enjoyable and informative content, ultimately enhancing their influence. The above evidence shows that SMIs' informative content plays a key role in engaging followers and forming close relationships, as sufficient information reduces uncertainty. Accordingly, we propose the subsequent hypothesis:

Recommendation intention refers to a customer's positive WOM regarding a brand and its products or services (Ma et al., 2019). While consumers may find it difficult to indicate their repurchase intentions, they can readily express their willingness to recommend, positioning recommendation intention as an indicator of brand loyalty (Kato, 2019). Al Nawas et al. (2021) confirmed that consumers' emotional RQ with e-retailers significantly affects WOM. Additionally, using satisfaction and trust as the two dimensions of RQ, Izogo (2016) found that satisfaction has a direct impact on customers' recommendation intentions toward retail banks, while trust does not. Extrapolating the above evidence to the SMI marketing context, this study postulated that:

Purchase intention is defined as the probability that a consumer is willing or intends to purchase a specific brand in the future (Bhardwaj et al., 2024). In an influencer marketing setting, purchase intention is the process of transforming a follower's brand consumption status from an emerging need into actual purchasing behavior (Alkan and Ulas, 2023). Social media research confirms that RQ positively influences sales performance (Franck and Damperat, 2023) and behavioral intention (Zhou et al., 2023). Given the support from existing literature, the following hypothesis is put forth:

Although previous studies (e.g. Ibrahim and Aljarah, 2023; Zhou et al., 2023) have empirically examined RQ as a mediator between various antecedents and outcomes, further evidence is needed in the influencer marketing context. Some scholars have studied how SMI attributes affect parasocial relationships (e.g. Ashraf et al., 2023) – the one-sided connections between consumers and SMIs (Aw and Chuah, 2021). However, less attention has been given to the role of SMI attributes in fostering long-term RQ. Additionally, SMIs' attributes, RQ and consumer behavioral intentions are often studied separately.

To address these shortcomings, this study integrates these concepts into a single framework to improve our knowledge about their intrinsic relationships. Based on the SOR model, we argue that when consumers are drawn to influencers' persona and content (stimuli), they form long-term relationships with the influencers (organism), leading to more positive behavioral intentions (response). In short, RQ mediates the effect of SMIs' attributes on followers' intentions. Based on this reasoning, the following hypotheses are developed:

In China, guanxi refers to a close, ubiquitous form of interpersonal connection that is crucial for business transactions (Zhou et al., 2023). Swift guanxi in an online context reflects users' rapid relationship-building with e-sellers through mutual understanding, harmony and reciprocal benefits (Ou et al., 2014). Studies show that in online environments, the formation of swift guanxi significantly promotes consumers' WOM and purchase intention (Cheng et al., 2020; Zhou et al., 2023). This is because guanxi enables consumer engagement with sellers (Guo et al., 2021), reinforcing connections and enhancing RQ.

Given the absence of in-person communication and information asymmetry, online shopping carries higher transaction risks, a prevailing issue in the influencer marketing domain as well. In high RQ scenarios, swift guanxi can alleviate risk concerns, accelerate followers' identification with the influencer and strengthen their behavioral intentions. Conversely, when RQ is low, the mutual understanding and reciprocal benefits offered by swift guanxi can partially repair the SMI-follower relationship, thus increasing followers' behavioral intentions.

To date, however, there is still a gap in understanding how swift guanxi moderates the correlation between RQ and behavioral intention in influencer marketing. Since swift guanxi is vital in providing social media users a sense of security and assurance in decision-making, we hypothesize that:

Based on the foregoing discussions, the proposed research framework in Figure 1 illustrates the influence of SMIs' persona and content attributes (inspiration, similarity, enjoyability, visual attractiveness, interactivity and informativeness) on RQ; the effect of RQ on followers' behavioral intentions (recommendation intention and purchase intention); the mediating role of RQ between SMIs' attributes and behavioral intentions; and the moderating role of swift guanxi between RQ and behavioral intentions.

Following a quantitative methodology, this study collected questionnaire data via Sojump, a reputable Chinese online survey platform. Purposive sampling was employed to select participants who had purchased products endorsed by SMIs. Accordingly, two screening questions were presented at the beginning of the questionnaire: “Are you over 18 years old?” and “Have you purchased products recommended or endorsed by SMIs at least once on social media platforms?” Those who did not meet these criteria were not allowed to proceed with the survey. To ensure clarity, we provided a definition of SMIs and asked respondents to name their favorite influencer, helping them better relate to the survey.

The required sample size was derived from both statistical and theoretical perspectives. Statistically, G*Power estimated a minimum of 139 participants for this research, while theoretically, Saunders et al. (2020) prescribe a sample size of 300 to represent a big population. Thus, this study set the sample size at 500. After confirming there were no missing values and removing straight-lining responses, 477 valid responses were retained for further analysis, exceeding the sample size requirements. Table 1 presents the demographic profile of the respondents. The majority were female (63.50%) and aged between 31 and 40 years (49.69%). Most respondents held an undergraduate degree (82.40%) and earned ¥100,001–200,000 annually (48.80%). Regarding shopping behavior, 33.50% reported purchasing twice weekly on social media, while 31.70% purchase more than three times a week.

A survey was developed for this study as the data collection instrument. Measurement items for each construct were adapted from validated scales in the literature. The instruments for SMIs' persona attributes (inspiration, similarity and enjoyability) were derived from Ki et al. (2020), whereas the scales for SMIs' content attributes (visual attractiveness, informativeness and interactivity) were obtained from Ki and Kim (2019). RQ items were adapted from Su et al. (2015) and De Wulf et al. (2001). Recommendation intention was assessed using Al-Ansi et al.’s (2019) three-item scale, while purchase intention was evaluated using items from Ki and Kim (2019). Finally, we drew from the scale developed by Lin et al. (2018) to measure swift guanxi. All variables were operationalized for a 7-point Likert scale (1 = strongly disagree, 7 = strongly agree). Demographic questions were also included to observe respondents' characteristics, namely gender, income, age, education and social media shopping frequency.

Since the survey was conducted in China, the translation of the questionnaire was conducted based on the guide from Ozolins et al.’s (2020) study. After the translation, the final translated questionnaire was then pretested by six experts from academia and industry. Subsequently, a pilot test was conducted on 44 participants, yielding Cronbach's alpha values of 0.659–0.960 for all constructs, confirming the reliability and validity of the scales (Hair et al., 2019).

Before analyzing the data, we first applied Harman's single-factor test and the full collinearity test to rule out common method variance. The results showed that the first factor explained 35.362% of the total model variance, below the 40% threshold. The collinearity analysis also produced variance inflation factor (VIF) values ranging from 1.144 to 2.623, lower than the 3.3 threshold suggested by Kock (2015) and Lim et al. (2022). Overall, we confirmed that the results of this study would not be affected by common method bias.

As for this study, we utilized an integrated PLS-SEM-ANN methodology for statistical analysis. First, PLS-SEM was applied to assess the proposed research framework. This approach is superior in interpreting and predicting behavioral intentions and analyzing complex structural models (Cham et al., 2022; Hair et al., 2019). It is also more flexible with regard to sample size, measurement scales and data distribution assumptions (Hair et al., 2019). Notwithstanding its advantages, PLS-SEM exhibits certain shortcomings when it comes to estimating nonlinear relationships (Wang et al., 2022). To overcome this, ANN can efficiently analyze nonlinear patterns, despite limitations in hypothesis testing (Zhang et al., 2024). Considering the strengths of both methods, this study integrated both ANN and PLS-SEM to develop a more robust and predictive analytical framework – PLS-SEM-ANN.

In the reflective measurement model assessment under PLS-SEM, the lower-order constructs' internal consistency reliability, convergent validity and discriminant validity were evaluated using confirmatory factor analysis. Table 2 shows Cronbach's alpha values (0.706–0.936) and composite reliability values (0.835–0.952), all exceeding the recommended threshold of 0.7, confirming satisfactory reliability (Hair et al., 2010). Next, convergent validity was assessed using factor loadings and average variance extracted (AVE). Table 2 shows item loadings of 0.748–0.918, exceeding the recommended 0.6 (Hair et al., 2010), and AVE values of 0.629–0.812, surpassing the threshold of 0.5 (Hair et al., 2019), which confirm the convergent validity of the constructs. Subsequently, discriminant validity was evaluated using the Fornell–Larcker criterion and the Heterotrait-Monotrait ratio of correlations (HTMT) criterion. As shown along the bolded diagonal columns in Table 3, the square root of each variable's AVE exceeded its correlations with other variables, satisfying the Fornell–Larcker criterion. Likewise, Table 4 indicates HTMT values < 0.90, meeting the requirement. Together, these results supported the model's discriminant validity.

Following Lo et al.’s (2017) research, we operationalized RQ as a higher-order construct comprising commitment, satisfaction and trust (which were part of the reflective measurement model assessment in Section 4.1). RQ was evaluated using the disjoint two-stage approach to ensure its reliability and validity. Table 5 shows that the factor loadings for all indicators of RQ were above the 0.6 benchmark, with both Cronbach's alpha and composite reliability surpassing 0.7, confirming the acceptable reliability of RQ. Its convergent validity was also established, with an AVE value of 0.749 (Table 5). Lastly, the discriminant validity of RQ was assessed using the HTMT and Fornell–Larcker methods, with Table 6 results verifying RQ met both criteria, ensuring satisfactory discriminant validity.

After confirming the measurement model's acceptability, we then assessed the PLS-SEM structural model. To begin, we tested for collinearity, with VIF values ranging from 1.318 to 4.276 (below the threshold of 5.0), confirming no multicollinearity issues (Hair et al., 2010). Subsequently, we used the bootstrapping technique with 5,000 subsamples to test the significance of the structural model relationships (results presented in Table 7). We found that similarity (β = 0.187, p < 0.001), enjoyability (β = 0.104, p < 0.01), visual attractiveness (β = 0.170, p < 0.001), interactivity (β = 0.220, p < 0.001) and informativeness (β = 0.338, p < 0.001) significantly and positively affect RQ, explaining 56% of its variance. However, inspiration (p > 0.05) shows no significant impact on RQ. Hence, H2, H3, H4, H5 and H6 were supported, while H1 was not. Next, as predicted, RQ significantly and positively influences both recommendation intention (β = 0.715, p < 0.001) and purchase intention (β = 0.677, p < 0.001), explaining 46.7 and 31.4% of their variance, respectively. These findings confirmed the acceptance of H7 and H8.

We then assessed the mediating effect of RQ between SMIs' persona and content attributes and followers' behavioral intentions. As reported in Table 8, RQ significantly mediates the effects of similarity (p < 0.001), enjoyability (p < 0.01), visual attractiveness (p < 0.001), interactivity (p < 0.001) and informativeness (p < 0.001) on both recommendation and purchase intention. Therefore, H9–H12 were partially supported, except for the inspiration–RQ–intentions pathways.

To test the moderating effect of swift guanxi, the bootstrapping technique was employed. Table 9 indicates that swift guanxi does not moderate the relationship between RQ and recommendation intention (β = 0.028, p = 0.149), rejecting H13. Nevertheless, it significantly moderates the relationship between RQ and purchase intention (β = 0.107, p = 0.004), supporting H14 with a small effect size (f² = 0.025). To illustrate this effect, simple slope analysis was conducted (Figures 2 and 3), depicting that when followers perceive swift guanxi to be high in their engagement with SMIs, the positive relationship between their RQ and purchase intention becomes stronger.

Compared to traditional regression methods, neural networks are thus superior in prediction accuracy and reliability. We employed SPSS 27 to conduct ANN analysis using the multilayer perceptron training method. A ten-fold cross-validation method was applied to prevent overfitting: 90% of data trained the model, while 10% evaluated its predictive performance as suggested by Almheiri et al. (2024). As illustrated in Figure 4, we analyzed SMIs' similarity, enjoyability, informativeness, visual attractiveness and interactivity. The ANN model's predictive accuracy was assessed via mean values of the root mean squared error (RMSE). Smaller RMSE values, reflecting smaller average differences between predicted and actual values, indicate better model performance. Table 10 shows the RMSE was 0.476 for training and 0.485 for testing, demonstrating high predictive accuracy and satisfactory model fit.

To determine the relative importance of factors, we subsequently conducted a sensitivity analysis ranking predictors by standardized percentage values. Table 11 shows informativeness as the most important predictor (97.81%), followed by visual attractiveness (72.49%), similarity (53.71%), interactivity (52.71%) and enjoyability (35.12%). Finally, we compared the ANN and PLS-SEM results (see Table 12), which highlight informativeness and visual attractiveness as the most important SMI attributes for RQ, with enjoyability being the least important. These findings reveal critical insights into SMI attributes driving high-quality relationships with their followers.

Grounded in the SOR model and RQ theory, this study developed an integrated model to examine how SMIs influence their followers' behavioral intentions by building high-quality relationships, with swift guanxi as the moderator. First, the findings reveal that all three content attributes (visual attractiveness, interactivity and informativeness) and two persona attributes (similarity and enjoyability) of SMIs significantly influence RQ. This scenario is plausible as SMIs build their communities by creating and sharing content, which inevitably involves establishing relational bonds with their followers (Cheung et al., 2022). The outcome from the present study confirms informativeness and visual attractiveness are the two most valuable attributes in enhancing RQ. In other words, SMIs who shared informative and visually appealing content are more likely to have high-quality relationships with their followers. These findings add to existing evidence (Ki and Kim, 2019; Ki et al., 2020) and establish the key drivers of consumer RQ in the influencer and digital marketing domain.

SMIs' similar and enjoyable persona attributes also contribute to high-quality relationships with followers, corresponding with prior findings on the influence of similarity on parasocial relationships (Ashraf et al., 2023) and the influence of enjoyability on RQ (Liu et al., 2023; Park and Ko, 2022). This evidence enriches the evidence of the persona-RQ link in influencer marketing. However, contrary to our expectations, inspiration does not significantly impact the SMI-follower RQ. Followers apparently place less importance on the inspiring persona constructed by SMIs, likely due to growing skepticism about SMI content authenticity as SMIs gain influence (Van Der Heide and Lim, 2016). This uncertainty makes it difficult for followers to build trust, thereby hindering the formation of high-quality relationships.

Second, this study establishes RQ as a mediator between SMIs' attributes and marketing outcomes, supporting previous research (Zhou et al., 2023; Izogo, 2016). Specifically, the findings indicate that RQ is the mechanism through which content attributes (visual attractiveness, interactivity and informativeness) and persona attributes (similarity and enjoyability) stimulate followers' intention to recommend and purchase the endorsed products. However, RQ does not mediate the correlation between inspiration and behavioral intentions. Considering empirical evidence that inspiring SMIs directly affects follower behaviors (e.g. Beckert and Naderer, 2023), it appears that RQ is not important in this process. Nevertheless, future research should explore cognitive pathways linking SMIs' inspiration to follower behaviors.

Third, we confirmed the direct influence of RQ on recommendation and purchase intentions in an influencer marketing setting. Consistent with previous literature, our findings highlight RQ as an important factor for SMIs to achieve marketing goals and encourage follower consumption. Subsequently, our results further demonstrate that swift guanxi, as a significant moderator, can enhance the effect of RQ on purchase intention. Swift guanxi is known to promote consumers' WOM and purchase intentions in online settings (Cheng et al., 2020; Zhou et al., 2023). Adding to this knowledge, this study shows that swift guanxi between SMIs and followers increases the likelihood of high-quality relationships leading to purchase intentions. Meanwhile, low-quality SMI-follower relationships can be strengthened through swift guanxi's mutual understanding and benefit, making RQ still able to influence consumers' purchase decisions. Nevertheless, swift guanxi does not affect the RQ–recommendation intention relationship, indicating that followers' recommendations rely strongly on trust, commitment and satisfaction in their SMI relationships, unaffected by external factors.

The outcomes of this study add significant value to the existing body of knowledge on influencer marketing. First, we expand the SOR Model and Relationship Quality Theory by integrating SMIs' attributes, RQ and followers' behavioral intentions within a comprehensive theoretical framework. In doing so, we deepen the understanding of the dynamic consumer behavior process in the influencer marketing context. Second, existing research (e.g. Aw et al., 2023) indicates that SMIs' persona and content attributes foster various positive outcomes, such as parasocial relationships, flow experiences and satisfaction. Scholars have also identified various antecedents of RQ in the social media context (i.e. Franck and Damperat, 2023; Song et al., 2023), but the existing literature still lacks evidence related to SMIs' attributes as key drivers of RQ in shaping consumer behavior. By addressing this gap, this research introduces a novel perspective to influencer marketing research.

Thirdly, our findings clarify that RQ mediates the influence of SMIs' attributes on follower behavioral intentions. Aligning with the SOR Model by Mehrabian and Russell (1974), SMIs' attributes are stimuli that activate followers' internal perceptions of RQ, subsequently driving their response in the form of behavioral intentions. This process deepens existing knowledge about the joint effects of SMIs' attributes and RQ on followers' behavioral outcomes. Finally, we confirm that swift guanxi acts as a moderator, enhancing the positive impact of RQ on followers' purchase intentions. As one of the first to investigate swift guanxi's moderating role, this study contributes to marketing literature by highlighting its significant impact on consumer behavior in digital marketing.

Our discoveries offer noteworthy practical insights for marketing professionals. With the rapid rise of social media, SMIs now compete fiercely to attract attention and build reputations (Aw et al., 2023). In such a competitive landscape, marketers must carefully and thoroughly evaluate SMIs before selecting them for collaborations. As a marketing core tenet is the development of long-term customer relationships, establishing high-quality follower relationships is a critical success factor in influencer marketing. Therefore, when selecting SMIs, brand marketers should prioritize those who are capable of building strong connections with followers to induce positive responses to endorsements. In this regard, brands can develop a tool to assess potential SMI partners' “relationship value” based on our findings about SMI attributes that enhance RQ. For instance, when collaborating with SMIs, brands should ensure that SMI-created content is both informative and visually appealing, favoring partners with these capabilities. This is because SMI content plays an integral role in sustaining customer-brand relationships.

In addition, SMIs should actively strengthen partner brands' consumer relationships through strategic content creation. Beyond sharing product information, they should create attractive, informative and interactive content while fostering two-way dialogue to build close relationships. Ultimately, well-crafted social media content has the potential to enhance the quality of consumer relationships, translating to recommendations and purchase intentions. Marketers, meanwhile, must recognize that SMIs are not merely attention-grabbing gimmicks, but brand ambassadors or even “friends” whom consumers are eager to interact with (Aw et al., 2023). Brands that value and effectively leverage SMI attributes are better positioned to achieve improved marketing outcomes.

This study found that swift guanxi intensifies the positive influence of RQ on followers' purchase intention. When brands consider collaborating with SMIs, they should evaluate the influencers' ability to create swift guanxi by observing engagement with followers and followers' responses. Partnering with influencers skilled in fostering mutual understanding, harmonious relationships and reciprocal benefits can boost product sales. At the same time, we encourage brands to establish an efficient, user-friendly feedback system to actively listen to and act on fans' feedback, tailoring marketing strategies to meet customers' needs. Lastly, for SMIs, we suggest continuously enhancing presentation skills and using warm, genuine language in live streams and daily posts. Such practice could build emotional connections, foster swift guanxi and strengthen relationships, ultimately boosting sales.

Notwithstanding its practical and theoretical contributions, this research also has limitations that must be improved upon in future research. First, the study was conducted in China, requiring further validation of our findings on SMI attributes influencing RQ in other countries and cultural contexts. Second, since follower behaviors vary across social media platforms and product types may affect marketing strategies (Vrontis et al., 2021), these points should be explored in future research.

Third, despite identifying key SMI attributes in this study, we may not have fully captured all relevant SMI characteristics. Even so, our research model serves as a valuable foundation for more comprehensive, refined and extensive investigations in the future. Lastly, reliance on a single-source survey may limit generalizability due to potential biases (Wang, 2025). Future research could use multiple methods (e.g. experiments, field studies, etc.) to enhance understanding and validity.