Visual Modality of Engagement: Conceptualization, Typology of Forms, and Outcomes

Customers proactively engage with firms’ offerings through behavioral manifestations such as brand-related social media posts, influencing other customers in online networks and, consequently, affecting brand value. With the growth of visually oriented social media platforms, interest has increased in understanding customer engagement behavior (CEB) using visual content. In this paper, we build on CEB, image acts, visual content, and communication theories to conceptualize the Visual Modality of Engagement (VME). Using both field and lab studies, we develop a typology of four distinct positive (experiential, evidential) and negative (mocking, dissuasive) forms of VME and offer empirical evidence revealing they induce different brand-related (purchase intentions, brand evaluation) and other customer-related (willingness to imitate, resharing intentions) outcomes. Additional results also reveal outcomes vary by the interplay of social and brand interactions with the various VME forms. The findings of this research offer guidance to content managers for the development of more effective engagement strategies in social media marketing. Graphical Abstract


Introduction
The use of visual content in social media has become an integral part of customer engagement strategies for many brands (Akpinar and Berger 2017;Rietveld et al. 2020). Platforms, such as Instagram and Facebook, facilitate the creation of brand-related posts using visual content as a form of Customer Engagement Behavior (CEB) (Beckers, van Doorn, and Verhoef 2017). Customer engagement behavior captures customers' behavioral manifestations that have a brand or firm focus, beyond purchase, and influencing other customers' decisions and the firm's value (Jaakkola and Alexander 2014;van Doorn et al. 2010).
CEB modality refers to "the different ways in which it can be expressed by customers," which online could be via text, photo, or video (van Doorn et al. 2010, p. 255). Existing research on services, engagement, and communication has identified three modalities: verbal, textual, and visual (e.g., Bakri, Krisjanous, and Richard 2020;Berger and Iyengar 2013;Blackwood 2018;Brodie et al. 2019;Rietveld et al. 2020;van Doorn et al. 2010). Verbal modality is spoken or oral (e.g., face-to-face, phone conversations, and word-of-mouth (WOM) recommendations), while textual modality is written (e.g., texting, tweeting, writing reviews, and e-WOM) (Berger and Iyengar 2013;Brodie et al. 2019;van Doorn et al. 2010). Visual modality, however, is nonverbal, where images substitute for words in online interactions (Bakri, Krisjanous, and Richard 2020). Despite the rapid proliferation of image use in social media (customers upload around 1.3 billion images on Instagram and 350 million images on Facebook daily (Statista.com 2023)), research remains focused on textual analysis, thus, indicating a gap in knowledge precipitating calls for research on visual modes (Babić Rosario, De Valck, and Sotgiu 2020;Hartmann et al. 2021;King, Racherla, and Bush 2014). Therefore, expanding CEB research to include visual modality is important for several reasons.
First, modality is a significant CEB dimension that influences its impact (van Doorn et al. 2010). However, prior engagement research has focused mainly on textual modality, identifying different typologies of textual CEB (e.g., Azer and Alexander 2018;Brodie et al. 2013;Hollebeek and Chen 2014); thus, the potential of images as forms of CEB has remained unexplored. CEB, through visual modality, makes the capture and sharing of intangible offline experiences possible and facilitates the visibility of services, brands, and products online (Akpinar and Berger 2017;Bakri, Krisjanous, and Richard 2020). Visual modality offers richer displays of contextual information, revealing cues both for customers seeking information and for services marketers struggling to understand the nature of subjective and intangible experiences (Ostrom et al. 2021).
Second, visual content, without the elaboration of text, is a powerful and credible vehicle for communication (Kress and van Leeuwen 2006;Suler 2008). According to image act theory, images communicate what users think (cognition) and feel (emotion) about a brand and can convey their intentions (Bakewell 1998;Searle 1976). According to engagement research, intentions are subsequently reflected in CEBs (Brodie et al. 2019).
Third, brain activation used to process words (verbal or written) and images differs (Khateb et al. 2002;Paivio 1986;Townsend and Kahn 2014). Images are processed more quickly, triggering greater emotional processing and cognitive elaboration and leading to higher levels of information retrieval (Blackwood 2018;Kjeldsen 2018;Lee et al. 2015;Lin et al. 2012). Therefore, differences are expected in how customers engage using visual modality which, hitherto, has been unclear. Fourth, the selection and creation of images are inherently subjective (Nicholson-Cole 2005;Suler 2008). However, an understanding of the images customers create to express CEB has not been explored.
Finally, according to image act theory, images communicate behavior that is intended to subsequently evoke behavior in recipients (Bakewell 1998;Searle 1976). The behavioral impact an image prompts depends on the creator's intention (Barinaga 2009) and the recipient's interpretation (Bakewell 1998;Berger and Iyengar 2013;Nicholson-Cole 2005). However, behaviors that images communicate and prompt have been overlooked in prior research which limits our understanding of visual modality and its impacts.
Drawing on the literature on CEB, image acts, communication, and visual content, we define the visual modality of engagement (VME) as customers' behavioral manifestations that have a service, product, or brand focus, using static or dynamic visuals (e.g., images or videos). This paper addresses the research gaps articulated above offering a comprehensive and nuanced understanding of VME using a combination of qualitative (field study) and quantitative (experimental) studies (see Figure 1). The field study reveals specific forms of VME, and the experimental studies evidence their outcomes.
First, this paper contributes to literature by introducing the concept of VME thereby informing and extending the engagement literature, specifically CEB modality. Second, this paper contributes to CEB, communication theory, and visual content research by exploring the behaviors that images intend to communicate and conceptualizing the first VME typology of two positive (experiential, evidential) and two negative (mocking, dissuasive) forms. Third, the paper contributes to visual content literature which had previously been limited to exploring the impacts of specific image characteristics with the first empirical evidence of the behaviors that images prompt in other customers. This paper reveals how VME forms induce different brand and customer-related outcomes and how outcomes vary when moderated by social and brand interactions. These findings present insights for managers seeking to leverage VME and increase customer engagement with their offerings. Finally, informed by the new conceptualizations, this paper offers a future research agenda to direct and craft research on visual modality.

Theoretical Background
Visual Modality of Engagement (VME) Understanding engagement has been an important focus of attention for marketing managers seeking to capture, for example, the enormous opportunities offered by social media Figure 1. Overview of studies. (Beckers, van Doorn, and Verhoef 2017;Harmeling et al. 2017;MSI. 2021). The wider engagement literature articulates a multidimensional concept comprising cognitive and emotional absorption resulting from interactive experiences with a firm or brand, which manifest in CEBs (Brodie et al. 2011). This paper focuses on CEB, representing customers' behavioral manifestations that have a service, product, or brand focus, beyond purchase (van Doorn et al. 2010, p. 254). Behavioral manifestations are behavioral expressions of customer engagement, which can be positive or negative (van Doorn et al. 2010).
Engagement research presents modality as a dimension of CEB referring to the different ways it is expressed by customers (van Doorn et al. 2010). The existing theoretical understanding of CEB has captured exclusively textual and verbal modalities: "voluntary, firm-focused customer behaviors-such as writing reviews or providing WOM recommendations-centered on the focal firm" (Brodie et al. 2019, p. 2). Hence, the focus in CEB research has been on forms such as WOM, e-WOM, referrals, recommendations, online reviews, and blogging (e.g., Azer and Alexander 2020a; Jaakkola and Alexander 2014). However, CEB is not limited to written or oral forms. Online CEB could include text, photos, or video (van Doorn et al. 2010). Extending the extant use of modality as a dimension of CEB is required to encompass visual content. Our view of modality builds on engagement literature, bridging it with communication theory that identifies three modalities: verbal, textual, and visual (e.g., Bakri, Krisjanous, and Richard 2020;Berger and Iyengar 2013;Blackwood 2018;Rietveld et al. 2020).
Among verbal, textual, and visual modalities, there are differences in brain activation (Paivio 1986;Pearce et al. 2018), processing, interpretation, andpreferences (Townsend andKahn 2014), and in motivation to engage using images (Lee et al. 2015;Nicholson-Cole 2005). Different cognitive and emotional needs result in different preferences for verbal, textual, and visual information. Processing visuals requires both cognition and emotion (LeDoux 1996;Parkinson 2022;Sojka and Giese 2006). Selecting images requires judgment of the creator, social context, brand-related experience, and desire to project various aspects of the self (e.g., self-branding) (Bakri, Krisjanous and Richard 2020;Blackwood 2018;Liu, Dzyabura, and Mizik 2020;Nicholson-Cole 2005;Pearce et al. 2018). The images used convey varying expressions toward a brand (Bakewell 1998;Kress and van Leeuwen 2006) and are likely to take a range of forms. For recipients, interpretation of visual content requires both cognitive and affective association that, subsequently, form impressions, attitudes, and behaviors toward the brand (Bakri, Krisjanous, and Richard 2020).
Beyond a hybrid concept of "visual e-WOM," understanding of VME remains limited. Table 1 summarizes existing studies of visual content on social media, revealing characteristics such as color, quality, human face presence (Kwon Jumbum et al., 2022;Li & Xie, 2020), self-representation (Blackwood 2018), and visual congruency (Argyris et al. 2020). In Table 1, we classify existing research analyzing visual content through three main categories and two sub-categories: the main categories are the cognitive processing of visual content by viewers, the emotional appeal of visual content, and the behavior that images intend to communicate. We also capture if images are created by the brand or customer. Table 1 reveals that existing studies capture, for example, visual versus textual content, image plotting based on time or color, and spotting functional and emotional content generated by brands in social media (e.g., Ordenes et al. 2019;Rietveld et al. 2020;Serrano and Ramjaun 2018;Townsend and Kahn 2014). However, the behavior images intend to communicate has been overlooked, which limits our understanding of VME and its impacts. Moreover, studies focus predominately on brand-created visual content and not that generated by customers.

Image Acts in C2C Communication
Image act theory-used here as an enabling theoryencompasses all human-made images and focuses on the behaviors that images communicate and prompt in viewers, which differ between individuals (Bakewell 1998;Barinaga 2009). Like speech acts, image acts convey thoughts, feelings, and intentions, which invoke behaviors in recipients and can flatter, promote, benefit, fight, accuse, denounce, or harm (Bakewell 1998;Searle 1976).
With the rise of social media platforms, sharing images is increasingly central to customer-to-customer (C2C) communications (Akpinar and Berger 2017;Ordenes et al. 2019), representing intended actions and communicating specific messages (Kjeldsen 2018;Kress and van Leeuwen 2006). In social media, image acts range from offering information to directing specific actions (Ordenes et al. 2019). Despite this holistic understanding of the nature of the image, the behaviors images intend to prompt are less well understood.
Image acts are captured in CEB directed at brands, products, or services utilizing visual content to communicate different behavioral manifestations, which could be positive or negative (Brodie et al. 2019). Customers use images in different ways for different reasons (Kress and van Leeuwen 2006). Therefore, understanding how customers use VME allows us to capture customers' behavioral manifestations through visual content and to consider their effect on other customers and the implications for firms (Babić Rosario, De Valck, and Sotgiu 2020).
Thus, analysis of literature on CEB, image acts, communication, and visual content allows us to make the following observations. First, we note that the predominate focus of CEB research relates to textual modality whilst, simultaneously, identifying that CEB is not limited to textual or oral forms. In fact, on social media use of visual forms is becoming the dominant mode of engaging. Theories on visual communication allow us to observe how image creation requires judgment from a creator and projects aspects of the self in a way that textual modality cannot-these projections also take a range of forms which existing literature does not currently capture (see Table 1). Finally, by adopting image acts as an enabling theory we note the centrality of visual modality to C2C communication and how images can both communicate behaviors and stimulate them in others. Having established the importance of visual modality to engagement and communication in general, but the lack of any existing typology for VME, our first study seeks to identify forms of VME through a field study using netnography. VME is undefined in literature; however, drawing on CEB, image acts, communication, and visual content literature streams, we define the VME as customers' behavioral manifestations that have a service, product, or brand focus, using static or dynamic visuals (e.g., images or videos). This definition will guide the inquiry.

Field Study-Netnography
Netnography is an ethnographic marketing research technique investigating communities and cultures emanating from computerized communications (Kozinets 2010). Netnography allows researchers to analyze information contained in naturally occurring data (Berger et al., 2020). This approach is useful when exploring online behavior and has been employed in multiple studies to identify behavioral forms of engagement (e.g., Azer and Alexander 2022; Azer, Blasco-Arcas, and Harrigan 2021; Brodie et al. 2013;Hollebeek and Chen 2014). We followed the recommendations for site selection proposed by Kozinets (2010). To ensure diversity of contexts and robustness of findings, we used Instagram and Facebook, as these are among the largest social networks worldwide, with almost 2.9 billion (Facebook) and 1.21 billion (Instagram) active users per month (Globaldata.com 2022;Statista.com 2022).
To ensure the stability and validity of findings, the NCapture facility of NVivo Pro software was used. We extracted 29,782 Facebook and Instagram pictorial posts created by individual users on the official pages of Amazon, Apple, American Airlines, and Nike. 1 We increased generalizability by researching a range of industries, including both services and products. Following recommendations for netnographic studies, we copied publicly shared archival data, comprising all posts, for an entire year and then filtered this for relevance (Kozinets 2010). Publicly communicated online messages are open to researchers; legally, it is the user's responsibility to identify what information to share publicly on social media (Kozinets 2010).
Accordingly, we included only public posts and removed all duplicate posts (to avoid redundancy), advertisements, pictures that included text, and promotional posts generated by companies or customers for their business. We proceeded with 18,985 relevant images for analysis.

Interpretation and Analysis
A pictorial analysis was conducted using NVivo Pro software to interpret the selected images, following thematic analysis procedures using open and axial coding (Corbin and Strauss 2008). Open coding involves breaking data apart and considering all possibilities before allocating conceptual coding labels. Axial coding involves crosscutting or relating concepts to one another. This process corresponds to the analytical sequence of abstracting and comparing, followed by checking and refinement, which is recommended for qualitative data analysis (Kozinets 2010). To ensure rigorous analysis, the study followed visual rhetoric theory, where visual images are viewed as communicative artifacts or symbols that perform communication (Bakri, Krisjanous, and Richard 2020). From an analytical perspective, visual rhetoric is an important tool when considering visual data. It casts light on the communicative dimensions of images and is characterized by considering aspects such as nature and function of images (Kjeldsen 2018). During our pictorial analysis, features, such as the presence or the absence of the brand in the image and how the consumers present the brand were noted. Also, if experiences of using the brand were included, or if the image presented a functional representation of the brand. Finally, if elements in the image inferred a specific view of that brand and the valences of this view-positive or negative. Crosschecking of coding within the research team was undertaken and discrepancies discussed ultimately reaching an overall agreement of 90% among coders.

Study 1: Results
This study introduces a typology of four VME forms classified as positive or negative. Positive forms of VME are evidential and experiential, while negative forms mocking and dissuasive. Table 2 offers examples of the forms as extracted and analyzed from the data.

Evidential
Evidential VME refers to customers' visual behavioral manifestations evidencing ownership/purchase of brands, products, or services. This category represents 35% 2 of images. Here, customers do not appear in photos; the focus is on the brand; they manifest brand ownership. For example, the images show the product off packaging or after assembly. In services, the images show the place of service (e.g., interiors of an aircraft). The image act is clarifying or illustrative of the brand (Bakewell 1998;Kress and van Leeuwen 2006;Searle 1976). Consumergenerated content that provides detail about a product or service is perceived as helpful by other customers when assessing products or services (Akpinar and Berger 2017;Filieri et al. 2021). Thus, engaging in evidential VME should make others feel they have a better sense of the brands' relevant features.

Experiential
Experiential VME refers to customers' visual behavioral manifestations experiencing brands, products, or services. In 40% of images, and unlike evidential VME, experiential VME see customers appearing in the image using the brand or service (Hartmann et al. 2021). Here, images tell stories of where they are or their enjoyment of using a brand. The image act reveals the customer's experience (Bakewell 1998

Mocking
Mocking VME refers to customers' visual behavioral manifestations ridiculing brands products, or services. In 15% of the photographs, customers use parodies of the brand logo, created by themselves to mock the brand or service. In mocking VME, customers do not appear in the image, nor are real photos used to evidence purchase or experience. Although mocking images does not focus on specific flaws of the brand they are used to convey negativity; capturing a negative verdictive act about the brand (Kress and van Leeuwen 2006;Searle 1976). Making fun of service providers has been observed in engagement literature using textual modality and has been identified as both retainable and memorable (Azer and Alexander 2020b).

Dissuasive
Dissuasive VME refers to customers' visual behavioral manifestations dissuading others from buying or using brands, products, or services. In 10% of photos, customers derogate the brand to influence others' impressions of the brand. Customers use alerting tools, such as a large cross over the brand logo. Thus, unlike mocking VME, customers engaging in dissuasive VME are not focusing on the brand's message but, rather, use images as a warning act to dissuade others from using the brand (Bakewell 1998;Searle 1976). Alerting images seize attention and produce greater inference (Schimmack and Derryberry 2005). Importantly, customers dissuade others not only from transacting with a brand but can also recommend competitors. The relative attractiveness of competitors can influence commitment to a brand (Azer and Alexander 2020b;Lemon, White, and Winer 2002).

Discussion
Study 1 contributes to engagement research by conceptualizing four forms of VME. Table 3 offers an explanation of each form and reveals that they are mutually exclusive. According to the engagement literature, modality is expected to influence the impact of CEB (van Doorn et al. 2010) and, in relation to image act theory, images differ in their effect on others (Bakewell 1998;Kress and van Leeuwen 2006). Therefore, as different forms of VME are unlikely to be equally influential or beneficial to the brand; it is crucial to explore how forms influence particular outcomes and whether effects differ between brandrelated and customer-related outcomes. Although exposure to images influences brand evaluations, purchase intentions, and sharing intentions (Akpinar and Berger 2017;Filieri et al. 2021;Ordenes et al. 2019), research has not examined the impacts of different forms of VME (generated by customers). Visual content research has considered the impact of brand-generated images on brand evaluation and behavioral outcomes (e.g., purchase and sharing intentions), but willingness to imitate has received less attention. Imitation behavior is important for generating demand and affects empathy, trust, and subsequent behaviors on social media (Ki, Park, and Kim 2022;Zulli and Zulli 2022). Therefore, in the following three studies we focus on four outcomes of VME that capture both customer-related and brand-related outcomes: brand evaluation, purchase intentions, resharing intentions, and willingness to imitate. Study 2 investigates the impact of VME forms on other customers and brands, while Study 3 and 4 investigate that impact moderated by social and brand interaction, respectively.
Study 2: Impact of VME According to image acts, when assessing products or services before purchase, evidential VME offers an illustrative image of how the product looks off its packaging or after assembly (Bakewell 1998). Viewers may feel a better sense of the product's relevant features (Akpinar and Berger 2017;Filieri et al. 2021). However, action images with a particular object, such as evidential VME, grab the attention of viewers toward the experience and the object more than the illustrative images. Images showing experiences or actions, especially pleasant ones, receive greater attention than static or non-action pictures (Bakewell 1998;Schimmack and Derryberry 2005) as the former evoke mental images that increase the intentions of others to try the product or imitate the post (Filieri et al. 2021;Kress and van Leeuwen 2006;Ordenes et al. 2019). Images of customers using the brand may result in more favorable brandand other customer-related outcomes than pictures that lack facial presence (Hartmann et al. 2021), therefore, experiential VME could stimulate more favorable outcomes compared to evidential VME. H1: Experiential VME will have a stronger positive impact than evidential VME on other customers' (i) brand evaluation, (ii) purchase intentions, (iii) resharing intentions, and (iv) willingness to imitate.
The impact of negative CEB can be understood in terms of intensity of impact (Azer and Alexander 2020a). Intensity of impact refers to the level of change effected within the target audience and associated brand-or other customer-related outcomes (van Doorn et al. 2010). Customers choose to use mockery instead of literal meanings to convey a verdictive negative image act toward brands or service providers (Bakewell 1998;Filik et al. 2016). Although verdictive acts enhance the critical effect and, hence, their negativity, these images provide no detailed information about the brand. However, mocking images are more memorable and more entertaining (Bakewell 1998), hence easily imitated (Zinkhan and Johnson 1994) than literal content describing brand flaws (Azer and Alexander 2020b).
Unlike mocking, customers engage in the dissuasive form by using alerts, such as crosses over the brand logo. These additions grab attention, produce more inference than other negative pictures (Kress and van Leeuwen 2006), and enhance performance on sensory processing tasks (Schimmack and Derryberry 2005). Warning image acts are more conclusive, leaving less room for ambiguity (Bakewell 1998). Importantly, to dissuade others from using the brand, customers explicitly show competitors' logos in a way that derogates brands. Recommending competitors by making them more attractive than the focal brand or service influences commitment to a brand relationship (Azer and Alexander 2020b;Lemon, White, and Winer 2002) and suggests greater intensity than mocking.
H2: Dissuasive VME will have a stronger negative impact than the mocking form on other customers' (i) brand evaluation, (ii) purchase intentions, (iii) resharing intentions, and (iv) willingness to imitate.

Study 2: Design and Procedures
We used an independent group experimental design to investigate the difference in impact for the four forms of VME. The stimulant material (see Web Appendix A) was developed using images analyzed and coded in the field study and simulated as an Instagram page to ensure realism and believability. To control brand familiarity, all the pictures related to a fictional technology brand named "Star." Following the recommendations of Hair et al. (2010) on sample size requirements (0.05 alpha, 0.8 statistical power, and large effect size), a sample of 220 participants (cell size = 55, females 39.1%, average age = 24.3, M = 1.44, SD = 0.93) was recruited through Prolific, a specialized panel provider. Sampling methods ensured the participants checked brand-related posts on social media, in addition to the screening question "I check brandrelated posts on social media" (1 = Never, 5 = Always). Using the randomization facility provided by Qualtrics, participants were assigned randomly to the four forms of VME. The experimental manipulations were developed using the definitions of the forms of VME (Web Appendix A) and manipulation checks for the VME forms were conducted using the chi-square test indicating different answer patterns between manipulations, x 2 (36, N = 220) = 190.1, p < .001. Based on previous research, we controlled for users' motivations for viewing brand-related posts on social media using four items adapted from Chan and Prendergast (2015). To control for quality of pictures and brand familiarity, we photoshopped the pictures and replaced the real brand "Apple" with a fictitious brand "Star," in addition to measuring brand familiarity using three items adapted from Kent and Allen (1994). After exposure to the scenarios, the participants completed a questionnaire comprising items to measure dependent variables brand evaluation index (Akpinar and Berger 2017), resharing intention (Akpinar and Berger 2017;Lee and Ma 2012), purchase intention (Coyle and Thorson 2001), willingness to imitate (Kasser et al. 2004), manipulation checks, and demographic items (age and gender). Factor loading and reliability of scales were above the recommended threshold of .7 (Hair et al. 2010) (see Web Appendix B, Table 1). Tests were undertaken to confirm convergent (AVE >.5) and discriminant validity and both maximum and average shared variance were less than AVE (Bagozzi and Yi 1988). Discriminant validity was confirmed as the square root of AVE for each construct was greater than the correlations between them and all other constructs (see Web Appendix B Table 2). Correlations among the study constructs showed no threats of multicollinearity (R < .80) (Hair et al. 2010). Finally, we examined CMV bias with Harman's single-factor test. The results from this test showed the greatest variance explained by one factor was 35%, indicating common method bias is not likely to be a contaminant of results (Podsakoff et al. 2003).
Post hoc analysis was conducted to offer a deeper understanding of differences between forms by testing each possible pair using a least significant difference test (LSD) (see Web Appendix B- Table 4). The results confirmed the above findings for all the dependent variables. Differences between the four forms were all significant. Experiential VME showed the highest impact among the forms on all dependent variables; however, it was not significantly different from mocking VME for resharing intentions. Evidential VME showed higher impact than both mocking and dissuasive; however, it did not have a significantly different impact from mocking on brand evaluation, willingness to imitate, and resharing intentions. Both mocking and evidential VME showed a favorable impact on brand evaluation and willingness to imitate, but the impact was more favorable for evidential VME. A comparison of resharing intentions showed only marginally significant differences between the four forms. Having established differences in the impact of VME forms, we continue by investigating their impact moderated by high and low social interaction.

Study 3: The Role of Social Interaction
Online social interaction (e.g., likes, shares, and comments) is an integral and frequently used aspect of social media engagement (Seo et al. 2019;Zell and Moeller 2018). The number of likes and comments received by a photo reflects the collective peer opinion of other social actors around the worth and attractiveness of the image (Li and Xie 2020b) and stimulates greater engagement with it (Dolan, Conduit and Fahy 2016). Although a higher number of likes on a photo should stimulate positive brand-and customer-related outcomes, it is unclear how these may differ across the different VME forms.
According to existing social media research (e.g., Bakhshi Saeideh et al., 2014;Hartmann et al., 2021), visible human faces in images can drive engagement. Such engagement, as represented by the number of "likes" displayed under each image, serves as a form of peer influence or social reinforcement. According to social reinforcement theory, visible human faces in images affect evaluations of and behaviors toward brands, products, or services (Seo et al. 2019). Therefore, following this theorizing, and based on the Study 2 results, we expect experiential VME to have a more favorable brand-and other customer-related impacts when accompanied by higher levels of social interaction and when compared to evidential VME. Thus, we suggest: H3: High (versus low) levels of social interaction will moderate the positive impact of experiential VME compared to evidential VME on other customers' (i) brand evaluation, (ii) purchase intentions, (iii) resharing intentions, and (iv) willingness to imitate.
Unlike positive VME, likes and comments on negative posts infer agreement with the negative form and its judgment on worth and, in some instances, attractiveness (e.g., mocking).
Prior research suggests that there would be a negative overall brand attribution based on the number of likes on a negative photo (Phua and Ahn 2016). Therefore, high levels of social interaction may moderate the impact of negative VME forms, yet this moderating effect may differ between the two forms. Negative humorous image acts positively affect memory and attitude towards the brand (Chung and Zhao 2003;Kress and van Leeuwen 2006). However, for more intense forms, for example, dissuasive VME, higher levels of social interaction will determine the extent to which a social network perceives the warning act to be truthful and believable (Bakewell 1998;Seo et al. 2019); hence, unfavorable brand-related outcomes and favorable actor-related outcomes (possible resharing and imitation) may be more likely.
H4: High (versus low) levels of social interaction will ameliorate the impact of mocking VME and increase the negative impact of the dissuasive VME on other customers' (i) brand evaluation, (ii) purchase intentions, however, will show more favorable impact on their (iii) resharing intentions, and (iv) willingness to imitate.

Study 3A: Positive VME Forms (Design and Procedure)
This experiment used a 2 (VME positive forms: experiential, evidential) × 2 (level of social interaction: high and low) between-subjects factorial design, resulting in four scenarios. Images from Study 2 were again used to represent each form, and the social interaction level was manipulated using the number of likes and comments (see Web Appendix A). A sample of 200 participants (cell size = 50, females 42.5%, average age = 25, M = 1.59, SD = 0.915) was recruited through Prolific. As with Study 1, procedures were followed to ensure sample representation and randomization of treatment assignment between subjects. The following questions were used to check the respondents' understanding of the positive forms of VME: "This picture shows... (a-people picturing themselves while experiencing the brand, b-people unboxing the brand, evidencing the purchase but not the usage)." The experimental manipulations to check the respondents' understanding of the level of social interaction used the following question "I think this picture got a ... level of social interaction from other social media users (a-high, b-low)." Manipulation checks were conducted using the chi-square test for the positive VME forms, indicating different answer patterns between manipulations, x 2 (4, N = 200) = 1200, p < .001; similarly, for the level of social interaction, x 2 (1, N = 200) = 200, p < .001.
This study used the same control and dependent variables as Study 2. Factor loading and reliability of scales were above the recommended threshold of .7 (see Web Appendix B, Table 1). As with Study 2, tests were undertaken to confirm convergent (AVE > .5) and discriminant validity (see Web Appendix B, Table 2).

Study 3A: Results
After satisfying preliminary checks on the assumption of homoscedasticity (Levene's test p >.05) for all dependent variables, and the equality of the entire variance-covariance matrixes (Box's test p = .351), MANOVA was conducted, and the results reveal a significant interaction effect between positive VME forms and level of social interaction (Wilk's lambda = .978, F (4, 213) = 5.111, p < .001); any effects for the control variables were nonsignificant under both conditions of social interaction. The interaction was significant for brand evaluation (F (1,216) = 9.44, p < .001), purchase intent (F (1,216) = 15.69, p < .001), and willingness to imitate (F (1,216) = 11.92, p < .001), but not significant for resharing intentions (F (1,216) = .003, p = .957) (rejecting H3iii). As shown in Web Appendix B (Table 5) and Figure 2, the level of social interaction, whether high or low, moderated the impact of positive VME forms, yet differently. High social interaction on experiential form showed a higher increase in brand evaluation (Mexp = 5.38, Mevi = 4.47; p < .001), purchase intent (Mexp = 5.18, Mevi = 4.12; p < .001), and willingness to imitate (Mexp = 5.06, Mevi = 4.16; p < .001) compared to evidential form. Similarly, the difference is also significant in the low social interaction condition, albeit both forms showed lower mean scores on all dependent variables, yet more favorable with experiential, thus confirming H3i, H3ii, H3iv.

Study 3B: Negative VME Forms (Design and Procedures)
This study also used a 2 (VME negative forms: mocking and dissuasive) × 2 (level of social interaction: high and low) betweensubjects factorial design, resulting in six scenarios (see exemplars in Web Appendix A). A sample of 200 participants (cell size = 50, females 30%, average age = 22.2, M = 1.55, SD = 0.922) was recruited through Prolific, with items from Study 3A used to check the respondents' understanding of the social interaction level. The following question was used to check the respondents' understanding of the negative forms: 'I think this picture...(a-mocks (makes fun of) the brand, b-discourages using the brand, encouraging the competitors). Manipulation checks were conducted using the chi-square test for the negative forms, indicating different answer patterns between manipulations, x 2 (4, N = 200) = 600, p < .001. Similarly, for the level of social interaction, x 2 (1, N = 200) = 200, p < .001. The same procedures were followed as for Study 3A (see Web Appendix B, Tables 1 and 2).

Study 3B: Results
Preliminary checks (Levene's Test p >.05, Box's Test p = .501) were satisfied, MANOVA was conducted, and results reveal a significant interaction effect between negative VME and the level of social interaction (Wilk's lambda = .820, F (4, 217) = 11,90, p < .001); any effects for the control variables were non-significant under both conditions of social interactions. The interaction was significant for brand evaluation (F (1,220) = 43.287, p < .001), purchase intent (F (1,220) = 30.512, p < .001), and willingness to imitate (F (1,220) = 10.795, p < .001); however, it was not significant for resharing intentions (F (1,220) = .797, p = .373) (rejecting H4iii). As shown in Web Appendix B (Table 5) and Figure 3, the level of social interaction, whether high or low, moderated the impact of negative VME forms. High social interaction on mocking VME showed the most favorable impact on brand evaluation (Mmock = 5.00, Mdiss = 2.40; p < .001), purchase intent (Mmock = 4.58, Mdiss = 2.39; p < .001), and willingness to imitate (Mmock = 5.00, Mdiss = 3.06; p < .001) compared to the dissuasive form. Mocking the brand-even with high social interaction from other social media users-did not negatively affect the brand. This is not the case with low social interaction, as mocking the brand negatively affected how other actors evaluated a brand (Mmock = 3.45) and their purchase intentions (Mmock = 2.94). Surprisingly, with low social interaction, dissuasive VME, although showing a negative impact on brand evaluation, was not as negative as with high social interaction (Mdiss = 3.60), demonstrating the power of social networks. Importantly, compared to low social interaction (Mdiss = 2.18), high social interaction on dissuasive forms increased the willingness of other customers to imitate such posts. Therefore, H4i, H4ii, and H4iv were confirmed.

Study 4: The Role of Brand Interaction
Individuals use social networks to connect with other individuals in their networks and with brands. Nevertheless, apart from literature on online reviews that notes the value of responses to negative consumer reviews (e.g., Azer and Alexander 2020a; Xie et al. 2016), the role of brand interaction on social media images posted has not been examined. Thus, our final experiment seeks to understand the role of brand interactions with VME forms. Specifically, the impact of brand interactions on other network members and whether that impact differs with different levels of brand engagement (e.g., comments vs. likes). The degree of brand engagement ranges from basic forms of engagement (e.g., "liking") to higher forms of engagement depicting a greater investment of resources (e.g., writing comments), which may prompt more elaboration and foster more positive effects than likes (Gable et al. 2004). A lack of response to positive contributions could stem from a lack of interest, or disapproval of the post (Zell and Moeller 2018). Thus, we predict "comments and likes" may reveal a higher level of brand engagement with consumers' brand-related posts than just "likes," resulting in more favorable brand-and other customer-related outcomes. In addition, no interaction from the brand side is expected to yield less favorable outcomes than likes only, and likes and comments that, in turn, depict greater investment of resources by the brand. Thus, we predict: H5: Brand interaction, through comments and likes vs. likes only vs. no interaction will moderate the positive impact of experiential VME over evidential VME on other customers' (i) brand evaluation, (ii) purchase intentions, (iii) resharing intentions, and (iv) willingness to imitate. Negative VME forms may be embarrassing for the brand, yet a response has always been advocated in prior literature to address consumer reviews (Xie et al. 2016), yet the effect of brand interaction on social media brand-related images is unclear. Customers use a dissuasive form attempt to provoke action, as opposed to merely ridiculing the brand with mocking. Recent engagement research suggests that managerial responses to negative CEB could mitigate the negative impact, however, warning acts may still result in unfavorable brand-related outcomes (Azer and Alexander 2020a;Bakewell 1998). Brand-generated comments on negative VME forms could also lessen other customers' intentions to imitate such negative forms. Such brand interaction may help restore a positive image and reduce the likelihood of other actors drawing negative inferences about the brand (Xie et al. 2016). Following this theorizing, we hypothesize that: H6: Brand interaction through comments vs. no interaction will ameliorate the negative impact of mocking VME over dissuasive VME on other customers' (i) brand evaluation, (ii) purchase intentions, (iii) resharing intentions, and (iv) willingness to imitate.

Study 4A: Positive VME Forms (Design and Procedure)
This experiment used a 2 (VME positive forms: experiential, evidential) × 3 (brand interaction: likes only, likes and comments, and no interaction) between-subjects factorial design, resulting in twelve scenarios (see exemplars in Web Appendix A). A sample of 300 participants (cell size = 50, females 40.1%, average age = 25.6, M = 1.26, SD = 0.955) was recruited through MTURK 3 . The same procedures were followed to ensure a representative sample and randomized assignment of treatment between subjects. Experimental manipulations from Study 3A were used to check respondents' understanding of the positive forms of VME. Understanding of brand interaction was checked using the following question: "I think the 'Star' brand… with this post (a-only liked, b-commented and liked, cdid not interact at all)." Manipulation checks for the positive VME forms indicated different answer patterns between manipulations, x 2 (4, N = 300) = 1800, p < .001, and this was similar for the brand interaction x 2 (3, N = 300) = 1200, p < .001. The same dependent and control variables were used in this study. Factor loading and reliability of scales and tests were undertaken to confirm convergent and discriminant validity (see Web Appendix B, Tables 1 and 2).

Study 4A: Results
Preliminary checks (Levene's test p >.05; Box's test p = .461) were satisfied, and a MANOVA was conducted. The results revealed a significant interaction effect between positive VMEs and brand interaction (Wilk's lambda = .930, F (8, 780) = 3.580, p < .001). Any effects for the control variables were nonsignificant under the three conditions of brand interaction. The interaction was significant for brand evaluation (F (2,393) = 8.673, p < .001), purchase intent (F (2,393) = 8.40, p < .001), and willingness to imitate (F (2,393) = 6.135, p = .002), but not significant for resharing intentions (F (2,393) = .032, p = .969) (rejecting H5iii). As shown in Web Appendix B (Table 6) and Figure 4, brand interaction moderated the impact of positive VME. Overall, the brand's interaction, whether with only likes or with likes and comments, increased other customers' evaluation of the focal brand, their purchase intentions, and their willingness to imitate such posts, compared to no interaction. However, this impact differs with each form. The brand's interaction with "likes" and "comments" on experiential VME showed a higher increase in brand evaluation (Mexp = 5.81, Mevi = 4.49; p < .001), purchase intent (Mexp = 5.52, Mevi = 3.71; p < .001), and willingness to imitate (Mexp = 5,45, Mevi = 3.78; p < .001) compared to evidential VME. Brand interaction with only "likes" showed a less favorable impact on both forms than interaction with "likes and comments." Importantly, with no brand interaction, experiential VME had a more favorable impact on other customers' purchase intentions and willingness to imitate, while brand evaluation was also favorable for both forms, albeit less favorably than when the brand interacted. Therefore, H5i, H5ii, and H5iv were confirmed.

Study 4B: Negative VME Forms (Design and Procedure)
This experiment used a 2 (VME negative forms: mocking, dissuasive) × 2 (brand interaction: comment and no interaction) between-subjects factorial design, resulting in six scenarios (see exemplars in Web Appendix A). A sample of 200 participants (cell size = 50, females 32.7%, average age = 25.3, M = 1.16, SD = 0.95) was recruited through MTURK. Experimental manipulations from Study 3B were used to check respondents' understanding of negative forms. Understanding of brand interaction was checked using the following question: "I think the 'Star' brand… on this post (a-commented, b-did not interact at all)." Manipulation checks for negative VMEs indicated different answer patterns between manipulations, x 2 (4, N = 200) = 600, p < .001, and this was similar for the brand interaction, x 2 (1, N = 200) = 300, p < .001. The same control and dependent variables were used in this study as in the previous study (see Web Appendix B, Tables 1 and 2).

Study 4B: Results
Preliminary checks (Levene's test p > .05; Box's test p = .545) were satisfied. A MANOVA was carried out. The results revealed a significant interaction effect between negative VMEs and brand interaction (Wilk's lambda = .889, F (4, 209) = 6.527, p < .001). Any effects for the control variables were non-significant under the two conditions of brand interaction. The interaction was significant for brand evaluation (F (1,212) = 6.93, p < .001), purchase intent (F (1,212) = 4.93, p = .027), and willingness to imitate (F (1,212) = 12.5, p < .001), but not significant for sharing intentions (F (1,212) = 1.096, p = .296) (rejecting H6iii). As shown in Web Appendix B (Table 6) and Figure 5, brand comments on negative forms of VME increased other actors' favorable evaluation of the focal brand and their purchase intentions compared to no interaction at all from the brand. However, with no brand interaction, users' willingness to imitate the negative forms increased compared to when the brand interacted with comments. Importantly, this impact differs with each form. For instance, when brands interact with the dissuasive form, a stronger negative impact on brand evaluation was still evident (Mmock = 5.20, Mdiss = 3.00; p < .001) and purchase intentions (Mmock = 4.65, Mdiss = 2.07; p < .001), while the brand's interaction with mocking form mitigated their negative effect. Importantly though, the absence of brand interaction increased the negative impact of both forms on brand evaluation and purchase intentions while encouraging willingness to imitate such posts (Mmock = 5.00, Mdiss = 3.99; p < .001). Therefore, H6i, H6ii, and H6iv were confirmed.

Theoretical Implications
This research offers contributions in three areas, informing and extending engagement, communication, and visual content research.
Conceptualization of VME. This paper introduces the VME concept, capturing forms of visual CEB and extending current knowledge of engagement modality. Thus, we respond to calls for research on visual modality motivated by the rapid proliferation of images in social media and the focus on textual analysis in previous research (Babić Rosario, De Valck, and Sotgiu 2020;Hartmann et al. 2021;King, Racherla and Bush 2014). Our research introduces a distinctive conceptualization of VME and evidence of its impact. We contribute to communication theory and visual content research through our investigation of the behaviors that images intend to communicate and prompt, which are central to image act theory (Bakewell 1998;Barinaga 2009;Kress and van Leeuwen 2006). We expand this literature, which had previously been limited to exploring the impacts of specific image characteristics (e.g., Kwon Jumbum et al., 2022;Li & Xie, 2020).
Exploration and Identification of Different VME Forms. We present the first typology of VME with two positive (evidential and experiential) and two negative (mocking and dissuasive) forms and investigate their impacts. We offer unique insights into visual CEB with brands through discrete forms of VME. Customer-generated images are pivotal for effective social media marketing, but existing research has focused on brandgenerated content and ignored differences between image types (e.g., Akpinar and Berger 2017;Hartmann et al. 2021;Ordenes et al. 2019). From a visual communication and social media perspective, our forms indicate a more nuanced view of customer-generated visual content and an improved understanding of its behavioral outcomes.
The Impact of VME Forms. Our three experimental studies reveal the different impacts of VME, across its various forms, on firms and on other customers in the network. We give evidence of the interplay of social interactions with various VME forms, how they reinforce evaluations of and behavior toward brands, and how this reinforcement differs between VME forms. Our findings extend previous research on visual communication and social media. The importance of how "likes" on social media posts serve as social reinforcement has been suggested (Seo et al. 2019) but differences between visual forms had not previously been considered. We provide new knowledge about the impact of brand interactions with various forms of VME. Prior research has suggested a role for brand interactions with customers on social media around customer loyalty and trust (Ferreira 2018). Our analysis, however, encompasses different levels of brand interaction, including comments, likes, and no interaction, which, to our knowledge, have not previously been studied either in combination or with various forms of VME.
Finally, we contribute by using willingness to imitate as a dependent variable. Prior research has focused primarily on customers' evaluations and behavioral outcomes (e.g., purchase and sharing intentions), rather than their willingness to imitate (Akpinar and Berger 2017;Ordenes et al. 2019). Willingness to imitate is, however, a prevalent phenomenon on social media platforms and important for potential brand-and consumerrelated outcomes (Zulli and Zulli 2022). It is a creative act, indicating both investment of resources and user participation (Blasco-Arcas et al. 2020;Jaakkola and Alexander 2014). In our research, we demonstrate that different forms of VME each have a distinct impact on others' willingness to imitate, offering insights into the visual drivers of imitation behavior.

Managerial Implications
Our study puts forward important managerial implications for social media practitioners in various industries. First, our findings present a guide to social media marketers in identifying critical VMEs when conducting brand-related social listening. A more nuanced understanding of the impacts of different forms of VME facilitates the potential to utilize the perceptions, attitudes, and behaviors enacted by customers to influence others through visual modes. Improved identification of potentially influential negative VMEs also allows organizations to act before harmful interactions become viral. The paper assesses the severity of different forms of negative VME, which can aid firms when evaluating customer interactions and determining the implications of negative forms of VME and their impacts on customer outcomes and performance.
Second, monitoring VME forms should be a priority. Our research suggests that companies should include image presence in their social "listening" metrics but also highlights the importance of deeper analysis to understand potential impacts better. This could be achieved through online image-processing tools, such as Google Cloud Vision API to analyze visual content, enabling firms to react appropriately to benefit from these manifestations or avoid potential risks.
Third, firms can use their existing customers' VMEs to demonstrate their value to new customers. Our results show that different forms of VME foster different reactions in other social media users. This is key from a managerial perspective, as it gives evidence to managers of behavioral manifestations they may wish to utilize for their marketing campaigns. Based on our results, managers should be more selective when choosing which VME to use due to differences in their impacts.
Fourth, managers could use our VME typology to incentivize specific VMEs and involve users who manifest these VMEs in buzz campaigns, specifically experiential VME. Finally, this paper offers managers a nuanced understanding of the role of brand interaction with customers' VMEs in social media and its influence over other customers in the network. Counterintuitively, a higher level of brand interaction with mocking VME reduces the willingness of other users to imitate this kind of manifestation. Consequently, managers need to be responsive in social media, especially to negative forms of VME with highintensity levels, to mitigate potential negative effects.

Limitations and Future Research
The study focused on Facebook and Instagram as photo-based social media platforms and found similar patterns between them. Future research could explore other platforms, such as Pinterest. The study provides definitions for different forms of VME, which can be used in future research. By using this typology, researchers can investigate how users' characteristics, psychological factors, and cultural traits influence their use of VME. This can contribute to understanding users' intentions and expectations, such as seeking popularity and likability, and can be useful for influencer marketing, user experience, and visual content.
Customers have multiple experiences with the same brand over time (van Doorn et al. 2010). As the relative quality of the experience changes, so does the likelihood of VME. Future research could investigate the dynamics of VME over time, considering different touch points in the customer journey. Future research might assess posting times or dominance in certain networks or certain brands, which would complement existing research.
This paper focuses on brand-related images shared voluntarily by customers on social media. However, future research could explore engagement that occurs less voluntarily (cf. Hollebeek, Kumar, and Srivastava 2022). Virtual reality (VR) and augmented reality (AR) offer different senses beyond the visual, such as haptic, tactile, and aural (Kim, Lee, and Jung 2020), which could enrich customer experience and engagement literature. Research could investigate how VR and AR could blur the line between reality and fantasy, leading to new forms of VME, particularly in immersive gaming and social interaction contexts. This paper offers forms of VME while the shared visual content is static images. Future research may use the conceptualization of VME this paper offers, to explore other VME forms that may emerge in dynamic visual content such as videos and social media stories.
Images are a powerful means of communication, hence, the focus on pictorial posts created by customers, but future research could compare different modalities, such as textual and verbal. The paper used a technology product as a fictitious brand to ensure the realism of the scenarios and product involvement. Future research could compare the impact of VME forms in services (such as travel services vs. financial services) as opposed to goods. Finally, this paper emphasizes the importance of studying customers' willingness to imitate VME. Future research could use VME forms to understand the brand-building consequences and potential impacts on various outcomes. The paper anticipates that its findings are likely to encourage further research in this area (see Table 4). a) It is valuable to know if such behavior is underpinned by selfrepresentation, rather than a brand promotion b) Would that behavior appear more with luxury brands, hence, customers reflecting that self-image (e.g., association with luxury brands, affordability, and financial stability)? c) Customers do not appear in such images, focusing mainly on the brand; would the firm be concerned if the poster is not an actual user of the brand, or is it all about the promotion? Would that ingenuine engagement affect the customer-firm relationship? Experiential Customers' visual behavioral manifestations experiencing services, products, or brands a) How the users' appeal (e.g., cheerful faces, beautiful figures, wellgroomed looks, or attractive gestures) interplay with the impact of such form? b) How the users' social status as appeared in the images (e.g., family vs. sole experiencers vs. romantic partners) may moderate its impact on other customers, does congruity with other customers' social status play a role in such impact? These are likely to contribute to influencer marketing and user experience research streams c) It is also important to explore if experiential VME could be used negatively and how firms could mitigate that if it occurs Mocking Customers' visual behavioral manifestations ridiculing brands, products, or services a) It is worth exploring mechanisms and strategies to ensure that mocking VME does not damage the integrity of brands, services, and products or other customers' evaluations of them b) How firms respond to such form is very important as well; would the use of humor counter the mockery effect or would it lead to a more severe brand-related impact? Dissuasive Customers' visual behavioral manifestations dissuading others from using or buying products, brands, or services a) Future research could elaborate on the implications of this behavior b) It is valuable to know if customers more easily engage in dissuasive behavior, which is more directive in nature, inclining to make the competitors look more attractive c) How that could affect C2C relationships is unclear, especially that directive acts are not popular in such relationships, as they project superiority demeanor