Chatbots in frontline services and customer experience: An anthropomorphism perspective

This study measures the effects of chatbot anthropomorphic language on customers' perception of chatbot competence and authenticity on customer engagement while taking into consideration the moderating roles of humanlike appearance and brand credibility. We conducted two experimental studies to examine the conceptual framework. Study 1 tests the moderating effect of a chatbot's anthropomorphic appearance on the relationship between chatbots' language and customer engagement. Study 2 tests the moderating effect of brand credibility on the relationship between a chatbot's anthropomorphic language and customer engagement. The findings confirm that the interaction between humanlike appearance via the use of avatars and anthropomorphic language, such as using emojis, in conversations with customers influences customer engagement, and that this effect is mediated by perceived chatbot competence and authenticity. Further, the positive effect of anthropomorphic language on perceived competence, and subsequently on authenticity and engagement, is only significant when the brand credibility was low (vs. high). This study offers insights into the effect of chatbots' anthropomorphic language and provides suggestions on how to devise efficient strategies for engaging customers using chatbots.

. Hence, it is important to further explore new horizons in customer experience.
Chatbots are an important strategic brand asset as they provide the chance to provide efficient customer service around-the-clock (Rajaobelina et al., 2021;Thomaz et al., 2020).For instance, over the course of a year, more than 100,000 chatbots were created to interact with customers on the Facebook Messenger platform (Johnson, 2017).Chatbots are capable of simple jobs like sending customers airline tickets as well as more difficult ones like giving them shopping, health, or financial advice (Jiménez-Barreto et al., 2021).By 2025, 95% of all consumer-brand interactions will either be improved upon or replaced by chatbots, according to market analysis (Servion, 2020).Thus, it is critical for customer experience researchers to investigate the insights of chatbots into customer experience and its implications (Mariani et al., 2022;Mehta et al., 2022).
Despite the fact that chatbots are frequently used to communicate with customers, recent reports show that their interactions with them frequently go wrong (e.g., Orlowski, 2017).Consumer scepticism toward user-chatbot interactions and a preference for interacting with human agents (as opposed to chatbots) are frequent outcomes of the accumulation of these negative experiences and a general customer mistrust of technology (Elsner, 2017).The lack of perceived competence and authenticity in the customer-chatbot interaction may be the cause of consumers' scepticism and opposition to chatbots, but they have not received sufficient attention from previous studies.It is crucial that chatbot designers and programmers comprehend how to give chatbots vital social-emotional and relational aspects to increase the humanlikeness of chatbot interactions, which raises the perceived competence and authenticity of the conversation.This is because advances in AI technology have made it possible for businesses to program chatbots to provide personalized responses based on cuttingedge speech recognition technologies, enabling a more anthropomorphic interaction (Mozafari et al., 2021;Wilson et al., 2017).
While some research has looked at how chatbots resemble human language, more research is needed to determine how different language styles affect customer perception of chatbot's humanlikeness which may lead to different perceived competence and authenticity (S.Y. Kim et al., 2019;Pizzi et al., 2023).Additionally, the effort to anthropomorphize chatbots faces the difficulty that consumers find it increasingly challenging to correctly distinguish humans from chatbots (Pizzi et al., 2023).A chatbot is the representative of a brand when interacting with customers.Thus, brand credibility as the information received from a brand also may form customer expectations which may influence the interaction with chatbots (Hamilton et al., 2021).However, brand credibility has been overlooked in the interaction between chatbots and online customers.There have not been any studies as far as the authors' literature review that examines the role of brand credibility in the interaction between chatbots and online customers (Appendix A).
The research question of this paper is: "How does chatbot anthropomorphic language affects customers' perception of chatbot competence and authenticity and customer engagement?".We draw on social response theory (e.g., Moon, 2000) to capture the set of drivers behind customer engagement resulting from chatbots and their anthropomorphism.Social response theory is an appropriate theoretical foundation because of its role in establishing how individuals treat technology (or for our study, chatbots) in their interactions and helps explain their behavioral outcomes (Miao et al., 2022).Discerning how anthropomorphic elements enhance individual chatbots' social responses will prove meaningful in understanding how anthropomorphism impacts the chatbots perceived competence and authenticity (Grazzini et al., 2023;Huang & Lee, 2022).Thus, social response theory in our paper to explicate and join this important and emerging arena of chatbot research.The current research has two main objectives: first, to examine how the anthropomorphic language and appearance of chatbots influence customers' perception of competence, authenticity, and customer engagement; and second, to explore the role of brand credibility in shaping consumers' interaction with chatbots.We conducted two experiments to test our conjectures.Thus, we are among the first to examine the effect of emojis as a form of anthropomorphic language.Using social response theory, we extend prior research and include emojis as a vital form of anthropomorphic language.In addition, our results show that brands with low credibility should avoid the use of emojis in their chatbots-but this effect is negated when brand credibility is high.
Our research contributes to the existing body of knowledge in the field of digital services by providing new insights into the role of chatbots' anthropomorphic language in predicting customer perception and engagement.Our study makes three main contributions to the literature.First, we propose that chatbots' use of emojis can positively influence customers' social responses, as they begin to perceive the chatbot as a social actor (Miao et al., 2022).Our findings also suggest that emojis can be used to further reduce feelings of anger when interacting with a chatbot.Second, our study provides a more nuanced understanding of how anthropomorphizing a chatbot can lead to customer engagement, mediated by authenticity and perceived competence.Finally, our research may pave the way for future studies to incorporate emojis in chatbots as a means of expressing emotions and reducing feelings of uncanniness in a nonintrusive manner.The current research provides some suggestions for marketers on how to create efficient plans for interacting with clients through chatbots.
The paper is structured as follows.First, the literature review and hypotheses development are presented.Next, we go over the two experiments to test the hypotheses.Finally, we address the general discussion, implications, limitations, and directions for future research to form the paper's conclusion.

| Anthropomorphism
Anthropomorphism is "the attribution of human characteristics to nonhuman entities" (Zhou et al., 2019, p. 954).Anthropomorphism is proposedly drawn from a theory of mind where humans attribute intentions, attitudes, and belief schemas to explain the actions of a human or nonhuman entity (Epley, 2018).Such attribution may include emotional traits (e.g., desire and openness; Fan et al., 2016), physical traits (e.g., a humanoid face or body), or cognitive traits (e.g., knowledge; Nguyen et al., 2022).Humans often anthropomorphize objects as a way of helping them understand and relate to the world around them.For example, humans often attribute agency to animals, natural forces, or deities to explain their behaviors, even if there is no intentionality behind these entities' actions (e.g., "The dog is affectionate" vs. "The dog loves me"; Epley et al., 2007).Therefore, anthropomorphism can be specifically thought of as a process that imbues nonhuman entities with a sense of agency as a means of forming a relationship with the focal object (Newman, 2018;Tam et al., 2013).
Anthropomorphism has been used in a marketing space to generate relationships with customers.For instance, brand logos and brand personas are attempts by a brand to be perceived as a living entity to encourage relationships with customers (Aggarwal & McGill, 2007).Indeed, the Michelin Man logo, invented in 1889, was amongst the first anthropomorphic brand logos created that (1) humanized a brand, and (2) provided an engaging brand experience that resonated with customers (Jurberg, 2020;Newman, 2018).
Anthropomorphism has also been used in autonomous vehicles to increase trust (Waytz et al., 2014), implemented as product features to increase product preference (MacInnis & Folkes, 2017), or in social cause campaigns to increase message compliance (Ahn et al., 2014).
Overall, brands strive for customers to develop a relationship with them and use anthropomorphic strategies to do so (Steinhoff & Palmatier, 2021).
In the digital age, technology has removed most semblance of direct human presence in social interactions (Steinhoff & Palmatier, 2021).To instil a sense of social presence, firms have begun anthropomorphizing customer's experiences across a wide variety of platforms (Brandão & Popoli, 2022) or devices (e.g., voice assistants such as Alexa; Mende et al., 2019).With the growth of online shopping, implementing anthropomorphic elements within a customer's journey has enabled the presence of a perceived partner (Hamilton et al., 2021).This generally leads to increased positive feelings and experiences on the part of the customer.However, anthropomorphism can be a double-edged sword, leading to feelings of reactance and anger (Crolic et al., 2022).For example, if an algorithm fails (Srinivasan & Sarial-Abi, 2021) or if the agency is attributed to a specific anthropomorphic actor (Waytz et al., 2014), then negative feelings may be greater than if those actors were non-anthropomorphic (Garvey et al., 2023).This effect is stronger when there is an interaction with an anthropomorphic actor, as negative attributions are more likely to occur if they are highly anthropomorphic (T.W. Kim et al., 2023).As a result, anthropomorphism can reap great benefits but can also be quite risky, as is most clear in the case of chatbots discussed next.

| Chatbots and current research
Chatbots can be broadly thought of as "any software application that engages in a dialog with a human using natural language" (Rese et al., 2020, p. 2), are often adopted as customer-facing agents in contexts ranging from retailing and travel to legal and medical services (Garvey et al., 2023).Chatbots can act as a 24/7 touchpoint answering customer queries and provide a 15%-90% cost reduction opportunity as it replaces the need of a human agent (Bakhshi et al., 2018).Indeed, 80% of firms have incorporated, or plan to use, chatbots in their service provision (W.Kim et al., 2022).
As a result, a flurry of research has emerged to determine what factors contribute to the efficacy of chatbots (Appendix A).
Most of this research has investigated the effectiveness of chatbots and their specific elements, namely (1) the effect of chatbot anthropomorphism, and (2) the effectiveness of the communication elements the chatbot uses.
First, anthropomorphism has been widely studied to determine what aspects of chatbots are most effective in enabling adoption intent (Sheehan et al., 2020), purchase intentions (Crolic et al., 2022), engagement (Kull et al., 2021) or satisfaction and loyalty (Esmark Jones et al., 2022).Studies have often compared the differences between the presence of humanlike (vs.non-humanlike) avatars to understand if the effect of anthropomorphizing a chatbot works to increase positive outcomes (Jin & Youn, 2022).Interestingly, this has produced a variety of mixed effects in the literature.For instance, anthropomorphizing a chatbot can be detrimental in scenarios where a customer is angry (Crolic et al., 2022).Chatbots have also been found to be better than humans at giving bad news to customers (Garvey et al., 2023).Further, Drouin et al. (2022) found that individuals who spoke to a chatbot had fewer negative emotions but reported a greater sense of homophily with a human-which contrasts with Pizzi et al. (2021), who found that lower anthropomorphizing of chatbots increased feelings of reactance.
The fragmented findings have been proposed due to situational factors (T.W. Kim et al., 2023).For example, in medical diagnosis (Longoni et al., 2019), financial contexts (Luo et al., 2019), or travel and banking (Kull et al., 2021).Yet even in the face of these Second, the ways in which a chatbot interacts with customers are a cause of great scrutiny in the literature (see Appendix A for previous studies on chatbots).The majority of papers focus on textbased communications (e.g., Rese et al., 2020), with less focusing on imagery (e.g., Roy & Naidoo, 2021) and speech (e.g., Luo et al., 2019).
Text-based elements often focus on using schemas to determine what is the most effective way to anthropomorphize chatbot-customer interactions (Pizzi et al., 2021), such as the use of humor (Shin et al., 2023) or warmth (Roy & Naidoo, 2021) to reduce customers uncertainty when interacting with a nonhuman entity (Yu et al., 2022).
These specific elements of chatbots are to simulate social presence, which leads to better customer experience (Rizomyliotis et al., 2022) and is a good predictor of chatbot usage continuance (Jin & Youn, 2022).
As chatbots become more advanced, imagery is an emerging factor in chatbot interactions.For example, chatbots can provide images of the product (W.Kim et al., 2022) or use GIFs (animated images) and memes (widespread inside jokes; Zhang et al., 2022).
Combining both text and imagery may contribute to stronger intentions and attitudes in chatbot-customer interactions.However, research has focused primarily on text and imagery but "ignored social media afforded language forms, e.g., emojis" (Ge & Gretzel, 2018, p. 1272).Scrutiny of emojis is important in chatbot interactions, especially as they can provide nonverbal cues to convey feelings-a sign of anthropomorphism (Wu et al., 2022).Emojis are proposed to generate 57% more likes on Facebook and increase click-through rates by 241% (Cyca, 2022a).Emojis are purported to convey meaning, and posts containing emojis have been found to increase positive emotions and purchase intentions over social media (Das et al., 2019).Further, the presence of emojis is proposed to prompt a greater effect on the helpfulness of online reviews (Wu et al., 2022).As emojis have become an integral part of online communications, only viewing chatbots via text and imagery elements is a major shortcoming (Ge & Gretzel, 2018).As a result, this study will build on prior research and focus on the use of emojis as a form of anthropomorphism in chatbot-customer interactions.

| Social response theory
Social response theory suggests that the more anthropomorphic a chatbot is, the greater the likelihood that people will react to it as a human actor and apply social rules toward it (Miao et al., 2022).Specifically, when any technology possesses a set of humanlike characteristics (such as language, turn-taking, politeness, and interactivity), individuals begin treating it as a social actor (Moon, 2000;Nass et al., 1996).We propose social response theory plays a central role in the reactions customers have toward chatbots where, as mentioned prior, the more anthropomorphic the chatbot, the greater human characteristics and social rules are attributed toward it (Crolic et al., 2022).It is the levels of anthropomorphism that are attributed to a chatbot that may increase individuals' tendency to perceive and behave toward it as more (vs.less) human.In essence, anthropomorphic elements of a chatbot, from its visual appearance to its language-ability will influence the level of social response humans will have toward it.
Turning toward our study's focus, these anthropomorphic cues that chatbots represent, for which we argue emoji-usage is prominent, will influence an individual's social response to a chatbot.
For example, emojis play a double-edged sword in customers' reactions due to contextual factors influencing social responses (e.g., emojis in professional vs. communal contexts; Li et al., 2019).In fact, the level of realism of ChatGPT and Bing's Chatbot using emojis has sparked ethical concerns wherein emoji usage may make a chatbot indistinguishable from humans and induce mistakes through empathetic social responses (Véliz, 2023).For instance, researchers placed a pair of eyes on an "honesty box" in a university coffee shop and found people pay up to three times as much as the control group due to feelings of observation (Bateson et al., 2006).
We extend such thinking, via social response theory, that emojis convey emotion and connection with individuals, increasing a chatbot's anthropomorphism and individuals' likelihood to treat the chatbot as a social actor (Moon, 2000).As such, we propose the use of emojis in chatbots is becoming increasingly important to understand as their use by AI agents begins to blur the distinction between humans and nonhumans leading to greater influence individuals' social responses (Véliz, 2023).Overall, we build our research on social response theory due to its explanatory power for why individuals engage with high (vs.low) anthropomorphic chatbots and their behavioral outcomes.

| Chatbot anthropomorphism and customer engagement
Customer engagement, for this study, will be defined as "a customer's behavioral manifestations that have a brand or firm focus, beyond purchase, resulting from motivational drivers" (Van Doorn et al., 2010, p. 254).We adopt this definition as we find traditional definitions of customer engagement frequently assume an affective, cognitive, and behavioral facet (e.g., Chi et al., 2022;Hollebeek et al., 2014).Such studies often view engagement through social media or online brand communities in relation to a focal actor (the brand) and attribute an emotional and cognitive component to the interactions (i.e., Khan et al., 2016).This is reasonable for brand communities due to their ongoing social components, but chatbot interactions are short, goaldriven and the interaction ceases upon goal attainment or failure (e.g., did the chatbot answer the customers' question correctly or not).Due to our studies' context, we propose customers' interactions with a chatbot influence their behaviors (i.e., word-of-mouth and review intentions) as driven by the customer's specific motivations to interact (e.g., information seeking).Besides, for customer engagement in the realm of chatbots, behavioral facets are most relevant, which coincides with prior research (see Kull et al., 2021;Mostafa & Kasamani, 2022).
Customer engagement has a wide array of outcomes ranging from word-of-mouth recommendations, knowledge sharing, and helping other customers (Mostafa & Kasamani, 2022).Through the lens of chatbot interactions, a customer's engagement may heighten based on their perceived demographic similarity to the chatbot and enable a better co-creation process (Esmark Jones et al., 2022).
Additionally, the language and tone used by a chatbot have been found to increase customer engagement and their social response (Kull et al., 2021).With regard to emojis, prior research has shown emojis increases the number of likes on social media posts by 72% and comments by 70% when compared to those that do not feature them (Ko et al., 2022).Further, customers who received a smiley face were significantly happier, while those who received negative emoji felt worse (Das et al., 2019).We thus propose that the inclusion of emojis (vs.none) in chatbot communication is a sign of anthropomorphism and will lead to a higher level of customer engagement.
The physical appearance of a chatbot has been the center of much anthropomorphism discussion in extant literature (Garvey et al., 2023;T. W. Kim et al., 2022;Sheehan et al., 2020).Research has essentially agreed that the greater the anthropomorphism of an agent, the greater customers' expectations are of that agent.For example, aligning with social response theory, Miao et al. (2022) propose a typology of avatar behavior and form realism.The authors' typology suggests that the higher (vs.lower) an avatar's form realism, the greater (vs.lesser) expectations of its behavioral realism are.To illustrate, Go and Sundar (2019) found that the anthropomorphism of chatbot agents triggered human/machine heuristics of how customers would evaluate a subsequent interaction.As discussed earlier, the effect of anthropomorphizing a nonhuman entity is to assign agency and intentionality to it (Epley et al., 2007).Therefore, as a chatbot's humanlike appearance increases, more human capabilities are attributed to the chatbot (Garvey et al., 2023).We propose that the anthropomorphic appearance of a chatbot (vs.none) will have an interaction effect with the use of emojis (i.e., humanlike communication).The interaction between these two factors will increase the overall anthropomorphism of the chatbot and impact customer's engagement.Formally, we propose: H1.The interaction of chatbot's anthropomorphic appearance and language affects customer engagement.
The presence of anthropomorphism increases perceived competence (Crolic et al., 2022).Competence reflects the ability of robots (such as chatbots) to perform a task with intelligence, skill, and efficacy (Belanche, Casaló, Schepers, et al., 2021).Despite the objective competence of chatbots, customers generally have a lesser preference to interact with them (Luo et al., 2019).This effect can be attenuated by the perceived realism of chatbot communications (Kull et al., 2021).For example, a higher (lower) level of anthropomorphism may increase (decrease) levels of perceived competence due to the attribution of human characteristics, even before any interaction has taken place (Miao et al., 2022).While performed in a human service employee setting, Li et al. (2019) found the use of emoji by human agents was perceived as less competent in most conditions.Contrastingly, the use of emotional expression, such as emojis, in chatbots may be perceived as enhancing competence (Rizomyliotis et al., 2022).Interestingly, the future (vs.present) time orientation of individuals reported higher levels of brand attitude and purchase intentions based on the perceived competence of a chatbot (Roy & Naidoo, 2021).Overall, these studies show that the level of perceived competence of a chatbot and emoji is mixed, but we agree with Rizomyliotis et al. (2022) and extend their findings to propose that the perceived competence of chatbots may be heightened in the presence of emoji in customer-chatbot interactions.
Authenticity can be thought of "as the real thing" (Rese et al., 2020).The authenticity of a chatbot can refer to a customer's ability to communicate with the chatbot in a natural way (Rese et al., 2020).Yet, algorithms are perceived as less authentic than their human counterparts (Yu et al., 2022).Anthropomorphic cues, such as language or visual elements, may provide a direct signal to a customer about the authenticity of a chatbot and heighten feelings of engagement and social response (Esmark Jones et al., 2022;Nass et al., 1996).The social presence of an anthropomorphic chatbot has been found to increase its perceived authenticity and extend behavioral intentions such as continued usage (Rese et al., 2020).
Therefore, in processing and interacting with chatbots with higher levels of anthropomorphism, we expect customers to view them as more authentic than their non-anthropomorphic counterparts.
As a result, we can expect that when customers perceive a chatbot with a higher (vs.lower) level of anthropomorphism, the authenticity and perceived competence of the chatbot will increase tandemly by shifting levels of perceived authenticity and perceived competence cues toward specific heuristics (i.e., anthropomorphic elements).The social response of individuals where, the higher (vs.lower) the anthropomorphism they perceive within a chatbot, the greater (vs.lesser) they will perceive the chatbot as a social actor.As emojis are proposed to increase the authenticity (Kull et al., 2021) and perceived competence (Rizomyliotis et al., 2022) of a chatbot, we argue that higher (vs.lower) anthropomorphic cues (such as text and visuals) will have different effects as mediated by authenticity and perceived competence.Consequently, we propose the following hypothesis.
H2.The influence of high (vs.low) anthropomorphic language on customer engagement, as moderated by anthropomorphic appearance, is serially mediated by (a) perceived competence and (b) authenticity.

| Chatbot anthropomorphic language and brand credibility
Brand credibility can be understood as the information received from a brand that is contingent on its willingness and ability to deliver what they promise (Spry et al., 2009).Generally, the more a brand invests in its marketing-mix consistency, the greater its perceived credibility (Erdem et al., 2006).A chatbot, which can be viewed as a part of a brand's marketing-mix, is a proxy frontline actor for a brand and, by default, represents them during customer interactions.As chatbots are replacing frontline employees as an integral touchpoint in a customer's experience and journey (Lemon & Verhoef, 2016), chatbots can begin to serve as a substitute for human entities and leverage brand credibility (Hamilton et al., 2021).For instance, in the context of a luxury brand (e.g., Burberry), even though consumers perceived the brand's chatbot as failing to provide a diversity of NGUYEN ET AL.
| 5 information, they perceived it as credible based on the brand operating it (Chung et al., 2020).Building on this, and turning toward our studies focus, Li and Shin (2023) found that the use of emojis was detrimental to traditional luxury brands but not masstige brands.
While further, the use of emojis in advertising (vs.general posting) on social media is found to be more effective (vs.less; Ko et al., 2022).
Building on these findings, we posit that brand credibility will display an interaction effect with the anthropomorphic language (i.e., emoji use) of a chatbot.
H3. Chatbot anthropomorphic language and brand credibility display an interaction effect on customer engagement.
Further, we argue that the effect of brand credibility, or the belief a brand will fulfill its promises (Spry et al., 2009), will play an important role in alleviating the potential detrimental effects of chatbots.As the brand's first point of contact with its customers, chatbots represent the brand.In this sense, we believe that, for high credibility brands, higher (vs.lower) anthropomorphic language (i.e., usage of emoji), will lead to higher levels of perceived competence and authenticity due to perceptions of the chatbot as more like a social actor (Moon, 2000).For instance, a highly credible brand (such as Microsoft) may have a semblance of power distance (Paharia & Swaminathan, 2019) and using emojis may help customers feel closer to the brand as they are commonly used to express emotions.We believe that the more credible a brand is, the greater the effect of a chatbot's anthropomorphic language.Thus: H4.The effect of a chatbot's high (vs.low) anthropomorphic language on customer engagement, as moderated by brand credibility, is serially mediated by (a) perceived competence and (b) authenticity.
We propose the conceptual framework as in Figure 1.

| METHODOLOGY
We conducted two experimental studies in which Studies 1 and 2 explore the moderating effect of anthropomorphic appearance and brand credibility, respectively, in the relationship among anthropomorphic language, perceived competence, perceived authenticity, and customer engagement.We used the experiment method to test our hypotheses as determining cause-effect relationships is a key motivation for experimental research (Viglia et al., 2021).Using experiments, researchers can manipulate the independent variable with a great deal of environmental control to ascertain the causal links between the independent and dependent variables (Kirk, 2012).
Experimental design can also be useful in verifying the underlying mechanisms at play (Rajavi et al., 2019).Study 1 tested the moderating effect of chatbot's anthropomorphic appearance in the relationship between chatbots' anthropomorphic language and customer engagement, and serial moderated mediation effects via competence and authenticity (Hypotheses 1 and 2).Study 2 tested the moderating effect of brand credibility in the relationship between chatbots' anthropomorphic language and customer engagement, and serial moderated mediation effects via competence and authenticity (Hypotheses 3 and 4).Across the two studies, we employed different scenarios of interactions with conversational agents from a fictional retailer (see Appendix B).We also conducted pre-tests of our manipulation of anthropomorphic language and appearance.In all studies, we recruited participants located in the United States from Prolific.
We choose US participants as the United States is a global leader in technology and innovation.Prolific has been found to provide high data quality (Peer et al., 2022).It has been confirmed by a number of studies on the data quality panels for online behavioral research studies (Chmielewski & Kucker, 2020;Palan & Schitter, 2018).North America (including the United States) is the largest player in the global chatbot market with a market share of around 30.72% in 2022 (Grand View Research, 2022).According to a recent survey by Ipsos, 68% of surveyed US consumers have used an automated customer service chatbot (Larini, 2023).To ensure data quality, we consistently used attention checks questions in which we asked the respondents to recall the name of the customer service agent, the product, and if the agent wore glasses.We then removed participants who failed all attention checks.
4.1 | Study 1: the interplay of chatbot's language and appearance on customer engagement

| Study design and procedure
For this first study, we recruited 318 participants (39% males, 59.2% females, and 1.8% others; M age = 36.86,SD = 13.823)located in the United States from the online crowdsourcing platform Prolific.Participants were randomly assigned to a 2 (anthropomorphic language: low, high) × 2 (humanlike appearance: chatbot, human) between-subjects design.Anthropomorphic language was manipulated using humanlike conversational traits such as the display of empathy and the use of emojis.To confirm that our scenario would be a suitable proxy for the manipulation of language anthropomorphism, we conducted a pre-test with a separate group of 42 participants (57% males) recruited from the same online panel.Results from this pre-test showed that the use of emojis and the display of empathy in language were a successful proxy for an anthropomorphized language.Specifically, participants from our pre-test indicated that they perceived the language used by the virtual assistant as warmer (M emj = 7.90, M noemj = 6.57, t( 40 For the humanlike appearance factor, participants were shown either the image of a human (human condition) or an avatar image (chatbot), which was created based on the human image using an online free tool.A sample of the manipulation scenarios is displayed in Appendix B. Participants were asked to consider an online shopping scenario where they decided to buy a jacket for the upcoming winter season from a fictitious online store (Maza) where a chat box popped up and a sales agent started a conversation with them.After reading the conversation script, participants were then asked to answer a series of dependent measures.Each participant only saw the scenario they were randomly assigned to and were not exposed to any other condition.

| Measures
The items we used to measure the latent constructs were taken from validated scales in prior research, which helps ensure the content validity of the measurement.To increase the internal validity of our experiment and avoid familiarity or attachment effects, we used a fictional brand in our studies (Belanche, Casaló, Schepers, et al., 2021).
Competence was measured by asking the respondents if the customer service agent has the following traits: competent and capable (S.Y. Kim et al., 2019).Authenticity was measured by asking the respondents to indicate how they feel about the customer service representative: unnatural/natural, inorganic/organic, and artificial/ real (Esmark Jones et al., 2022).To measure customer engagement, participants were asked to indicate the extent to which they agree or disagree with the following statement: I feel motivated to learn more about X, I encourage friends and relatives to do business with X that offers this online chat support; I consider X that offers this online chat support to be my first choice when buying products (Kull et al., 2021;Mostafa & Kasamani, 2022).Technical anxiety was measured by 4 items "I have difficulty understanding most technological matters relating to online chat support services," "I feel apprehensive about using online chat support services," "I hesitate to use online chat support services for fear of making mistakes I cannot correct," and "Online chat support services are somewhat intimidating to me" (Meuter et al., 2003;Thatcher & Perrewe, 2002) was used as a control variable.Finally, participants reported their demographics.To minimize common method bias, we applied several strategies, including developing a research information coversheet with a clear research purpose and set of instructions, and improve scale item clarity via pretest, and remove common scale properties such as using both Likert and unipolar scales (Jordan & Troth, 2020;Podsakoff et al., 2012).

Perceived competence
The effect of anthropomorphic language on competence would be moderated by anthropomorphic appearance.We conducted a twoway ANOVA with anthropomorphic language and appearance, and their interaction as independent variables, and perceived competence as the dependent variable.We also controlled for technical anxiety.

Authenticity
We conducted a two-way ANOVA with anthropomorphic language and appearance, and their interaction as independent variables, and authenticity as the dependent variable.We also controlled for technical anxiety.Results revealed nonsignificant main effects of anthropomorphic language (F(1, 313) = 0.183, p = 0.669), appearance F I G U R E 2 Customer engagement by anthropomorphic language and appearance.

Serial mediation analysis
We propose that the effect of anthropomorphic language on customer engagement is sequentially mediated by competence and authenticity.We conducted a serial moderated mediation analysis using Hayes' PROCESS Model 85 (Hayes, 2017).This analysis examined the indirect effects of anthropomorphic language (high = 1, low = −1), moderated by anthropomorphic appearance (human avatar = 1, chatbot avatar = −1), on customer engagement via competence and authenticity (as serial mediators).We also controlled for technical anxiety.The results are shown in Table 2.
Perceived competence has a significant impact on authenticity.
Competence and authenticity were significant predictors of customer engagement.The index of moderated mediation was significant and negative (b = −0.(1 = low, 10 = high).Results from a repeated-measures ANOVA revealed that participants perceived H&M to be more honest (M H&M = 6.68 vs. M Forever21 = 3.04, F(81) = 214.54;p < 0.001) and more credible (M H&M = 6.39 vs. M Forever21 = 3.08, F(81) = 234.48;p < 0.001) than Forever21, thus supporting the choice of H&M as the higher credibility brand and Forever21 as the lower credibility brand for our brand credibility manipulation.
Similar to Study 1, participants were asked to consider a shopping scenario where they decided to buy a jacket for the upcoming winter season and that they either decided to visit the H&M or Forever21 online store where they engaged with the online store's chatbot agent.After reading the conversation script, participants were then asked to answer the same dependent measures as Study 1. Besides, participants in this study were specifically told that the sales agent was a chatbot.Each participant only saw the scenario they were randomly assigned to and were not exposed to any other condition.A sample of the manipulation scenarios is displayed in Appendix B.

Manipulation checks
As expected, participants in the high level of anthropomorphic language condition (vs. the low level of anthropomorphic language condition) perceived the language to be warmer (M high = 7.56, .This shows that our manipulation of humanlike language was successful.
To check that the manipulation of brand credibility worked as planner, brand credibility was measured using a four items scale adapted from Morhart et al. ( 2015) "X is a brand that accomplishes its value promise," "X is an honest brand," "X is a brand they will not betray you," and "I am very familiar with X" (Morhart et al., 2015).As expected, participants in the high brand credibility condition (i.e., H&M) perceived the brand to be more credible than those in the low brand credibility condition, that is, Forever21 did (M low = 4.88, M high = 6.09, t(359) = −5.031,p < 0.001).This shows that our manipulation of brand credibility was successful.

Perceived competence
The effect of anthropomorphic language on competence would be moderated by brand credibility.We conducted a two-way ANOVA with anthropomorphic language and brand credibility, and their interaction as independent variables, and competence as the dependent variable.Figure 3).When the brand credibility was low, participants in the high level of anthropomorphic language condition (vs. the low level of anthropomorphic language condition) reported a lower level of competence (M high = 7.360, SD = 2.246, M low = 8.086, SD = 1.873, t(1, 178) = −2.357,p = 0.020).However, when the brand credibility was high, there was no difference between participants in the high and low levels of anthropomorphic language conditions (M high = 8.115, SD = 1.889,M low = 7.936, SD = 1.723,F(1, 179) = 0.666, p = 0.506).

Serial mediation analysis
We propose that the effect of anthropomorphic language on customer engagement is sequentially mediated by competence and authenticity.We conducted a serial moderated mediation analysis using Hayes' PROCESS Model 85 (Hayes, 2017).This analysis examined the indirect effects of anthropomorphic language (high = 1, low = −1), moderated by brand credibility (high = 1, low = −1), on customer engagement via competence and authenticity (as serial mediators).We also controlled technical anxiety.The results are shown in Table 4.
Competence has a significant impact on authenticity.The effects of competence and authenticity on customer engagement were significant.The index of moderated mediation was significant

| GENERAL DISCUSSION
The goal of this research was to provide empirical insights into the influence of chatbot anthropomorphic language, in the interaction with chatbot anthropomorphic language and brand credibility, on customer engagement.Prior research reflects the increasing interest in chatbots and the contrasting effects of their anthropomorphism (e.g., Garvey et al., 2023) and their communication elements (e.g., Jin & Youn, 2022).Most research has neglected social media afforded communication tools as an anthropomorphic mechanism (Ge & Gretzel, 2018)-a gap this research sought to address.Study 1 shows that the humanlikeness of chatbots, which is combined with human appearance with an avatar and anthropomorphic language such as using emojis in conversation with customers, increases customer engagement.
Humanlike languages make customers feel more intimate, personal, friendly, and less formal.Human appearance or anthropomorphic language only is not sufficient to engage customers, but the combination makes a great difference.
In Study 1, we found that chatbot humanlikeness with the combination of language and appearance affects customer perception of competence which in turn affects perceived authenticity and customer engagement.While previous literature has shown that AI chatbots with communication styles lead to positive outcomes (e.g., perceived competence and warmth; Roy & Naidoo, 2021), while also addressing the effects of anthropomorphism (T.W. Kim et al., 2022), our research is the first to show that the effect of perceived competence on consumer intentions is mediated by perceived authenticity.We not only extend the anthropomorphism literature where previous studies either focus on chatbot appearance or language (e.g., Esmark Jones et al., 2022), but our study is the first to examine emoji in chatbot language as a cue for perceptions of chatbot anthropomorphism.
F I G U R E 3 Perceived competence by anthropomorphic language and brand credibility.
In addition, we show that when individuals are aware they are engaging with an AI (i.e., avatar condition), they become more motivated to engage in central route processing and therefore are more likely to critically evaluate information and form attitudes based on the content of the information (e.g., text, emojis).That is, when consumers are exposed to a chatbot sales agent, an AI schema is triggered, and consumers judge text interactions in more detail than if they were to engage with a human sales agent.As such, our results demonstrate that higher (vs.lower) humanlike language leads to higher perceived competence for chatbot interactions but not when consumers interact with a human.Furthermore, we show that consumers attribute higher levels of authenticity when a chatbot is perceived as more (vs.less) competent.However, because consumers do not have a reason to question the authenticity of human-tohuman interactions, we did not find evidence that consumers engaged in more nuanced analysis of text-based interactions.
In Study 2, the findings indicated that brand credibility set an expectation in the head of customers which influences the impact of chatbot anthrophonic language on customer engagement.Customers often wonder whether chatbots can represent a brand well (Chung et al., 2020).Brand credibility has been overlooked in the anthropomorphism literature and the link between brand credibility and chatbot characteristics has not been yet explored.Previous research in AI and chatbots have mostly found that AI/chatbot anthropomorphism has a positive influence on perceptions of competence (e.g., Pizzi et al., 2023) and warmth (e.g., S. Y. Kim et al., 2019;Pizzi et al., 2023).However, it should be noted that the relationship between anthropomorphism and perceived competence/authenticity is not always be linear or universal.In this study, we show that this effect is moderated by brand credibility such that there is a reversal of the effect of anthropomorphism on perceived competence and perceived authenticity.In particular, our results are consistent with the notion that when consumers interact with a brand that is perceived to be more credible, they are less motivated to engage in central route processing because their previous interactions with the brand give customers assurance that the brand will perform as promised, and therefore new encounters (e.g., interaction with a chatbot sales assistant) are likely to be processed through the peripheral route.On the other hand, when consumers interact with a brand perceived to be less credible, they are more likely to process information thoroughly, via central processing, as there is higher uncertainty about the brand's ability which is demonstrated by lower consumer ratings of perceived competence.In consequence, when less credible brands try to improve perceptions of chatbot competence with anthropomorphism, consumers scepticism is triggered and instead, the attempt backfires such that interactions are perceived to more artificial, less organic, more inauthentic, a complete reversal of the chatbot anthropomorphism effect.

| IMPLICATIONS
In addressing the emerging interest in the anthropomorphism of chatbots, this study has attempted to contribute to this extant literature.The significant findings found within this study will have implications for practitioners and academics alike.We provide a summary of our findings and their implications in Table 5.

| Theoretical implications
The current study contributes to the literature in three ways.First, a growing stream of literature has begun investigating the replacement of human agents with AI agents and customer preferences between them (Yu et al., 2022).Literature in this area has shown the influence of humor (Shin et al., 2023), language schemas and (Zhang et al., 2022) -but this study is, to the authors' knowledge, the first to empirically test the effect of emojis in chatbots.As the capabilities of chatbots emerge, the use of emojis may humanize not only the chatbot, but the brand.Further, when combined with anthropomorphic visual cues, the use of emojis leads to higher levels of customer engagement.We propose that the chatbots use of emojis positively influence customers' social response as they begin to perceive it more as a social actor (Miao et al., 2022).Taken together, the use of emojis can thus be utilized in conjunction with similar research, such as Garvey et al. (2023), who found sending an anthropomorphized agent can be more effective than a human.Using our findings, emojis may be also used to further attenuate feelings of anger when using such an agent (Crolic et al., 2022).
Second, our findings provide a more nuanced piece of the puzzle on how anthropomorphizing a chatbot can lead to customer engagement.
We established the mediating role of authenticity and perceived competence.Prior research identified the importance of these factors (e.g., Kull et al., 2021;Roy & Naidoo, 2021) but rather identified their direct effects in conjunction with the perceived warmth of the chatbot.
Our results indicate a chatbot's perceived competence and authenticity serially mediate the influence of anthropomorphic chatbots under two specific conditions (1) the anthropomorphic appearance of the chatbot and (2) low brand credibility.In particular, the presence of emojis enhanced low (vs high) anthropomorphized chatbot agents, while the use of emojis had a negative effect when a chatbot with low brand credibility used them.These findings are interesting as emojis can thus act as a double-edged sword where individuals' social response and perception of a chatbot as a social actor can benefit overall engagement, but harm those smaller and less credible brands.We validated these constructs for future researchers as strong mediators of anthropomorphism on customer engagement.
Finally, we contribute to the overarching literature on anthropomorphism.Prior research has often dealt with high levels of anthropomorphism using the uncanny valley hypotheses (e.g., Mende et al., 2019).For instance, if a robot physically resembles a human and acts in an agentic manner, feelings of discomfort may emerge (Crolic et al., 2022).This was further addressed by extant literature on chatbots by predominantly using a human (vs.none) avatar (e.g., Esmark Jones et al., 2022).However, utilizing emojis, a form of cultural and emotional expression (Li et al., 2019), in a chatbot context addresses this area of research.Our contribution to this arena may allow future researchers to incorporate emojis to T A B L E 5 Summary of implications.

Findings Summary of Implication(s)
Theoretical implication

Emoji usage
• Addressing the growing stream of literature on chatbots, and in conjunction with social response theory, this paper is the first to analyze the effect of emojis.• Emoji usage is recommended for chatbots where anthropomorphism is required.

Authenticity and competence
• Perceived competence and authenticity serially mediate anthropomorphic chatbots' effect on customer engagement.• Emojis had a negative effect when low-credible brands utilized them, but this effect was not present for brands with high credibility.

Anthropomorphism
• Using emojis may overcome feelings of unease in highly anthropomorphic chatbots.

Managerial implication
Brand credibility

| Managerial implications
Our findings indicate that the use of emojis may act as a double-edged sword and backfire if used incorrectly.For instance, in brands that have low credibility, emojis may have a negative effect.Therefore, marketing managers of brands with lesser credibility, such as those with lower brand awareness (Spry et al., 2009), should use chatbots without emojis.
Further, emojis and the anthropomorphic appearance of a chatbot had a main, and mediating effect, on customer engagement.This suggests marketing managers should create a multidimensional chatbot (i.e., anthropomorphic appearance and language) are the most effective at building customer engagement overall.Taken together, we recommend marketing managers of brands with lower brand credibility, such as in small-medium enterprises (SMEs), to be cautious in their use of emojis with chatbots but, if doing so, use chatbots with lower anthropomorphic appearance (Miao et al., 2022).As such, the more anthropomorphic a chatbot is, the more a mistake can harm a less credible brand (Moon, 2000).
In addition, our results can be beneficial in understanding how competence and authenticity increase customer engagement.It is critical that the chatbot is perceived as competent and authentic, particularly if anthropomorphic, as social response theory indicates that customers will attribute human characteristics to it (Mende et al., 2019).It is important that marketing managers test their chatbots with customers-particularly chatbots (1) with more humanlike elements and (2) for lesser credible businesses such as SMEs.
Developing metrics such as churn, time spent with a chatbot, goal completion rates or feedback scores are essential to improve the chatbot, its competen,cy and authenticity as well as the customer's experience (Cyca, 2022b).For example, Telenor implemented specific metrics for their chatbot and increased customer satisfaction and revenue by 20% and 15%, respectively (Leah, 2022).These metrics will help managers understand the effectiveness of their chatbots and which characteristics of anthropomorphizing, such as via emojis or humanlikeness, is appropriate (Li et al., 2019;Pizzi et al., 2021).

| LIMITATIONS AND FUTURE RESEARCH DIRECTIONS
Despite the current research provides insights into chatbots' anthropomorphic language, it still has a number of limitations.First, we collected data from the online Prolific platform, where each participant completed the questionnaire on his or her own chosen devices.Yet providing results that are more generalizable and trustworthy than samples made up of college students, this way of data collection may bring "noise" into the data collection methods used in the study (Goodman et al., 2013).Second, the study context was the customer service environment in the apparel sector.It may be worthwhile to validate the findings for chatbots used for customer service in different industries, even though we do not expect the research setting we chose to have a substantial impact on our findings.To examine the findings' broader application, it could be worthwhile to replicate the study in a new environment, such as one that makes use of chatbots for teaching or medical support.Third, the chatbot's persona and conversational intelligence were left out of the current research and could have an effect on how the user perceives the chatbot's humanlikeness (Go & Sundar, 2019).Despite the aforementioned drawbacks, we think that our work has greatly advanced theory and practice and will serve as a source of inspiration for more study.

| CONCLUSION
fragmented findings, research has forged forward to finding what aspects of chatbots work or not.Overall, research has forged forward to finding what aspects of chatbots work and do not work.This has led to research classifying what communication elements are most effective.
Serial mediation moderated by brand credibility.
T A B L E 4 Kim et al., 2019) credibility should not use chatbots with emojis.But if emoji usage is desired, this negative effect can be offset by lower anthropomorphic appearance.13ommunicate feelings and placate feelings of uncanniness in a nonintrusive manner (S.Y.Kim et al., 2019). |