Sweet escape: The role of empathy in social media

engagement with human versus virtual influencers

Agata Mirowska, Jbid Arsenyan

To cite this version:

Agata Mirowska, Jbid Arsenyan. Sweet escape: The role of empathy in social media engagement
with human versus virtual influencers. International Journal of Human-Computer Studies, 2023, 174,
pp.103008. �10.1016/j.ijhcs.2023.103008�. �hal-04021848�

HAL Id: hal-04021848

https://rennes-sb.hal.science/hal-04021848v1
Submitted on 31 Mar 2025

HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est

archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents
entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non,
lished or not. The documents may come from émanant des établissements d’enseignement et de
teaching and research institutions in France or recherche français ou étrangers, des laboratoires
abroad, or from public or private research centers. publics ou privés.

Distributed under a Creative Commons Attribution - NonCommercial 4.0 International License

Version of Record: https://www.sciencedirect.com/science/article/pii/S1071581923000149
Manuscript_adcfed81c1bb9dc9ad16d60cbbe9abf0

Sweet escape: The role of empathy in social media engagement with human versus virtual

influencers

Agata Mirowskaa, Jbid Arsenyamb

aNeoma Business School, 1 Rue du Maréchal Juin, 76130 Mont-Saint-Aignan, France

b
Rennes School of Business, 2 rue Robert d'Arbrissel, 35000 Rennes, France

© 2023 published by Elsevier. This manuscript is made available under the CC BY NC user license
https://creativecommons.org/licenses/by-nc/4.0/
 Sweet escape: The role of empathy in social media engagement with human versus

virtual influencers

Abstract

Virtual influencers engage in emotional sharing to gain and keep followers. However, given

that many people use social media for diversion purposes, this emotional sharing may hinder

users’ ability to escape from everyday emotional experiences, particularly for highly

empathetic individuals. Using a between subjects, randomised experimental design, we

explore how empathy affects reactions to virtual vs. human influencers, showing that those

highest on empathy are more likely to follow a virtual influencer, and rate her as more

socially attractive, than a comparable human influencer; these results disappear when the

influencers’ true nature is unknown to participants. We postulate that these results represent

an “escapism effect”, where the virtual influencer is expected to provide greater diversionary

benefits from everyday human emotional experiences and require fewer cognitive resources

in the form of emotional sense making. We present practical implications and future research

opportunities arising from this effect.

Keywords

Virtual agent, virtual influencer, social media influencer, empathy, Uncanny Valley
 2

Sweet escape: The role of empathy in social media engagement with human versus

virtual influencers

1 Introduction

The social sharing of emotions is a common human tendency, one that predates the

use of social media and mobile devices (Rimé, 2009). The democratisation of the Internet has

enabled individuals to move this social sharing online and the evolution to Web 2.0 has given

rise to platforms such as Twitter and Instagram, which showcase individuals’ emotional

experiences on a much larger scale (X. Liu et al., 2020). Although digital entities such as

chatbots and avatars are already prevalent on social media (Nowak et al., 2015; Xu et al.,

2017), we have recently seen the emergence of virtual agents in these human networks, who

operate independently from user input (Arsenyan & Mirowska, 2021; Park et al., 2021). In

the form of virtual influencers, virtual agents are already used for marketing purposes (Tan &

Liew, 2020; Yen & Chiang, 2021) and are quickly becoming opinion leaders in areas of

fashion and entertainment (Tietjen, 2018). These virtual agents are created to blend into the

social network in which they operate, in order to compete with their human counterparts in a

context of short attention spans and abundant content (Chang, 2010) and, to this end, may

seek to mimic the human sharing of emotional experiences (Arsenyan & Mirowska, 2021).

However, the rise in popularity of social media influencers has not been without

criticism. Detrimental effects on followers in terms of body image issues, excessive social

comparisons, and depression (Alfasi, 2019) have resulted in questions being raised regarding

the appropriateness of influencer use, particularly in marketing campaigns, and potential

alternatives given that social media use continues to grow amongst youth segments of the

population. As virtual influencers are increasingly deployed on mainstream social media

platforms, they may present a viable alternative to human influencers. Therefore, it is

 3

important to understand what reactions to virtual vs. human influencers are to be expected

(Arsenyan & Mirowska, 2021) and how they may be affected by individual differences (von

der Pütten et al., 2010). Identifying individual differences in willingness to engage with

virtual influencers may allow for virtual influencers to be designed and deployed in a way

that protects potentially vulnerable individuals (Vossen & Valkenburg, 2016).

Individuals engage in social media usage for specific purposes (McLaughlin et al.,

2022). One of the primary reasons reported for social media use is diversion and the ability to

experience emotional escape from one’s regular life, trials, and tribulations (Huang & Su,

2018; Schmuck, 2021; Sheldon & Bryant, 2016). However, individuals’ pursuit of diversion

may be hindered by influencers’ attempted emotional sharing as it affects their ability to

emotionally disconnect when interacting with other users. This ability to emotionally

disconnect may, in turn, be affected by individual level empathy – an individual’s ability to

“step into a character’s life” (MacDorman, 2019, p. 140) and experience, understand, and/or

distance themselves from the other’s emotional experience (Carré et al., 2013, p. 20).

We posit that virtual influencers may provide a potential “sweet escape” for those

highly susceptible to being drawn into others’ emotional experiences by providing

entertainment content that does not elicit the same level of emotional entanglement and

sense-making as that provided by human influencers. We use a between subjects, randomised

experimental design, to examine differences in willingness to engage with a virtual vs. human

influencer, both when participants were aware of the influencers’ true nature and when they

were not. We look for interaction effects between experimental conditions and participant

empathy scores to explore whether individuals who are more susceptible to being caught up

in others’ emotional experiences are more likely to follow a virtual influencer, and rate her as

more attractive, than a comparable human influencer.

 4

In the sections that follow, we will define the phenomena of virtual influencers and

present our hypotheses and experimental methodology. We conclude with a discussion of the

results, implications, and potential future research directions for this burgeoning field of

study.

2 (Virtual) Influencers

Social media influencers are micro-celebrities who specialise in cultivating a unique

public image transmitted via their online presence on one or more social media platforms

(Djafarova & Bowes, 2021; Khamis et al., 2016; Schmuck, 2021). Influencers succeed in

creating a personal brand (Abidin, 2018), thereby transforming themselves “from citizens to

corporations” (Senft, 2013, p. 351). This success, in turn, provides opportunities for income-

generation as companies and brands seek to reach target audiences via the influencers’ social

media reach. As such, the activity of influencers has global economic, legal, social, and

cultural consequences (Abidin, 2018).

More recently, the social media world has seen the rise of virtual influencers (VIs), an

amalgamation of digital avatars and virtual agents (Arsenyan & Mirowska, 2021), who

present a crafted reality not dissimilar from the manufactured authenticity of their human

counterparts, but with the key distinction of being authentically fake (Wills, 2019). VIs are

presented as human-adjacent, engaging in activities such as modelling, singing, socialising,

and opinion-sharing (Arsenyan & Mirowska, 2021). Similarly to human influencers (HIs),

VIs create their own public personas and elaborate storylines presenting varying degrees of

anthropomorphism (Hanus & Fox, 2015), offering a potential alternative for both companies

(seeking product placement opportunities) and users (as influencers to follow within various

domains).
 5

Since the popularisation of the first widely known VI, Lil Miquela (Shieber, 2018),

several additional VIs have been crafted and launched in the digital arena (Ahn et al., 2022).

From Noonouri, a computer-generated character stylized with anime features, to Shudu, the

world’s first digital supermodel, they represent the extremes of anthropomorphizing of virtual

agents. Similar to their human counterparts, these VIs engage in the practice of

microcelebrity (Marwick, 2018). As the influencer strives to appeal to audiences (by feeding

content to) and communities (engaging in content with), the follower becomes an active

agent, and even the commodity, in the exchange (Senft, 2013). By deciding where their

much-valued attention (commodified in the form of clicks, likes, and comments) is directed,

followers dictate which influencers and narratives get more attention and which are ignored.

As such, VIs in highly visible human networks create a whole new context in which the

Uncanny Valley Hypothesis can be applied (UVH; Mori et al., 2012). UVH argues that, as

the human-likeness of an artificial entity increases, peoples’ responses shift from affinity to

eeriness as the entity, despite showing a lifelike appearance, is not quite human. Even though

Mori (1970) makes a distinction between mobility (e.g. robots) and stillness (e.g. images) in

eliciting eeriness, the UVH has more recently been applied to computer generated and

moving images (Seymour et al., 2017).

Regardless of their level of human resemblance, VIs attempt to recreate a similar

experience for users as HIs, with emotionally charged content, realistic-sounding story lines,

and even producing music and video content (Ahn et al., 2022; Arsenyan & Mirowska, 2021;

Thomas & Fowler, 2021). Yet, it is not clear whether these attempts on the part of VIs to use

similar means as humans to make an emotional connection with their audience have uniform

effects for different types of users.

Previous research looking at the reactions of users to VIs on Instagram has relied on

users who had already chosen to engage with the VI in some form. Arsenyan and Mirowska
 6

(2021) found that, although VIs attempted to mimic the positive emotions that are prevalent

on Instagram, they earned more derogatory and sceptical language in the comments from the

audience compared to their human, as well as more cartoon-like, counterparts. Additionally,

despite earning higher traditional engagement metrics, such as likes and views, the human-

adjacent VI garnered fewer positive emotional reactions than the human-like and cartoon-like

influencer, even showing an “uncanny hill” with respect to higher negative reactions as

compared to the other influencers in the study (Arsenyan & Mirowska, 2021). Other research

has similarly found that VIs can elicit scepticism on the part of the audience that knows their

true nature (Sands et al., 2022), and even affect attitudes toward the brand they are endorsing

(Thomas & Fowler, 2021).

In contrast, we are interested in first impressions of HIs vs. VIs – that is, reactions on

the part of people who do not yet follow an influencer and are not aware of their existence.

Evidence suggests that such first impressions are key to the formation of behavioural and

affective reactions (Lindgaard et al., 2006), which can in turn shape the evolution of the

relationship between user and influencer, such as the likelihood and frequency of subsequent

encounters (Cafaro et al., 2016). Affective reactions can include perceived compatibility with

one’s self and social circle, also known as social attractiveness (Spence et al., 2019), while

behavioural reactions, such as willingness to affiliate (Philipp-Muller et al., 2020), can affect

an influencers’ engagement metrics (Taillon et al., 2020), and therefore their earnings

potential (Abidin, 2018). Although limited research exists to date looking at VIs, evidence

suggests that users have very different reactions to HIs vs. VIs, particularly when the VI has a

human-like appearance (Arsenyan & Mirowska, 2021) or when a transgression, such as a

politically incorrect statement, has been committed (Thomas & Fowler, 2021).
 7

3 Hypothesis Development

3.1 Need for Diversion and Emotional Contagion

Uses and gratification theory posits that people actively calibrate their social media

activity based on specific needs, such as the search for diversion from their offline lives

(Sheldon & Bryant, 2016; Yen & Chiang, 2021). To this end, social media, particularly

Instagram, are reported as important platforms for those seeking escape from their everyday

problems and experiences (Huang & Su, 2018). However, exposure to emotionally charged

content may actually hinder such escape, particularly for people who are highly susceptible to

being caught up in the emotions on display (Weiss & Cohen, 2019). Nevertheless, HIs often

play on emotional bonds with users to gain followers and build parasocial relationships to

increase their visibility and online status (Hwang & Zhang, 2018), and VIs similarly attempt

to use emotion to build bonds with followers (Arsenyan & Mirowska, 2021).

However, the UVH effect, where VIs who appear too human-like are dehumanised

and referred to as “robots” or “fake”, is pervasive on social media (Arsenyan & Mirowska,

2021). The use of such dehumanising language may point to expectations that any

emotionally charged content posted by a VI will not have the same effect as that posted by

fellow humans (Nowak et al., 2015). Therefore, we explore the role of empathy, and how it

may interact with knowledge about an influencer’s human vs. virtual origins, in predicting

individuals’ willingness to follow and attractiveness of HIs versus VIs.

The popularly used term empathy is seen as a “fundamental skill for navigating

everyday life” (Cameron et al., 2019, p. 4) and its absence has been linked to problematic

behaviours and depressive states (Carré et al., 2013). Empathy serves as an important

foundation in building a variety of relationships (Bickmore & Picard, 2005) and it is

considered a key factor in audience reactions to human versus virtual agents (Misselhorn,
 8

2009). Interestingly, Worral et al. (2021) found a lack of empathy when studying user

motivations to engage in online interactions, although in follow-up questioning some of these

users highlighted the empathetic dimension of their participation. These results suggest a

subconscious desire to “shut off” ones’ empathetic tendencies in certain contexts, something

that may be much more easily justifiable when faced with a non-human agent’s emotional

experiences

As a higher order construct which encompasses multiple dimensions (J. A. Hall &

Schwartz, 2019), empathy has traditionally been understood to include an affective and

cognitive dimension (D’Ambrosio et al., 2009; Wieseke et al., 2012). Affective empathy, or

emotional contagion, is defined as the “transfer of emotional states from one person to

another” (Czarna et al., 2015, p. 318; Verhagen et al., 2014). It is considered an integral part

of empathy, having been included in both two and three-dimensional models of the construct

(Carré et al., 2013; D’Ambrosio et al., 2009; Jolliffe & Farrington, 2006). It is considered an

automatic, bottom-up process, where individuals reproduce the emotions that another is

experiencing within themselves (Carré et al., 2013). Of course, this can refer to the

experiencing of both positive and negative emotions.

For individuals high on emotional contagion, who easily experience others’ affective

states, this empathetic reaction may block their ability to gain the true diversion benefits of

social media use. The above-mentioned dehumanisation of VIs may serve to dampen one’s

empathetic response, as the terms found to be often used to refer to VIs by Arsenyan and

Mirowska (2021) serve to place the VI in a category separate from the human user. This

distancing may work in favour of users high on emotional contagion, in that they will expect

to gain a greater escape experience from following a VI than a HI:

Hypothesis 1 (H1): Individual level emotional contagion will moderate the effects of

influencer type on a) willingness to follow and b) attractiveness of the influencer, such

 9

that those higher on emotional contagion will be more willing to follow the virtual

influencer and find her more attractive than the human influencer.

3.2 Resource Conservation and Cognitive Empathy

The second dimension of empathy, cognitive empathy, is the ability to understand and

reason about the affective states of others. Considered less automatic than contagion, this

dimension of empathy activates brain regions responsible for the processing of environmental

stimuli (Carré et al., 2013). Importantly, cognitive empathy should specifically involve an

understanding of emotional situations, not just a broader recreation of the emotional

experience or adoption of others’ positions (Bensalah et al., 2016).

Recent research has found that empathetic reactions are cognitively taxing (Cameron

et al., 2019) and it is well established that individuals prefer to conserve resources when their

expenditure is not expected to lead to resource gains (Hobfoll, 1989). As argued above, the

dehumanisation of VIs may serve to lighten the cognitive load required of users when

interacting with them, as their experience is seen as less real and therefore not requiring in-

depth processing or understanding. Therefore, in order to meet their goals of diversion, users

high on cognitive empathy may expect to experience a greater emotional escape as a result of

interacting with VIs rather than HIs:

Hypothesis 2 (H2): Individual level cognitive empathy will moderate the effects of

influencer type on a) willingness to follow and b) attractiveness of the influencer, such

that those higher on cognitive empathy will be more willing to follow the virtual

influencer and find her more attractive than the human influencer.

3.3 Escapism and Emotional Dissociation

More recently, scholars have expanded to a three dimensional model of empathy,

which has been validated with psychometric and neuropsychological evidence (Carré et al.,
 10

2013). The third dimension, emotional disassociation, is defined as the ability to disengage

and protect oneself from others’ affective experiences. As a top-down process, disassociation

is a self-regulatory process drawing on areas of the brain responsible for executive function

and the attending elements of self-control (Carré et al., 2013; Neumann et al., 2015; Yuan &

Raz, 2014). For those low on this dimension, exposure to others’ emotional experiences may

put them at risk for bearing the brunt of the emotional experience. Therefore, such individuals

may pursue “proactive person-environment interactions” (Mehl et al., 2006, p. 872),

preferring to put themselves into situations suited towards their personal characteristics (Wilt

& Revelle, 2019). In the case of a low ability to disassociate oneself from others’ emotional

experiences, the VI, given its non-human nature, may be a more attractive target of

engagement if their social media presence is expected to be either lower in emotional content

or less invasive as an emotional experience for the observer:

Hypothesis 3 (H3): Individual level emotional disassociation will moderate the effects

of influencer type on a) willingness to follow and b) attractiveness of the influencer,

such that those lower on emotional disassociation will be more willing to follow the

virtual influencer and find her more attractive than the human influencer.

3.4 Awareness

The above arguments depend on the social media user being able to discern an

influencer’s true nature. If an individual is seeking out profiles that will allow them some

degree of escape from the emotional entanglement that their high empathy provokes in

interactions with human profiles, then this escape is most likely to occur when the user is sure

that the profile is not a human. We therefore propose one final hypothesis:

Hypothesis 4 (H4): Hypotheses 1, 2, and 3 will only be supported when participants

are explicitly aware of an influencer’s human versus virtual nature.

 11

4 Methodology

We tested our hypotheses using a between-subjects, random assignment experimental

design. The main variable manipulated was exposure to visual content of an HI versus VI.

We conducted the experiment under two conditions: one where participants were made aware

of the human vs. virtual nature of the influencer they viewed, and another where this

information was not shared with participants.

4.1 Participants

Participants included 224 young adults who were given the opportunity to participate

in the study in exchange for points towards course credit on a university-wide research

participant recruitment platform. Ten participants reported enough familiarity with the social

media influencers to whom they were exposed that their responses were removed from the

study. The final sample consisted of 214 participants, majoring in Marketing, 37% female,

and on average 22 years old (s.d. = 1.67), which is in the typical age range for Instagram

users (Statista, 2019). All participants reported using some form of social media, with the

majority accessing their social media accounts daily (21%) or several times a day (76%).

Almost 80% of the participants reported spending at least one hour per day on social media,

with over a quarter reporting spending over 3 hours per day. Additionally, 86% of

participants reported using the target social media of this experiment, Instagram, regularly.

Please see Tables 1 and 2 for additional descriptive information about the final sample.

4.2 Procedure

4.2.1 Choice of Virtual and Human Influencers

For our experimental stimuli, we chose pictures posted on Instagram for several

reasons. Instagram is a highly visual platform (Casaló et al., 2020; Sheldon & Bryant, 2016;
 12

Waterloo et al., 2018), which includes more personal and intimate content based on one’s

personal (rather than relational) identity (Sheldon & Bryant, 2016). It is also the fastest

growing and most used social media (Sheldon & Bryant, 2016) and particularly popular with

influencers (Johnson et al., 2019). Finally, most users choose to maintain their information as

publicly visible on Instagram (Waterloo et al., 2018), making any profile accessible to

anyone.

Prior to launching the focal study, we asked a sample of 402 Instagram users recruited

from the Prolific online research participant pool (45.5% female, average age 23.31, SD =

4.00), to evaluate six potential VI profiles for human likeness, eeriness, and likeability

(Katsyri et al., 2017). All six of the VI profiles chosen depicted influencers presenting as

female. This was due to the fact that the majority of Instagram users and influencers are

female (Djafarova & Rushworth, 2017; Roberti, 2022). Additionally, at the time of this study,

there were few male virtual influencers active on social media, which would have unduly

limited the pool from which we could select an appropriate VI.

In line with the UVH (Mori, 1970), it was decided that the VI scoring the lowest on

these dimensions may be too easily confounded with a human, while the VI scoring the

highest on these dimensions was too obviously a virtual entity to allow for valid comparisons

with an HI. We therefore chose Lil’ Miquela (LM; Figure 1) as the focal VI given that her

scores on all three dimensions were closest to the average across all of the VIs.

Our next step was to identify an HI with a similar social media presence: single,

female, 18 to 25 years old with artistic endeavours, preferably living on the east coast of the

United States, whose posts focused primarily on lifestyle content targeting a young adult

audience. The authors independently identified six potential HI accounts. A pilot study was

then conducted, asking 151 Instagram users to evaluate the similarity of each profile in terms

of style, content, and expected follower characteristics to LM (44.5% female, average age
 13

23.54, SD = 4.23). This second pilot group was also recruited via Prolific, but it was ensured

that participants were different from those in the first pilot study outlined above. Participants

were asked to indicate the extent to which they agreed (1 = strongly disagree, 7 = strongly

agree) with the following statements: “These two influencers have a similar personal style”,

“These two influencers post similar content”, and “These two influencers have similar

followers”. The authors then proceeded to review the Instagram profiles of the chosen VI and

those HIs who were rated as ‘most similar’ on these criteria. The goal was to identify pictures

of typical everyday situations that projected consistency across the VI and HI personas.

It was decided that the pictures posted by the HI who was rated as most similar in

terms of style and followership (and second most similar in terms of content) were more

sexually suggestive than the content posted by LM, making it difficult to identify pictures

that were similar enough to avoid potential contamination effects. Consequently, this HI was

eliminated from further consideration. We then proceeded to view the profile of the HI who

was rated by the participants as second most similar in terms of style and followership, and

first in terms of content. Upon closer comparative examination, it was decided that, given

their predominant emphasis on fashion and beauty, this HIs pictures were similarly

incompatible with the types of pictures posted on LM’s profile. For all three criteria, the third

most similar profile was that of Meghan DeAngelis (MD; Figure 2). The authors reviewed

her profile and determined that there were enough pictures of similar style and content to

enable the creation of paired visual stimuli for LM and MD, in order to ensure the required

consistency across the VI and HI personas. We then proceeded to identify specific images

posted by both influencers in their accounts, with our initial screening resulting in 24 pairs of

pictures. Upon further review and discussion, a final set of 6 photos were agreed upon to be

used as experimental stimuli, given the similarities of the pictures in terms of context and
 14

comparable situations across the two influencers. Please see Appendix A for list of themes in

the pictures and the set of photos displayed to participants.

4.2.2 Details of Participation in Focal Experiment

Participation took place over two online sessions one week apart, in order to minimise

the risk of common method bias (MacKenzie & Podsakoff, 2012). In session one, participants

were asked to answer questions about their social media usage and demographic

characteristics, and to complete the basic empathy scale (see below). A week after the first

session, the link for the second session was made available to those who had successfully

completed the first part of the study. At this stage, participants were randomly assigned to

one of four experimental conditions (“human vs. virtual influencer” X “know vs. don’t know

the influencer’s true nature”). Participants assigned to the “know” condition were told,

“Below are several pictures of a social media influencer who is a real person (virtual

personality) currently active on several social media platforms”. In contrast, those in the

“don’t know” condition were told: “Below are several pictures of a social media

influencer who is currently active on several social media platforms”. Participants were then

shown a collage of the six chosen photos of the respective influencer (Appendix A). After

spending a minimum of one minute looking at the pictures, participants were asked to write a

short description of each picture, which served as an attention check. As a manipulation

check, participants in both conditions were asked whether the influencer they saw was a

human or a virtual personality. In order to avoid any potential priming effects, this latter

question only appeared at the very end of the survey for participants in the “don’t know”

condition.
 15

4.2.3 Measures

Unless otherwise indicated, all variables were measured on a 1 to 7 Strongly

Disagree/Strongly Agree Likert scale.

Willingness to Follow. A single-item measure was employed to assess willingness to

follow the influencer (“I would be willing to follow her using my personal social media

account”), approximating a public representation of engagement with the influencer that is

visible to others within one’s social media network. We chose to use a single-item measure

given that they have been found to be equivalent to multiple-item scales (Fisher et al., 2016)

and the fact that we were measuring a specific behavioural intention, rather than a latent

construct.

Attractiveness. In order to capture the importance of attractiveness in determining

followership on social media (Djafarova & Rushworth, 2017) we also measured social

attractiveness (5 items; sample item: “I think she could be a friend of mine”; α = .75) and

physical attractiveness (7 items; sample item: “I don’t like the way she looks”; reverse coded;

α = .85) using scales adapted from McCroskey and McCain (1974).

Basic Empathy Scale. We used the basic empathy scale (BES), developed by Carré

et al. (2013), which measures empathy along three dimensions: emotional contagion (6 items;

sample item = “I get caught up in other people’s feelings easily”; α = .73); cognitive empathy

(8 items; sample item = “When someone is feeling ‘down’ I can usually understand how they

feel”; α = .681); and emotional disassociation (6 items; sample item = “My friends’ emotions

don’t affect me much”; α = .80).

1A closer inspection of these results showed that the removal of one item in the cognitive empathy
subscale, “I can usually work out when my friends are scared”, would increase the reliability to α =
0.74. We re-ran all of the analysis with this attenuated cognitive empathy variable; however, this did
not significantly change the results reported below. We therefore proceed to report the results using
the full cognitive empathy scale.
 16

Control Variables. Participants were asked to provide information about their age

and gender (coded as Male = 1). We wanted to control for Instagram usage but were

cognizant of the risk of priming participants to the purpose of the study. Therefore, instead of

asking directly whether they were Instagram users, we asked them to choose the social media

platforms they frequented regularly from a list of popular social media sites. This information

was then coded as 1 for those reporting using Instagram regularly, and 0 otherwise.

5 Results

For simplicity of presentation, and given the potentially different cognitive processes

involved, we split the data for analysis, first examining those participants who did not know

the influencers’ true nature, and then analysing those who did have this information. We used

stepwise regression to test the relationships between our variables, as this allowed us to

examine not only the overall interaction effects but also any incremental effects of adding

control variables, independent variables, and the interaction terms. We entered age, gender,

and the binary Instagram use variable as controls in the first step, followed by the

independent variable(s) of interest and finally any interaction terms. All continuous variables

were mean-centred prior to analysis (Aiken et al., 2003), and we used Dawson’s (2014)

recommendations for evaluating simple slopes when investigating visual graphs of interaction

effects.

5.1 Manipulation & Attention Checks

In the condition where participants were aware of the influencer’s true nature, a

manipulation check was included after participants had viewed the influencer photos, asking

whether the influencer was human or virtual. Any participant who answered this question

incorrectly was not permitted to proceed any further to complete the experiment. To avoid the

risk of participants guessing the purpose of the experiment, any participant who attempted the
 17

study again after having been removed had their responses deleted from the final data set. In

the condition where participants were not informed about the true nature of the influencer, for

those who were shown pictures of the human influencer, 77% correctly guessed that she was

human; for those who were shown pictures of the virtual influencer, 46% guessed correctly

that she was a virtual personality. In both conditions, participants were also asked to briefly

describe the pictures they were shown. A visual inspection of these results did not identify

any participants who raised suspicion of low attention during their participation.

5.2 Hypothesis Testing in the “Know” Condition

Prior to formal hypothesis testing, we checked whether the influencer condition

(human vs. virtual) had a direct effect on any of the dependent variables. As Table 3 shows,

the experimental condition was only significant for social attractiveness, with the VI being

more highly rated than the H1 (β = 0.24, p < .05).

To test H1, we interacted the human vs. virtual influencer condition with emotional

contagion scores. As Table 4 shows, this interaction term was approaching significance for

willingness to follow (β = 0.25, p < .10). Graphing these results in Figure 3 shows that H1a is

supported: participants scoring low on emotional contagion showed no preference for either

influencer (simple slope = -0.11, n.s.), however those scoring high on emotional contagion

reported a significantly higher willingness to follow the virtual influencer (simple slope =

1.18, p < .05). There was no significant effect of emotional contagion on perceptions of

neither social nor physical attractiveness, leaving H1b unsupported.

To evaluate H2, Table 5 shows the coefficients on the influencer type X cognitive

empathy interaction terms, which was significant for willingness to follow (β = 0.37, p < .01)

and social attractiveness (β = 0.31, p < .05). The graphs for both of these results show similar

patterns and are presented in figures 4 and 5. Participants low on cognitive empathy do not
 18

show a differential willingness to follow (slope = -0.47, n.s.) or perceptions of social

attractiveness (simple slope = .06, n.s.) for either type of influencer. However, those who

score high on cognitive empathy report significantly higher willingness to follow the VI

(simple slope = 1.35, p < .01) and see her as more socially attractive (simple slope = 0.95, p <

.01). Therefore, H2a and H2b were supported.

Table 6 shows the results of the regression analysis with influencer type interacted

with emotional disassociation, in order to test H3. We see a significant effect of the

interaction for willingness to follow (β = -0.34, p < .01) and social attractiveness (β = -0.24, p

= .05), supporting H3a and H3b. Figures 6 shows that, as predicted, participants who scored

high on emotional disassociation showed no difference in willingness to follow either type of

influencer (simple slope = -.43, n.s.). However, those scoring lower on this type of empathy

are significantly more likely to follow the virtual influencer (simple slope = 1.57, p < .01).

Figure 7 shows similar results for social attractiveness: high emotional disassociation

individuals showed no preference (simple slope = .09, n.s.), while those low on emotional

disassociation found the VI significantly more socially attractive (simple slope: 0.92, p <

.01).

5.3 Hypothesis Testing in the “Don’t Know” Condition

In addition to the control variables included in the above analyses, we also included a

binary variable to capture whether participants correctly guessed the influencers’ true nature

(see manipulation check section above). As in the know condition, we first looked at whether

influencer type had an effect on any of the dependent variables, with no coefficients on the

experimental condition showing significance in any of the regressions.

Additionally, the interaction of influencer type with each type of empathy only

approached significance in the case of emotional disassociation and willingness to follow (β

 19

= -0.27, p < .10). However, graphing these results showed that neither of the slopes were

significant. Therefore, H4 is supported in that none of the effects seen in the “know”

condition were recreated in the “don’t know” condition2.

A summary of the hypotheses and results is shown in Table 7.

6 Discussion

6.1 Sweet Escape

Our results show that individual differences in empathy may play an important role in

engagement with human versus virtual agents in the context of social media. When

participants are aware of the true nature of the agent, individuals scoring high on emotional

contagion and cognitive empathy and low on emotional disassociation show a greater

willingness to follow the VI. Additionally, those scoring high on cognitive empathy and low

on emotional disassociation believed the VI would be a better fit with their social circle

(social attractiveness). Overall, these findings are consistent with an “escapism effect”, where

the VI is expected to provide greater diversionary benefits from everyday human emotional

experiences and require fewer cognitive resources in the form of emotional sense making.

This effect manifests through a preference for VIs, with their authentically fake storylines

over HIs, with authentic, human ones.

In the case of cognitive empathy and emotional disassociation, the same patterns were

found for social attractiveness; these are the two dimensions of empathy with the greatest

long-term consequence, as they may affect cognitive resources available for other activities

and long-term emotional experiences, respectively. The construct of social attractiveness, as

measured here, captures the willingness to interact with the focal person and/or include them

2 In the interest of brevity, detailed results for this analysis are not presented here but are available
from the corresponding author upon request.
 20

in one’s social circle; these results may indicate an even stronger escapism effect, in that the

expected lower emotional entanglements of the VI may even gain them (hypothetical) access

to users’ social circles. Social identity theory postulates that individuals will want to be

affiliated with others who bring benefits to the individual’s social standing (Jin, 2018; Tajfel

& Turner, 1986). Although such social identity effects were outside the scope of this study, it

is possible that these inclusion criteria may also include benefits to the individual that extend

beyond their external social status – e.g., lower empathetic responses.

There was no effect of any of the three empathy dimensions on physical attractiveness

ratings. This is not altogether surprising, as physical attractiveness is a cognitive evaluation

of the target’s physical appearance, which may not have a clear link to behavioural intentions

(Santiago & Serralha, 2022), nor intention to interact with the influencer in the future

(Sokolova & Kefi, 2020). Therefore, in the context of empathetic reactions, there is no real

emotional consequence to judgements of physical attractiveness. Additionally, the limited

exposure to the influencers in restricted contexts, and using pictures that were clearly chosen

(by the influencer) to be attractive may have limited the variability in physical attractiveness

ratings, thereby precluding any differences across conditions or sub-groups.

6.2 Disclosing Virtuality

As expected, all of these relationships disappeared when participants were not aware

of the influencer’s nature. This again supports the existence of the escapism effect, that when

participants are aware that the influencer may offer a less realistic emotional experience, they

show a preference for engaging with that influencer. Importantly, however, simple suspicion

about the influencer’s nature does not appear to be enough; the VI presented in the

experiment is ambiguous enough to raise doubt regarding her true nature (Arsenyan &

Mirowska, 2021). Yet, such doubt was not enough to engage empathy to alter reactions. Only
 21

when participants were explicitly aware of her virtual nature were they able to indulge in her

profile with confidence that a less intense emotional experience, and therefore an escape from

the emotional entanglements of their daily lives, awaited them.

These results also provide further support to earlier work that pitted the Computer Are

Social Actors Paradigm (CASA; Nass & Moon, 2000) against the UVH in the context of HIs

vs VIs (Arsenyan & Mirowska, 2021). Although under CASA we would expect VIs to be

treated similarly to their human counterparts, as in this prior research, we see that this is not

the case here. Given the lack of consensus in recent CASA studies studying digital entities on

social media (Meng & Dai, 2021; Park et al., 2021; Spence et al., 2019), the specific context

of social media and user pursuit of diversion in this realm may act as a boundary condition to

the treatment of digital agents as humans (Nowak et al., 2015).

The current work also sheds a more favourable light on some potential consequences

of the UVH. Although the virtual agent may elicit some feelings of creepiness and negative

emotions (Arsenyan & Mirowska, 2021; Mori, 1970), there may be an upside to these

reactions in the form of greater diversionary benefits. The virtual agent with a high human

resemblance, but who maintains their virtual status, may provide just enough reality to allow

for users to be transported into the virtual agents’ world (Brown, 2015), but also allow the

user to dehumanise or distance themselves just enough to protect themselves from the

emotional repercussions of positive, affiliative reactions.

Worral et al. (2021) have called for the recognition of a social-emotional paradigm in

studying user motivations to engage in online communities. We echo this call and add to it a

consideration of the coherence of existing norms and practices with the various profiles of

potential other users – human vs. virtual, and in the latter case the degree of autonomy – in

studying engagement, information-sharing, and social and emotional support in online social

contexts. Much has been written of the potential negative effects of social media exposure,
 22

particularly on young adults and their emotional and psychological well-being (Boer et al.,

2022; Schemer et al., 2021; Schmuck, 2021; Valkenburg et al., 2022). However, our results

imply that the presence of virtual agents in human social networks may require a

reconsideration of the exact role of individual differences (Nowak et al., 2015) such as self-

esteem (Djafarova & Rushworth, 2017; Schmuck, 2021), empathy, or shyness (Birnie &

Horvath, 2002; von der Pütten et al., 2010) in the study of social media usage and

consequences. Between human users, empathy serves to build connections (Vossen &

Valkenburg, 2016) and parasocial relationships (PSR; Hwang & Zhang, 2018) by allowing

users to relate to each other; however, between human and virtual users, it seems that

empathy builds connections not via relatedness, but via the escape potential of interacting

with a non-human agent. Thus, empathy is now seen as not necessarily an asset but a

potential liability, while virtual agents become not just a robotic presence but also a possible

safe haven that has the potential to buffer vulnerable users from some of the detrimental

influence of social media influencers (Vossen & Valkenburg, 2016).

6.3 Practical Implications

These results have practical implications for individuals and companies looking to

leverage the power of influencers, human or virtual, to connect with their target audience via

social media. For individuals who are not yet followers of a given VI, trying to play on

empathy may backfire. VIs may be expected to have less rich emotional lives, or to elicit less

genuine empathetic reactions, which may explain part of their appeal. Trait activation theory

(Tett & Guterman, 2000) argues that certain dispositional traits are only activated when

necessary given the context. Highly emotional content, although intended to build

relationships with the audience, may actually turn off potential followers by overly activating

empathetic responses; the presence of the escapism effect may mean that users want a break
 23

from empathising with others, and the associated increased cognitive load (Cameron et al.,

2019).

Another implication is that, to leverage this escapism effect of VIs, their true nature

should be clearly advertised. Although this may appear to be a counterintuitive

recommendation, VIs that “pass” for human may come with the same expected emotional

baggage as HIs. High-empathy potential followers may appreciate the assurance of a less

authentic emotional experience as a way to minimise their emotional investment in the

influencer and their storylines. This recommendation also builds on Arsenyan and

Mirowska’s (2021) findings with respect to the UVH, where the unambiguously non-human

influencer received positive reactions comparable to the HI, but the more ambiguous, human-

like VI received much more negative reactions.

A further implication is that, if individuals are seeking an escapism effect from social

media, messages concerning emotionally charged content should be carefully calibrated and

transmitted. Highly emotionally charged content, particularly from human influencers, may

lead to “emotional overload” where attentional processes are disrupted and the intended

message may be filtered out as a protective mechanism (Ytre-Arne & Moe, 2021). If highly

empathetic individuals avoid such content or profiles that they believe tend to display such

content, then such messaging may need to be transmitted via different channels or methods.

Conversely, it appears that ratings of physical attractiveness are immune to the effect

of individual level empathy. Therefore, highly emotional content that is meant to elicit

positive visual evaluations only may be appropriate. However, the stimuli chosen to convey

such messaging (e.g., phonological, lexical, grammatical, denotational; Cardona-Rivera et al.,

2020) should be carefully considered to avoid any unintended consequences.

Finally, Senft (2013) argues that becoming a celebrity requires the transformation

from “citizen to corporation” (p. 351). For the human influencer, this happens in the form of
 24

self-branding, where the product being proposed is intimacy with the influencer, and the price

being asked is follower attention. For this exchange to be successful, followers need to be

emotionally invested in the influencers’ life and narratives. Although questions have been

raised about the potential economic, legal, and social consequences of such exchanges

(Abidin, 2018), the presence of humans on both sides of the transaction makes things rather

straight-forward to understand. VI’s on the other hand, are commodifying intimacy with, by

their very nature, fake storylines and fake people. As the results of the current research show,

there may be sub-groups of social media users who will be more drawn to this “authentically

fake” storytelling, which opens up additional questions of potential real-world consequences

arising from virtual-world events.

For instance, reactions to highly emotional content for certain categories of

individuals may have far-reaching consequences in terms of the types of narratives that are

shared and valorised in the wider social media realm, particularly as online activists rely on

users’ ability to empathise with their cause (Martínez García, 2020). Senft (2013) argues that

online identity is no longer the property of the individual, but now resides with the

“perceiver”, taking the power of dissemination away from the storyteller and granting it to

the receiver of this narrative, in line with the received narrative argument from Castricato et

al. (2021). As algorithms become “key factors in how attention is distributed” (Marwick,

2018, p. 163), the individuals and narratives that garner more attention online will find

themselves in a virtuous cycle of follower growth, leading to greater market value and

income-generating opportunities for the influencer (Abidin, 2018). There are two potential

consequences of this process that we believe deserve highlighting. On the one hand, highly

emotionally charged content recounting difficult or unjust circumstances may be ignored in

favour of more frivolous fare, particularly as algorithms adapt to preferences among highly

empathetic individuals. However, if this is also the same sub-population more likely to
 25

engage in offline action to counter injustices or difficulties (Craddock, 2020), then this

activism may end up muted in the long run. Alternatively, if virtual influencers have an

advantage in garnering attention for highly emotional content amongst this same sub-group of

social media users, then the causes that become visible on social media (and, potentially,

acted upon offline) may be decided by the corporations (and potentially their algorithms)

behind these virtual personas rather than by the human users of the platform. Future research

should draw on theories and techniques within anthropology, sociology, race and gender

studies, media and cultures studies (Senft, 2013) to consider such potential long-term and

offline effects.

6.4 Limitations

The findings of this study are limited to the initial perception of the virtual agents, as

we purposely did not sample followers of the focal influencers in our study. Although this

was by design, it means that our study does not capture the reduced uncertainty and increased

trust required in the human-agent interaction (B. Liu, 2021) that may develop as a result of

sustained interactions (Bickmore & Picard, 2005). Therefore, the present study is not able to

study the differences in longer-term outcomes of user-HI vs. user-VI interactions, such as

PSR formation. Given that empathy is an interpersonal skill that helps to build relationships

(Cameron et al., 2019), including parasocial ones (Hartmann, 2017; Hwang & Zhang, 2018),

it is possible that the propensity for higher empathy participants to follow VIs indicates a

potential budding PSR. However, PSR requires parasocial interactions over time, which we

were not able to capture in the current study.

Additionally, the limited experimental stimuli of one human and one virtual influencer

and the exclusion of participants who were familiar with the focal influencers means we are

not able to consider any effects of persona popularity or social status markers on willingness
 26

to interact with a virtual influencer on a publicly visible social media platform. Interestingly,

social attractiveness was the only dependent variable significantly affected by the

experimental manipulation, with the virtual influencer being seen as more socially attractive.

Future longitudinal studies should consider how social identity related characteristics such as

social standing, popularity, or social image and their expected benefits may affect

engagement with human versus virtual influencers.

Another limitation is the use of pictures in the experiment, as opposed to videos or

written content. This was by design to minimise potential confounding effects of text

interpretation or video playback issues, and to create a stimulus very close to the majority of

content on Instagram. However, as non-verbal behaviour and attitudes have been found to

elicit positive impressions in virtual human-agent encounters (Cafaro et al., 2016), a

replication of this study with videos would allow for an exploration of the potential

moderating effect of the first encounter medium.

Our data collection relied on a young adult, student population, who participated in an

online survey. Although this sample was representative of the demographics of Instagram

users, we acknowledge that the implications of this study may not be appropriately

extrapolated to other user populations or social media platforms. Furthermore, researchers'

choices pertaining to which virtual and human influencers and stimuli were included in the

pilot study and final experimental materials may reflect unconscious biases arising from our

specific backgrounds and experiences (Das et al., 2022). For instance, deciding to exclude

one of the “more similar” human influencers due to what we perceived to be overly

sexualized content means that the stimuli ultimately chosen for the final experiment are

moderately explicit at most. Although we acknowledge that it is not possible to eliminate all

such biases, all attempts were made to develop the materials, collect, analyse, and report data

as objectively as possible. The difficulty that we found in matching up the highly sexualized
 27

human profile with the virtual one may also indicate a potential boundary condition for

virtual influencers’ mimicking of human social media content, as well as interesting potential

cross-cultural differences. For example, developers may be simultaneously relying on

common social taboos against sexual intimacy with inanimate objects (Su et al., 2019) and

the acceptability of emotional attachment to inanimate objects (Sharabiani, 2021) to allow

their VIs to gain popularity and followers. However, if similar norms do not hold in other

cultural contexts (Motschenbacher, 2018), then this approach may perpetuate the

Anglocentrism of studies of internet-based phenomenon (Marwick, 2018).

Additionally, although several steps were taken to minimise the risk of common

method variance (CMV) as a driver of the results, it is still possible that some CMV effects

persisted if participants recalled their participation from one part of the study to another.

However, the finding of several significant moderation results reduces CMV concerns, given

the difficulty of identifying moderation results in its presence (Siemsen et al., 2010).

Finally, one general limitation in current VI research is the limited knowledge that the

public and academics have regarding the mechanisms that operate behind the virtual

influencers - i.e. whether their content is manually authored, automatically generated or some

hybrid methodology. The current results, therefore, rely on individuals’ personal beliefs about

how a “virtual influencer” operates, which may interact with their own ideas, beliefs, and

fears regarding artificial intelligence and automated systems.

7 Conclusion

This study contributes to research into interactions with virtual agents by considering

the role of empathy as an individual difference variable predicting preferences for human vs.

virtual influencers in a social media context. Results show that there is no difference in the

initial perception of the agent with regards to their human counterpart when the virtual nature
 28

of the agent is not disclosed to the user. When the virtual agent is clearly described as such,

we observe an escapism effect, where users turn to virtual influencers as a sweet escape from

the potentially excessive emotional experience of interacting with fellow humans. These

findings have implications for the development, deployment, and employment of virtual

influencers, as well as the understanding of proliferation of highly emotionally charged

content in social media settings. We believe that this research provides insight for future

virtual influencer, as well as virtual agent research, making a case for the importance of

empathy in engaging with these digital entities.

Future research should investigate the emotional experiences of individuals, both

when choosing which influencers to follow and when interacting with their chosen

influencers on social media. Diary studies or interviews could be utilised to explore these

interactions in more depth. Additionally, the potential role of virtual agents in offsetting some

of the established negative effects of social media usage, such as low self-esteem, excessive

social comparisons, or depression (Alfasi, 2019), represents a fertile area of study. As AI

technologies and, by extension, virtual agents, continue to evolve, insights into automated

story generation (e.g., Riedl, 2010) may allow for new methods of exploring human

engagement with the interactive storylines generated by virtual influencers.

Finally, if the escapism effects embodied by a willingness to follow a VI represents a

defensive psychological mechanism (Labrecque et al., 2011), then future research should

explore whether highly empathetic followers of VIs are more likely to dehumanise them than

low empathetic followers, implying that such defence mechanisms may be deployed

collectively for greater psychological protection.

 29

8 References

Abidin, C. (2018). Internet celebrity: Understanding fame online (First edition). Emerald Publishing.

Ahn, R. J., Cho, S. Y., & Sunny Tsai, W. (2022). Demystifying Computer-Generated Imagery (CGI)

Influencers: The Effect of Perceived Anthropomorphism and Social Presence on Brand

Outcomes. Journal of Interactive Advertising, 22(3), 327–335.

https://doi.org/10.1080/15252019.2022.2111242

Aiken, L. S., West, S. G., & Reno, R. R. (2003). Multiple regression: Testing and interpreting

interactions (Nachdr.). SAGE.

Alfasi, Y. (2019). The grass is always greener on my Friends’ profiles: The effect of Facebook social

comparison on state self-esteem and depression. Personality and Individual Differences, 147,

111–117. https://doi.org/10.1016/j.paid.2019.04.032

Arsenyan, J., & Mirowska, A. (2021). Almost human? A comparative case study on the social media

presence of virtual influencers. International Journal of Human-Computer Studies, 155,

102694. https://doi.org/10.1016/j.ijhcs.2021.102694

Bensalah, L., Stefaniak, N., Carre, A., & Besche-Richard, C. (2016). The Basic Empathy Scale

adapted to French middle childhood: Structure and development of empathy. Behavior

Research Methods, 48, 1410–1420. https://doi.org/10.3758/s13428-015-0650-8

Bickmore, T. W., & Picard, R. W. (2005). Establishing and maintaining long-term human-computer

relationships. ACM Transactions on Computer-Human Interaction, 12(2), 293–327.

https://doi.org/10.1145/1067860.1067867

Birnie, S. A., & Horvath, P. (2002). Psychological predictors of internet social communication.

Journal of Computer-Mediated Communication, 7(4), 0–0. https://doi.org/10.1111/j.1083-

6101.2002.tb00154.x

Boer, M., Stevens, G. W. J. M., Finkenauer, C., & van den Eijnden, R. J. J. M. (2022). The complex

association between social media use intensity and adolescent wellbeing: A longitudinal

investigation of five factors that may affect the association. Computers in Human Behavior,

128, 107084. https://doi.org/10.1016/j.chb.2021.107084

 30

Brown, W. J. (2015). Examining four processes of audience involvement with media personae:

Transportation, parasocial interaction, identification, and worship. Communication Theory,

25, 259–283. https://doi.org/10.1111/comt.12053

Cafaro, A., Vilhjálmsson, H. H., & Bickmore, T. (2016). First impressions in human-agent virtual

encounters. ACM Transactions on Computer-Human Interaction, 23(4), 1–40.

https://doi.org/10.1145/2940325

Cameron, C. D., Hutcherson, C. A., Ferguson, A. M., Scheffer, J. A., Hadjiandreou, E., & Inzlicht, M.

(2019). Empathy is hard work: People choose to avoid empathy because of its cognitive costs.

Journal of Experimental Psychology: General, 148(6), 962–976.

https://doi.org/10.1037/xge0000595

Cardona-Rivera, R. E., Zagal, J. P., & Debus, M. S. (2020). GFI: A Formal Approach to Narrative

Design and Game Research. In A.-G. Bosser, D. E. Millard, & C. Hargood (Eds.), Interactive

Storytelling (Vol. 12497, pp. 133–148). Springer International Publishing.

https://doi.org/10.1007/978-3-030-62516-0_13

Carré, A., Stefaniak, N., D’Ambrosio, F., Bensalah, L., & Besche-Richard, C. (2013). The basic

empathy scale in adults (BES-A): Factor structure of a revised form. Psychological

Assessment, 25(3), 679–691. https://doi.org/10.1037/a0032297

Casaló, L. V., Flavián, C., & Ibáñez-Sánchez, S. (2020). Influencers on Instagram: Antecedents and

consequences of opinion leadership. Journal of Business Research, 117, 510–519.

https://doi.org/10.1016/j.jbusres.2018.07.005

Castricato, L., Biderman, S., Cardona-Rivera, R. E., & Thue, D. (2021). Towards a Formal Model of

Narratives (arXiv:2103.12872). arXiv. http://arxiv.org/abs/2103.12872

Chang, T. (2010). Gaming will save us all. Communications of the ACM, 53(3), 22–24.

https://doi.org/10.1145/1666420.1666431

Craddock, E. (2020). The Affective, the Normative and the Everyday: Exploring What Motivates and

Sustains Anti-Austerity Activism. In E. Craddock (Ed.), Living Against Austerity: A Feminist

Investigation of Doing Activism and Being Activist (p. 0). Policy Press.

https://doi.org/10.1332/policypress/9781529205701.003.0004
 31

Czarna, A. Z., Wróbel, M., Dufner, M., & Zeigler-Hill, V. (2015). Narcissism and Emotional

Contagion: Do Narcissists “Catch” the Emotions of Others? Social Psychological and

Personality Science, 6(3), Article 3. https://doi.org/10.1177/1948550614559652

D’Ambrosio, F., Olivier, M., Didon, D., & Besche, C. (2009). The basic empathy scale: A French

validation of a measure of empathy in youth. Personality and Individual Differences, 46,

160–165. https://doi.org/10.1016/j.paid.2008.09.020

Das, D., Podder, A., & Semaan, B. (2022). Note: A Sociomaterial Perspective on Trace Data

Collection: Strategies for Democratizing and Limiting Bias. ACM SIGCAS/SIGCHI

Conference on Computing and Sustainable Societies (COMPASS), 569–573.

https://doi.org/10.1145/3530190.3534835

Dawson, J. F. (2014). Moderation in Management Research: What, Why, When, and How. Journal of

Business and Psychology, 29, 1–19. https://doi.org/10.1007/s10869-013-9308-7

Djafarova, E., & Bowes, T. (2021). ‘Instagram made Me buy it’: Generation Z impulse purchases in

fashion industry. Journal of Retailing and Consumer Services, 59, 102345.

https://doi.org/10.1016/j.jretconser.2020.102345

Djafarova, E., & Rushworth, C. (2017). Exploring the credibility of online celebrities’ Instagram

profiles in influencing the purchase decisions of young female users. Computers in Human

Behavior, 68, 1–7. https://doi.org/10.1016/j.chb.2016.11.009

Fisher, G. G., Matthews, R. A., & Gibbons, A. M. (2016). Developing and investigating the use of

single-item measures in organizational research. Journal of Occupational Health Psychology,

21(1), 3–23. https://doi.org/10.1037/a0039139

Hall, J. A., & Schwartz, R. (2019). Empathy present and future. The Journal of Social Psychology,

159(3), Article 3. https://doi.org/10.1080/00224545.2018.1477442

Hanus, M. D., & Fox, J. (2015). Persuasive avatars: The effects of customizing a virtual salesperson‫׳‬s

appearance on brand liking and purchase intentions. International Journal of Human-

Computer Studies, 84, 33–40. https://doi.org/10.1016/j.ijhcs.2015.07.004

 32

Hartmann, T. (2017). Parasocial interaction, parasocial relationships, and well-being. In The

Routledge handbook of media use and well-being: International perspectives on theory and

research on positive media effects (pp. 131–144). Routledge/Taylor & Francis Group.

Hobfoll, S. E. (1989). Conservation of resources: A new attempt at conceptualizing stress. American

Psychologist, 44(3), 513–524. https://doi.org/10.1037/0003-066X.44.3.513

Huang, Y.-T., & Su, S.-F. (2018). Motives for Instagram Use and Topics of Interest among Young

Adults. Future Internet, 10(8), 77. https://doi.org/10.3390/fi10080077

Hwang, K., & Zhang, Q. (2018). Influence of parasocial relationship between digital celebrities and

their followers on followers’ purchase and electronic word-of-mouth intentions, and

persuasion knowledge. Computers in Human Behavior, 87, 155–173.

https://doi.org/10.1016/j.chb.2018.05.029

Jin, S. V. (2018). “Celebrity 2.0 and beyond!” Effects of Facebook profile sources on social

networking advertising. Computers in Human Behavior, 79, 154–168.

https://doi.org/10.1016/j.chb.2017.10.033

Johnson, B. K., Potocki, B., & Veldhuis, J. (2019). Is that my friend or an advert? The effectiveness

of Instagram native advertisements posing as social posts. Journal of Computer-Mediated

Communication, 24, 108–125. https://doi.org/10.1093/jcmc/zmz003

Jolliffe, D., & Farrington, D. P. (2006). Development and validation of the Basic Empathy Scale.

Journal of Adolescence, 29(4), Article 4. https://doi.org/10.1016/j.adolescence.2005.08.010

Katsyri, J., Makarainen, M., & Takala, T. (2017). Testing the ‘uncanny valley’ hypothesis in semi

realistic computer-animated film characters: An empirical evaluation of natural film stimuli.

International Journal of Human-Computer Studies, 97, 149–161.

https://doi.org/10.1016/j.ijhcs.2016.09.010

Khamis, S., Ang, L., & Welling, R. (2016). Self-branding, ‘micro-celebrity’ and the rise of Social

Media Influencers. Celebrity Studies, 8(2), 191–208.

https://doi.org/10.1080/19392397.2016.1218292
 33

Labrecque, L. i. ( 1 ), Krishen, A. s. ( 2 ), & Grzeskowiak, S. ( 3 ). (2011). Exploring social

motivations for brand loyalty: Conformity versus escapism. Journal of Brand Management,

18(7), 457–472. https://doi.org/10.1057/bm.2011.12

Lindgaard, G., Fernandes, G., Dudek, C., & Browñ, J. (2006). Attention web designers: You have 50

milliseconds to make a good first impression! Behaviour & Information Technology, 25(2),

115–126. https://doi.org/10.1080/01449290500330448

Liu, B. (2021). In AI We Trust? Effects of Agency Locus and Transparency on Uncertainty Reduction

in Human–AI Interaction. Journal of Computer-Mediated Communication, 26(6), 384–402.

https://doi.org/10.1093/jcmc/zmab013

Liu, X., Min, Q., & Han, S. (2020). Understanding users’ continuous content contribution behaviours

on microblogs: An integrated perspective of uses and gratification theory and social influence

theory. Behaviour & Information Technology, 39(5), 525–543.

https://doi.org/10.1080/0144929X.2019.1603326

MacDorman, K. F. (2019). In the uncanny valley, transportation predicts narrative enjoyment more

than empathy, but only for the tragic hero. Computers in Human Behavior, 94, 140–153.

https://doi.org/10.1016/j.chb.2019.01.011

MacKenzie, S. B., & Podsakoff, P. M. (2012). Common method bias in marketing: Causes,

mechanisms, and procedural remedies. Journal of Retailing, 88, 542–555.

https://doi.org/10.1016/j.jretai.2012.08.001

Martínez García, A. B. (2020). Women activists’ strategies of online self-presentation. AI &

SOCIETY. https://doi.org/10.1007/s00146-020-01102-y

Marwick, A. E. (2018). The Algorithmic Celebrity: The Future of Internet Fame and Microcelebrity

Studies: The Future of Internet Fame and Microcelebrity Studies. In C. Abidin & M. L.

Brown (Eds.), Microcelebrity Around the Globe (pp. 161–169). Emerald Publishing Limited.

https://doi.org/10.1108/978-1-78756-749-820181015

McCroskey, J. C., & McCain, T. A. (1974). The measurement of interpersonal attraction. Speech

Monographs, 41(3), 261–266. https://doi.org/10.1080/03637757409375845

 34

McLaughlin, C., Haverila, K., & Haverila, M. (2022). Gratifications sought versus gratifications

achieved in online brand communities: Satisfaction and motives of lurkers and posters.

Journal of Brand Management, 29(2), 190–207. https://doi.org/10.1057/s41262-021-00262-6

Mehl, M. R., Gosling, S. D., & Pennebaker, J. W. (2006). Personality in its natural habitat:

Manifestations and implicit folk theories of personality in daily life. Journal of Personality

and Social Psychology, 90(5), Article 5. https://doi.org/10.1037/0022-3514.90.5.862

Meng, J., & Dai, Y. (Nancy). (2021). Emotional support from AI chatbots: Should a supportive

partner self-disclose or not? Journal of Computer-Mediated Communication, 26(4), 207–222.

https://doi.org/10.1093/jcmc/zmab005

Misselhorn, C. (2009). Empathy with Inanimate Objects and the Uncanny Valley. Minds and

Machines, 19(3), 345–359. https://doi.org/10.1007/s11023-009-9158-2

Mori, M. (1970). Bukimi no tani [the uncanny valley]. Energy, 7, 33–35.

Mori, M., MacDorman, K., & Kageki, N. (2012). The Uncanny Valley [From the Field]. IEEE

Robotics & Automation Magazine, 19(2), 98–100.

https://doi.org/10.1109/MRA.2012.2192811

Motschenbacher, H. (2018). Language and Sexual Normativity. In K. Hall & R. Barrett (Eds.), The

Oxford Handbook of Language and Sexuality (1st ed.). Oxford University Press.

https://doi.org/10.1093/oxfordhb/9780190212926.013.14

Nass, C., & Moon, Y. (2000). Machines and mindlessness: Social responses to computers. Journal of

Social Issues, 56, 81–103. https://doi.org/10.1111/0022-4537.00153

Neumann, David. L., Chan, R. C. K., Boyle, Gregory. J., Wang, Y., & Rae Westbury, H. (2015).

Measures of empathy: Self-report, behavioral, and neuroscientific approaches. In Measures of

personality and social psychological constructs (pp. 257–289). Academic Press.

https://doi.org/10.1016/B978-0-12-386915-9.00010-3

Nowak, K. L., Fox, J., & Ranjit, Y. S. (2015). Inferences About Avatars: Sexism, Appropriateness,

Anthropomorphism, and the Objectification of Female Virtual Representations. Journal of

Computer-Mediated Communication, 20(5), 554–569. https://doi.org/10.1111/jcc4.12130

 35

Park, G., Nan, D., Park, E., Kim, K. J., Han, J., & del Pobil, A. P. (2021). Computers as Social

Actors? Examining How Users Perceive and Interact with Virtual Influencers on Social

Media. 2021 15th International Conference on Ubiquitous Information Management and

Communication (IMCOM), 1–6. https://doi.org/10.1109/IMCOM51814.2021.9377397

Philipp-Muller, A., Wallace, L. E., Sawicki, V., Patton, K. M., & Wegener, D. T. (2020).

Understanding When Similarity-Induced Affective Attraction Predicts Willingness to

Affiliate: An Attitude Strength Perspective. Frontiers in Psychology, 11, 1919.

https://doi.org/10.3389/fpsyg.2020.01919

Riedl, M. O. (2010). Story Planning: Creativity Through Exploration, Retrieval, and Analogical

Transformation. Minds and Machines, 20(4), 589–614. https://doi.org/10.1007/s11023-010-

9210-2

Rimé, B. (2009). Emotion Elicits the Social Sharing of Emotion: Theory and Empirical Review.

Emotion Review, 1(1), Article 1. https://doi.org/10.1177/1754073908097189

Roberti, G. (2022). Female influencers: Analyzing the social media representation of female

subjectivity in Italy. Frontiers in Sociology, 7.

https://www.frontiersin.org/articles/10.3389/fsoc.2022.1024043

Sands, S., Ferraro, C., Demsar, V., & Chandler, G. (2022). False idols: Unpacking the opportunities

and challenges of falsity in the context of virtual influencers. Business Horizons, 65(6), 777–

788. https://doi.org/10.1016/j.bushor.2022.08.002

Santiago, J. K., & Serralha, T. (2022). What more influences the followers? The effect of digital

influencer attractiveness, homophily and credibility on followers’ purchase intention. Issues

in Information Systems, 23(1).

Schemer, C., Masur, P. K., Geiß, S., Müller, P., & Schäfer, S. (2021). The Impact of Internet and

Social Media Use on Well-Being: A Longitudinal Analysis of Adolescents Across Nine

Years. Journal of Computer-Mediated Communication, 26(1), 1–21.

https://doi.org/10.1093/jcmc/zmaa014
 36

Schmuck, D. (2021). Following social media influencers in early adolescence: Fear of missing out,

social well-being and supportive communication with parents. Journal of Computer-Mediated

Communication, 26(5), 245–264. https://doi.org/10.1093/jcmc/zmab008

Senft, T. M. (2013). Microcelebrity and the Branded Self. In J. Hartley, J. Burgess, & A. Bruns (Eds.),

A Companion to New Media Dynamics (pp. 346–354). Wiley-Blackwell.

https://doi.org/10.1002/9781118321607.ch22

Seymour, M., Riemer, K., & Kay, J. (2017). Interactive Realistic Digital Avatars—Revisiting the

Uncanny Valley. Hawaii International Conference on System Sciences 2017 (HICSS-50).

https://aisel.aisnet.org/hicss-50/cl/hci/4

Sharabiani, A. (2021). Genuine empathy with inanimate objects. Phenomenology and the Cognitive

Sciences, 20. https://doi.org/10.1007/s11097-020-09715-w

Sheldon, P., & Bryant, K. (2016). Instagram: Motives for its use and relationship to narcissism and

contextual age. Computers in Human Behavior, 58, 89–97.

https://doi.org/10.1016/j.chb.2015.12.059

Shieber, J. (2018, April 24). The makers of the virtual influencer, Lil Miquela, snag real money from

Silicon Valley | TechCrunch. https://techcrunch.com/2018/04/23/the-makers-of-the-virtual-

influencer-lil-miquela-snag-real-money-from-silicon-valley/

Siemsen, E., Roth, A., & Oliveira, P. (2010). Common Method Bias in Regression Models With

Linear, Quadratic, and Interaction Effects. Organizational Research Methods, 13(3), 456–

476. https://doi.org/10.1177/1094428109351241

Sokolova, K., & Kefi, H. (2020). Instagram and YouTube bloggers promote it, why should I buy?

How credibility and parasocial interaction influence purchase intentions. Journal of Retailing

and Consumer Services, 53. https://doi.org/10.1016/j.jretconser.2019.01.011

Spence, P. R., Edwards, A., Edwards, C., & Jin, X. (2019). “The bot predicted rain, grab an

umbrella”: Few perceived differences in communication quality of a weather Twitterbot

versus professional and amateur meteorologists. Behaviour & Information Technology, 38(1),

101–109. https://doi.org/10.1080/0144929X.2018.1514425
 37

Statista. (2019). Instagram influencers by age 2019 l Statistic. Statista.

https://www.statista.com/statistics/893733/share-influencers-creating-sponsored-posts-by-

age/

Su, N. M., Lazar, A., Bardzell, J., & Bardzell, S. (2019). Of Dolls and Men: Anticipating Sexual

Intimacy with Robots. ACM Transactions on Computer-Human Interaction, 26(3), 1–35.

https://doi.org/10.1145/3301422

Taillon, B. J., Mueller, S. M., Kowalczyk, C. M., & Jones, D. N. (2020). Understanding the

relationships between social media influencers and their followers: The moderating role of

closeness. Journal of Product & Brand Management, 29(6), 767–782.

https://doi.org/10.1108/JPBM-03-2019-2292

Tajfel, H., & Turner, J. (1986). C.(1986). The social identity theory of intergroup behavior.

Psychology of Intergroup Relations, Edited by Worchel, S. and Austin, WG Nelson-Hall

Publishers: Chicago.

Tan, S.-M., & Liew, T. W. (2020). Designing Embodied Virtual Agents as Product Specialists in a

Multi-Product Category E-Commerce: The Roles of Source Credibility and Social Presence.

International Journal of Human–Computer Interaction, 36(12), 1136–1149.

https://doi.org/10.1080/10447318.2020.1722399

Tett, R. P., & Guterman, H. A. (2000). Situation Trait Relevance, Trait Expression and Cross-

Situational Consistency: Testing a Principle of Trait Activation. Journal of Research, 34(4),

Article 4.

Thomas, V. L., & Fowler, K. (2021). Close Encounters of the AI Kind: Use of AI Influencers As

Brand Endorsers. Journal of Advertising, 50(1), 11–25.

https://doi.org/10.1080/00913367.2020.1810595

Tietjen, A. (2018). Shudu: Fashion’s First Avatar Supermodel? She May Not Be Able to Talk or

Predict Your Favorite New Product, but Don’t Underestimate Her Power. Here, She Poses for

Wwd in Her Debut Fashion Editorial. WWD: Women’s Wear Daily, 10–11.

Valkenburg, P. M., Beyens, I., Pouwels, J. L., van Driel, I. I., & Keijsers, L. (2022). Social Media

Browsing and Adolescent Well-Being: Challenging the “Passive Social Media Use
 38

Hypothesis.” Journal of Computer-Mediated Communication, 27(1), zmab015.

https://doi.org/10.1093/jcmc/zmab015

Verhagen, T., van Nes, J., Feldberg, F., & van Dolen, W. (2014). Virtual Customer Service Agents:

Using Social Presence and Personalization to Shape Online Service Encounters. Journal of

Computer-Mediated Communication, 19(3), 529–545. https://doi.org/10.1111/jcc4.12066

von der Pütten, A. M., Krämer, N. C., Gratch, J., & Kang, S.-H. (2010). “It doesn’t matter what you

are!” Explaining social effects of agents and avatars. Computers in Human Behavior, 26(6),

1641–1650. https://doi.org/10.1016/j.chb.2010.06.012

Vossen, H. G. M., & Valkenburg, P. M. (2016). Do social media foster or curtail adolescents’

empathy? A longitudinal study. Computers in Human Behavior, 63, 118–124.

https://doi.org/10.1016/j.chb.2016.05.040

Waterloo, S. F., Baumgartner, S. E., Peter, J., & Valkenburg, P. M. (2018). Norms of online

expressions of emotion: Comparing Facebook, Twitter, Instagram, and WhatsApp. New

Media & Society, 20, 1813–1831. https://doi.org/10.1177/1461444817707349

Weiss, J. K., & Cohen, E. L. (2019). Clicking for change: The role of empathy and negative affect on

engagement with a charitable social media campaign. Behaviour & Information Technology,

38(12), 1185–1193. https://doi.org/10.1080/0144929X.2019.1578827

Wieseke, J., Geigenmüller, A., & Kraus, F. (2012). On the role of empathy in customer-employee

interactions. Journal of Service Research, 15, 316–331.

https://doi.org/10.1177/1094670512439743

Wills, K. (2019, March 6). Meet Lil Miquela: The AI influencer taking over Instagram. Evening

Standard. https://www.standard.co.uk/lifestyle/esmagazine/lil-miquela-ai-influencer-

instagram-a4084566.html

Wilt, J., & Revelle, W. (2019). The Big Five, everyday contexts and activities, and affective

experience. Personality and Individual Differences, 136, 140–147.

https://doi.org/10.1016/j.paid.2017.12.032
 39

Worrall, A., Cappello, A., & Osolen, R. (2021). The importance of socio‐emotional considerations in

online communities, social informatics, and information science. Journal of the Association

for Information Science and Technology, 72, 1247–1260. https://doi.org/10.1002/asi.24489

Xu, A., Liu, Z., Guo, Y., Sinha, V., & Akkiraju, R. (2017). A New Chatbot for Customer Service on

Social Media. Proceedings of the 2017 CHI Conference on Human Factors in Computing

Systems, 3506–3510. https://doi.org/10.1145/3025453.3025496

Yen, C., & Chiang, M.-C. (2021). Trust me, if you can: A study on the factors that influence

consumers’ purchase intention triggered by chatbots based on brain image evidence and self-

reported assessments. Behaviour & Information Technology, 40(11), 1177–1194.

https://doi.org/10.1080/0144929X.2020.1743362

Ytre-Arne, B., & Moe, H. (2021). Doomscrolling, Monitoring and Avoiding: News Use in COVID-19

Pandemic Lockdown. Journalism Studies, 22(13), 1739–1755.

https://doi.org/10.1080/1461670X.2021.1952475

Yuan, P., & Raz, N. (2014). Prefrontal cortex and executive functions in healthy adults: A meta-

analysis of structural neuroimaging studies. Neuroscience and Biobehavioral Reviews, 0,

180–192. https://doi.org/10.1016/j.neubiorev.2014.02.005
 40

9 Tables & Figures

Figure 1: Virtual Influencer, Lil' Miquela (LM)

Figure 2: Human Influencer, Meghan DeAnglis (MD)

 41

Table 1: Demographic characteristics of final sample

Characteristics Number Percentage of Sample3

Gender
Male 130 61%
Female 80 37%
Don’t identify as either 4 2%
male or female
Age
20-23 193 90%
24-26 17 8%
28 and over 4 2%
Ethnicities
African 3 1%
Black 1 1%
Caribbean 5 2%
Western European 103 48%
East Asian 22 10%
Latino/Hispanic 6 3%
Mixed 6 3%
Native Indian 1 1%
South Asian 2 1%
North African 3 1%
White North American 2 1%
Mediterranean 22 10%
Eastern European 22 10%
Other 16 8%

3
Some totals may not equal exactly 100% due to rounding effects
 42

Table 2: Social media usage characteristics of final sample

Characteristics Number Percentage of Sample4

Frequency of Social Media

Usage
Never 1 1%
Monthly, but less than once 2 1%
per week
Weekly, but less than daily 3 1%
Daily 45 21%
Several times a day 163 76%
Daily social media usage
Less than 30 minutes 6 3%
30 minutes to 1 hour 37 17%
1-3 hours 114 53%
Over 3 hours 57 27%
Social Media Platforms used5
Facebook 170 79%
Instagram 185 86%
LinkedIN 127 59%
Snapchat 115 54%
Twitter 77 36%
TikTok 57 27%
Whats App 114 53%
Other 32 15%

4
Some totals may not equal exactly 100% due to rounding effects
5
Participants were able to choose multiple responses
 43

Table 3: Regression results of effect of influencer type when participants know influencer is
human or virtual (“know” condition)
DV: Willingness to DV: Social DV: Physical
Follow Attractiveness Attractiveness
Variable B SE B β B SE B β B SE B β

Age -.09 .13 -.07 -.01 .07 -.02 .02 .07 .02
Gender -.52 .38 -.14 -.33 .22 -.14 -.58 .21 -.27**
IG Use .01 .52 .00 -.12 .30 -.04 .21 .28 .07
Influencer .56 .36 .15 .51 .21 .24* .36 .20 .18
n = 105
Gender: 0 = Female, 1 = Male
IG Use: 0 = Do not use Instagram, 1 = Use Instagram regularly
Influencer: 0 = Human Influencer, 1 = Virtual Influencer
* p < .05, ** p < .01
 44

Table 4: Regression results showing effect of interaction of influencer type and emotional
contagion when participants know influencer is human or virtual (“know” condition)

DV: Willingness to DV: Social DV: Physical

Follow Attractiveness Attractiveness

Variable B SE B β B SE B β B SE B β

Fixed Effects
Age -.08 .13 -.07 -.01 .07 -.01 .02 .07 .03
Gender -.51 .45 -.13 -.42 .27 -.18 -.61 .25 -.28*
IG Use .06 .51 .01 -.09 .31 -.03 .21 .29 .07
Influencer .54 .36 .15 .49 .21 .23* .35 .20 .17+
Emotional
-.43 .26 -.23 -.18 .16 -.16 -.04 .15 -.04
Contagion

Interaction
Effects
Influencer X
Emotional .66 .37 .25+ .13 0.22 0.08 .02 .21 .01
Contagion

Overall F 1.51 1.61 2.14

2
Adjusted R .03 .03 .12
∆F 3.22+ .37 .01
∆ R2 .03 .00 .00
n = 105
Gender: 0 = Female, 1 = Male
IG Use: 0 = Do not use Instagram, 1 = Use Instagram regularly
Influencer: 0 = Human Influencer, 1 = Virtual Influencer
+
p < .10, * p < .05, ** p < .01
 45

6
Willingness to Follow

Low
4 Contagion

3 High
Contagion

1
Human Virtual

Figure 3: Interaction of influencer type with emotional contagion on willingness to follow

influencer when know influencer is human or virtual
 46

Table 5: Regression results showing effect of interaction of influencer type and cognitive
empathy when participants know influencer is human or virtual (“know” condition)

DV: Willingness to DV: Social DV: Physical

Follow Attractiveness Attractiveness

Variable B SE B β B SE B β B SE B β

Fixed Effects
Age -.03 .12 -.02 -.02 .08 -.03 -.01 .07 -.01
+
Gender -.65 .36 -.17 -.30 .23 -.13 -.51 .21 -.24
IG Use .22 .49 .04 -.13 .31 -.04 .14 .29 .05
Influencer .44 .33 .12 .51 .21 .24* .39 .20 .19
Cognitive
-1.38 .33 -.58** -.39 .20 -.27+ -.09 .19 -.07
Empathy

Interaction
Effects
Influencer X
Cognitive 1.20 0.44 .37** .59 .28 .31* .36 .26 .20
Empathy

Overall F 3.97** 2.25* 2.58*

2
Adjusted R .15 .07 .08
∆F 7.39** 4.52* 1.90
∆ R2 .06 .04 .02
n = 105
Gender: 0 = Female, 1 = Male
IG Use: 0 = Do not use Instagram, 1 = Use Instagram regularly
Influencer: 0 = Human Influencer, 1 = Virtual Influencer
+
p < .10, * p < .05, ** p < .01
 47

6
Willingness to Follow

5
Low Cog
Empathy (full
4
scale)

3 High Cog
Empathy (full
scale)
2

1
Human Virtual

Figure 4: Interaction of influencer type with cognitive empathy on willingness to follow

influencer when know influencer is human or virtual
 48

5
Social Attractiveness

Low Cog
Empathy (full
4
scale)

3 High Cog
Empathy (full
scale)
2

1
Human Virtual

Figure 5: Interaction of influencer type with cognitive empathy on social attractiveness to

follow influencer when know influencer is human or virtual
 49

Table 6: Regression results showing effect of interaction of influencer type and emotional
disassociation when participants know influencer is human or virtual (“know” condition)

DV: Willingness to DV: Social DV: Physical

Follow Attractiveness Attractiveness

Variable B SE B β B SE B β B SE B β

Fixed Effects
Age -.10 .12 -.08 -.03 .07 -.04 -.01 .07 -.02
Gender -.53 .37 -.14 -.27 .23 -.12 -.46 .21 -.21
IG Use .09 .50 .02 -.09 .30 -.03 .23 .28 .08
Influencer .57 .34 .16 .51 .21 .24* .34 .19 .16
Emotional
.62 .21 .36** .12 .13 .12 -.09 .12 -.10
Disassociation

Interaction
Effects
Influencer X
Emotional -.94 .33 -.34** -.39 .20 -.24* -.28 .19 -.18
Disassociation

Overall F 2.74* 2.09+ 3.41**

2
Adjusted R .09 .06 .12
∆F 8.10** 3.86* 2.24
∆ R2 .07 .04 .02
n = 105
Gender: 0 = Female, 1 = Male
IG Use: 0 = Do not use Instagram, 1 = Use Instagram regularly
Influencer: 0 = Human Influencer, 1 = Virtual Influencer
+
p < .10, * p < .05, ** p < .01
 50

6
Willingness to Follow

5
Low Emo
4 Disassociation

3 High Emo
Disassociation

1
Human Virtual

Figure 6: Interaction of influencer type with emotional disassociation on willingness to

follow influencer when know influencer is human or virtual
 51

5
Social Attractiveness

Low Emo
4 Disassociation

3 High Emo
Disassociation

1
Human Virtual

Figure 7: Interaction of influencer type with emotional disassociation on social attractiveness

to follow influencer when know influencer is human or virtual
 52

Table 7: Summary of Hypotheses

No. Relationship Tested Result Summary

H1a Emotional Contagion moderates Supported Individuals high on emotional contagion report a higher willingness to follow the virtual,
Influencer Type Willingness to rather than human influencer; individuals low on emotional contagion report similar
Follow willingness to follow the human versus virtual influencer.
H1b Emotional Contagion moderates Unsupported No difference in social attractiveness ratings for human versus virtual influencer based on
Influencer Type Social participant emotional contagion scores
Attractiveness
H1b Emotional Contagion moderates Unsupported No difference in physical attractiveness ratings for human versus virtual influencer based on
Influencer Type Physical participant emotional contagion scores
Attractiveness
H2a Cognitive Empathy moderates Supported Individuals high on cognitive empathy report a higher report a higher willingness to follow
Influencer Type Willingness to the virtual, rather than human influencer; individuals low on cognitive empathy report
Follow similar willingness to follow the human versus virtual influencer.
H2b Cognitive Empathy moderates Supported Individuals high on cognitive empathy rate the virtual influencer as being more socially
Influencer Type Social attractive, as compared to the human influencer; individuals low on cognitive empathy rate
Attractiveness the human and virtual influencers similarly in terms of social attractiveness
H2b Cognitive Empathy moderates Unsupported No difference in physical attractiveness ratings for human versus virtual influencer based on
Influencer Type Physical participant cognitive empathy scores
Attractiveness
H3a Emotional Disassociation moderates Supported Individuals low on emotional disassociation report a higher willingness to follow the virtual,
Influencer Type Willingness to rather than human influencer; individuals high on emotional disassociation report similar
Follow willingness to follow the human versus virtual influencer.
H3b Emotional Disassociation moderates Supported Individuals low on emotional disassociation rate the virtual influencer as being more socially
Influencer Type Social attractive, as compared to the human influencer; individuals high on emotional
Attractiveness disassociation rate the human and virtual influencers similarly in terms of social
attractiveness
H3b Emotional Disassociation moderates Unsupported No difference in physical attractiveness ratings for human versus virtual influencer based on
Influencer Type Physical participant emotional disassociation scores
Attractiveness
H4 H1, H2, H3 only supported in Supported The moderating effects of emotional contagion, cognitive empathy, and emotional
“know” condition disassociation were only supported when participants were expressly told that the influencer
was human or virtual
53

Appendix A: Visual Stimuli Presented to Participants

Table: Picture Descriptions

1 Influencer’s birthday

2 Influencer with a dog

3 Influencer with a male friend

4 Influencer at an open-air festival

5 Influencer at the beach with a female

friend

6 Influencer’s portrait
54

Virtual Influencer:
55

Human Influencer:
56

Data Availability Statement

The data collected and used in this study will be shared on reasonable request to the

corresponding author.