Linguistic elements of conversational human voice in online brand communication: Manipulations and perceptions

The conversational human voice (CHV) is an extensively studied and adopted communication style in online brand communication. However, in previous research the way in which CHV is operationalized differs considerably: the type and the number of linguistic elements used to establish a sense of CHV in online brand messages varies. Moreover, it is still unknown how CHV operationalizations contribute to consumers’ perceptions of CHV, which consequently could affect their evaluation regarding the message and the brand. In this paper, we addressed these issues by conducting an integrative literature review and a perception experiment, and consequently present a taxonomy of linguistic elements related to message personalization, informal speech, and invitational rhetoric that can be used to operationalize CHV systematically in future studies in online brand communication. Directions for future research and managerial implications are discussed.


Introduction
With the emergence of the internet and especially since the development of Web 2.0, both consumers and brands increasingly communicate in an online environment. In the online public atmosphere of blogs, review websites, and social media, information and opinions can be shared, questions can be asked, and complaints can be expressed easily, reaching a great number of people. In the context of marketing communications, these developments shifted the traditional top down, one-sided communication to equal, two-sided communication between consumers and brands (e.g., Kaplan & Haenlein, 2010).
The interactive online environment requests adjustments to the brand's communication style to accommodate to the register of Web 2.0 (e.g., Fournier & Avery, 2011;Kaplan & Haenlein, 2010). In The Cluetrain Manifesto, Searls and Weinberger (2000) advice brands to adopt a 'markets as conversations' approach that builds collaborative relationships through conversational communication with consumers rather than targeting them with unsolicited promotional messages like direct mail (i.e., 'markets as target' approach). Their notions were later conceptualized as conversational human voice.
The conversational human voice (CHV) is mostly attributed to Kelleher who defined the concept as "an engaging and natural style of organizational communication as perceived by an organization's publics based on interactions between individuals in the organization and individuals in publics" (Kelleher, 2009, p. 177). CHV reflects both human voice attributes (in contrast to a corporate tone of voice) such as using a sense of humor and treating others as humans, and conversational attributes such as providing prompt feedback and being open to dialogue (Kelleher & Miller, 2006;Kelleher, 2009).
van Noort, Willemsen, Kerkhof, and Verhoeven (2014) distinguish three tactics that can be used to establish a sense of CHV in online brand communication: Message Personalization, Informal Speech, and Invitational Rhetoric. Message Personalization refers to the degree to which a specific individual (brand and consumer) can be addressed in a message (cf. Walther, 2011). Informal Speech is described as casual, everyday language that differs from formal, corporate language (cf. Kelleher & Miller, 2006). Invitational Rhetoric represents the conversational aspect; i.e., the extent to which the brand's communication style stimulates consumers to engage in conversations and creates mutual understanding between them (cf. Foss & Griffin, 1995).
The past decade, the use and effects of CHV has been examined extensively in the fields of corporate communication, public relations, and marketing communications. In general, the research shows the use of CHV in online brand messages has positive effects on consumers' brand evaluations, such as reputation and trust (e.g., Kelleher, 2009;Kerkhof, Beugels, Utz, & Beukeboom, 2011;Yang, Kang

. Search and selection strategy
Databases Google Scholar and Web of Science were used to search relevant studies that have been conducted from 2006, the appearance of Kelleher's first paper about CHV (Kelleher & Miller, 2006), until July 2020. In order to find papers in which concrete linguistic operationalizations of CHV were examined, we narrowed our search scope from conversational human voice without any filters to "conversational human voice" followed by relevant search terms, such as brand communication, operationalization, language, and linguistic. The search was limited to English papers which were published in international peer-reviewed journals to ensure mutual comparability.
We identified 158 unique records through database searching and removed Kelleher's two core papers that defined the concept of CHV (Kelleher & Miller, 2006;Kelleher, 2009). In-depth reading of the remaining full papers 1 was performed to determine the relevance of the article regarding the linguistic operationalization of CHV. Papers were eligible for inclusion in the literature review if they met the following criteria: (1) CHV was the main research objective, (2) CHV was an independent variable in the study, and (3) concrete examples of CHV operationalizations were provided.
With regard to the first inclusion criterion, it appeared that 86 papers were eliminated since the concept of CHV was only described in the theoretical sections of the paper (68 papers) or CHV-related references were only present in the bibliography (18 papers). The second criterion led to the exclusion of 17 papers since researchers included the concept to measure the perception of CHV rather than manipulating it (e.g., Dijkmans, Kerkhof, Buyukcan-Tetik, & Beukeboom, 2015;van Noort & Willemsen, 2012). Lastly, 7 papers were eliminated because they did not met the third criterion: the CHV manipulations were not operationalized with concrete linguistic elements (e.g., Chen, Ji, & Men, 2017;van Wissen & Wonneberger, 2017). In total 38 papers met the eligibility criteria. Inspection of the reference lists of the included papers did not lead to the inclusion of additional articles in the literature review. An overview of the papers included in the final dataset is online available via Open Science Framework (OSF), https://osf.io/sxp8u/.

Results
To structure the findings of the literature review, we use the three main tactics of CHV that can be used to establish a sense of CHV in online brand communication as proposed by van Noort et al. (2014).
Personal Greeting. Eleven papers operationalized CHV by personally greeting consumers using greeting words (Hi, Hello, Dear), and/or addressing the consumer with his or her name (Hi John, Hello Günther). Remarkable examples were reported by Zhang, Tao, and Kim (2014) and Zhang and Vásquez (2014), who stated that consumers can also be addressed with proper names (Dear valued guest), family names (Dear Mr. Smit), or with special nicknames (such as Beliebers). These personal greetings could also differ in informality (compare Hi John with Dear Mr. Smith) although this is not explicitly addressed by scholars.
Personal Addressing Consumer. The second subcategory of message personalization is the use of personal pronouns to address the consumer. In fourteen research papers, CHV was manipulated by second-person pronouns as you, your, yours, and yourself.  also reported the personal pronoun U. This so-called textism can be described as orthographically unconventional language in which standard spelling conventions and grammar rules are disregarded (Verheijen, 2013) and thus can also be regard as informal language.
Personal Addressing Employee. In 26 papers, first-person pronouns were used to refer to the sender of the message (i.e., employee). Pronouns, such as I, my, me, myself, we, us, our and ourselves, enhance the receivers' perceptions they are communicating with an individual brand representative rather than a faceless brand. These personal pronouns are often used instead of addressing to the brand as a whole (such as we versus the colleagues or the webcare team, our IT department versus the IT department, I instead of the brand name). However, a potential difference in informality could be distinguished as well. Zhang and Vásquez (2014) stated that first-person singular pronouns (I, my, me) are more humanized expressions than first-person plural pronouns (we, our, us) which gives the impression of a brand collective. Ten papers explicitly mentioned the use of first-person singular pronouns as a tactic to address the employee personally (i.e., Hamby & Ilyuk, 2019;Hong & Kim, 2019;Jahng & Hong, 2017;Kim, Sung, & Moon, 2020;Lillqvist & Louhiala-Salminen, 2014;Merry, 2010;Oh & Ki, 2019;Park & Cameron, 2014;van Prooijen & Bartels, 2019;Zhang & Vásquez, 2014).
Personal Signature. The brand message can be signed personally which enables consumers to identify the employee behind the brand message. Sixteen research papers mentioned personal signature, but the manipulations differed substantially: using only first names or initials (Thomas, or TO, in online brand messages it is common that signatures are preceded by a ^), including last names (Morgan Smith), and/or professional titles, and brand names (Niina from customer services, Anna from the [car brand] team). These additions could be perceived as more formal personal signatures.

Informal speech
Informal speech can be defined as a casual, familiar, and common style of communication (Gretry et al., 2017) that mimics spoken language (Chua et al., 2012). An informal style is believed to reinforce relationships between consumers and brands because informality decreases the feeling of hierarchy between the interlocutors which in turn reinforces trust (Gretry et al., 2017). In 21 papers CHV was operationalized by means of informal speech, that can be classified into Nonverbal and Verbal cues (see Table 2).
Nonverbal Cues. This subcategory conveys features of nonverbal communication in computer-mediated written interactions that replace gestures, intonation, body language, and facial expressions in everyday face-to-face communication. According to Luangrath, Peck, and Barger (2017) paralinguistic features in computer mediated written interaction could convey audible (e.g., the usage of capitals to represent volume in spoken language), visual (e.g., emoticons and emoji to visualize facial expressions), or tactile elements (e.g., animated videos (GIFs) that refer to tactile experiences).
Interjections, the second subcategory, were only mentioned in the papers of Johnen and Schnittka (2019) and Luangrath et al. (2017). Interjections are spoken language-based words that often mimic sound, such as ah, wow, haha. In their study, Luangrath et al. (2017, p. 101) explicitly mark this alternant of vocalization, and state these utterances, fillers, terms, or sounds can be spoken or produced by the body and result in an audible noise that is not necessarily an English word, but conveys meaning.

Table 1
Manipulations of message personalization in previous CHV studies.

Invitational rhetoric
Invitational rhetoric refers to strategies brands use to invite consumers to conversation in order to build a relationship with consumers and in addition declare their commitment to this relationship (Avidar, 2013, Kwon & Sung, 2011Pollach, 2005;Willemsen, Neijens, & Bronner, 2013;van Noort et al., 2014), as proposed in the manifest of Searls and Weinberger (2000). Invitational rhetoric appeared in 22 papers (see Table 3).
Stimulating Dialogues. Employees can explicitly invite consumers to share their thoughts and experiences about the brand, for example by asking questions (Could you explain what is the matter?) and expressing commitment to the uniqueness and intrinsic value of each consumer's thoughts and experiences about that brand (Please share your thoughts with us!). 2 Phrases to invite consumers to engage in a conversation were mentioned in seventeen papers, for example by asking for feedback (Let us know what you think, submit your suggestions), maintaining a relationship (Follow, stay tuned, become a fan, join us at an event), or redirecting consumers to other media (Check out the links, watch new ads, learn about) (e.g., Kwon & Sung, 2011).
Acknowledging. The appreciation of consumers' (critical) feedback comes across explicitly when brands thank their consumers for their thoughts, suggestions, comments and questions. Several studies have shown social cues, such as acknowledging, enhance the perceived humanness of brands (Aggarwal & McGill, 2007;Puzakova, Kwak, & Rocereto, 2013;Reynolds & Quinn, 2008). Nine papers were found in which acknowledging was used to manipulate CHV. Examples of manipulations are: Thank you for approaching us (Avidar, 2013), and Thank you for writing your review, which we value very much (Sparks et al., 2016).
Apologizing. A natural way of responding to negative experiences or events, such as a consumer' complaint or a crisis, is to offer an apology. By showing regret, the brand shows it takes the consumers' comments seriously which may invite them to engage in a conversation with the brand to solve the problem instead of badmouthing about the brand's failure (Grégoire, Salle, & Tripp, 2015). The manipulation of CHV by means of an apology appeared only in six research papers. Commonly used words to offer an apology were sorry, apology, apologize, regret. Zhang and Vásquez (2014) state a first-person singular pronoun often co-occurs with apologies (i.e., I am sorry).
Showing Sympathy or Empathy. A brand could stimulate conversations with consumers when it shows sympathy and/or empathy to their situation. By using phrases, such as We understand that this has been an inconvenient situation for you, and we can imagine that this is not how you envisioned your experience (Javornik et al., 2020) the brand shows it feels along with consumers and understands their concerns, which will enhance the consumers' perceptions regarding the brand. This subcategory appeared in seven papers only. Some scholars used sorry to show the brand's compassion to the consumer's feelings (i.e., I am sorry for your frustration!, We feel so sorry; Barcelos et al., 2018;Liu & Ji, 2019;Song & Kim, 2016). Others used adverbs of modality (naturally, unfortunately) and emotional adjectives, such as 'happy' in We are happy that you shared your positive experience with your [car brand] in our community (Hong & Kim, 2019;Javornik et al., 2020;Schamari & Schaefers, 2015).
Using Humor. If brands want their consumers to engage with them, they 'need to give them a reason for doing so' by being enjoyable and Table 2 Manipulations of informal speech in previous CHV studies.

Category Description References
Nonverbal cues Linguistic features that represent audible, tactile or visual elements in face-toface communication

Audible elements
Capitalization ( 2 In some studies this category is labelled as 'interactivity' (Avidar, 2013); according to Kelleher's work (2009) (2010), who investigated social media messages that refer to an earthquake on Haiti: #Uknowubrokewhen Haiti sends YOU some money. Well-wishing. Brands can express positive wishes to consumers to ensure the relationship will be maintained in the future. This can be achieved by including well-wishing in the closing remarks of the brand messages, for example by phrasing Have a nice day, or by encouraging the consumer to revisit the brand like We look forward to welcoming you back in the near future. According to speech act theory (Austin, 1975;Searle, 1969), well-wishing is a common and natural way to close conversations. However, well-wishing only occurred in five papers. In response to hotel reviews, Liu and Ji (2019) for example used the sentence Thank you and wish you more wonderful and memorable journeys in your life.

Conclusion
The integrative literature review showed CHV is operationalized differently in the 38 papers that met our eligibility criteria. Based on the three main tactics of CHV (van Noort et al., 2014), we reviewed the literature on concrete linguistic elements that can be used to establish a sense of CHV in online brand communication. The first main category, Message Personalization, is most commonly used in previous CHV research. Especially, personally addressing employee (e.g., I, we) and the employee's signature (e.g., Niina, ^Thomas) are frequently mentioned and used in experimental materials. The second frequently used main category in prior research is Invitational Rhetoric. Especially, stimulating dialogues (e.g., Could you explain what is the matter?) is frequently employed. The third main category, Informal Speech, is least present in the reviewed research. If scholars used informal language in their experimental materials, visual elements (e.g., ;-), ) and informal vocabulary and phrases (e.g., That's awesome) are most frequently used.
Based on the integrative literature review, we have created a taxonomy which consists of seventeen categories related to Message Personalization, Informal Speech, and Invitational Rhetoric that can be used to establish a sense of CHV in online brand messages. The taxonomy summarizes the three tables with linguistic elements of CHV and is online available via Open Science Framework (OSF), https://osf. io/sxp8u/. Scholars can use this taxonomy to systematically operationalize CHV in experimental materials in future research. Moreover, marketing communication practitioners can use the taxonomy to systematically adopt CHV in online brand messages, or to create their own online communication style.
Although the integrative literature review provides insights in which linguistic elements are used to establish a sense of CHV in online brand messages, it is not clear how these elements are related to the perceived CHV. In the reviewed research, a reflection is missing on whether the type of linguistic elements and the number of linguistic elements differ in consumers' perception of CHV. Therefore, a perception experiment was conducted to investigate to what extent the type and the number of linguistic elements in online brand messages lead to differences in the perception of CHV.

Introduction
Nowadays, brands seem to have embraced the Searls and Weinberger's (2001) 'markets as conversations' approach. They are predominantly present on Facebook and Twitter -94 percent uses Facebook and 68 percent uses Twitter (Stelzner, 2017) and have appointed representatives to monitor online consumer messages containing questions and complaints about brand-related topics, and, when deemed necessary, to engage in conversations. These practices have become known as webcare (van Noort & Willemsen, 2012).
In the highly conversational context of webcare, Liebrecht and van Hooijdonk (2020) already examined the contribution of linguistic elements to the perception of CHV in online brand messages, albeit amongst webcare employees. The scholars conducted an explorative perception experiment in which 47 representatives rated and ranked the perceived CHV of webcare messages. Nine linguistic elements were included in the experiment (i.e., three elements for Message Personalization, three elements for Informal Speech, and three elements for Invitational Rhetoric). Each webcare message contained only one of nine linguistic elements. Also, a basic webcare message was included in the experiment which did not contain any linguistic element of CHV.
The results of the perception experiment showed that a webcare message containing a linguistic element of CHV led to higher perceived CHV, with the exception of messages in which contractions were used. Moreover, Message Personalization contributed most to the perception Expressing positive wishes to consumers to ensure the relationship will be maintained in the future (Have a nice day!, We sincerely hope that you will return) Barcelos et al., 2018;Javornik et al., 2020;Liu & Ji, 2019;Sparks et al., 2016;Zhang & Vásquez, 2014 Note. Categories with superscripts are ( C. Liebrecht et al. of CHV, whereas Informal Speech contributed least. The current study will examine on a larger scale whether these findings hold for consumers who read webcare messages. Based on Liebrecht and van Hooijdonk's (2020) findings, the following hypotheses are formulated with regard to webcare messages containing one linguistic element of CHV.
H1: Webcare messages containing one linguistic element of CHV will lead to higher levels of perceived CHV compared to webcare messages without linguistic elements of CHV. H2: Webcare messages with Message Personalization will lead to higher levels of perceived CHV compared to webcare messages with Informal Speech or Invitational Rhetoric.
The integrative literature review showed that in previous research CHV is oftentimes operationalized using multiple linguistic of CHV, varying from elements of only one main category (e.g., Kerkhof et al., 2011;Schamari & Schaefers, 2015) to elements of multiple main categories (e.g., Gretry et al., 2017;Huibers & Verhoeven, 2014). Arguably, the latter operationalization could contribute more to the perceived CHV than messages containing elements of only one main category, which is reflected in hypothesis 3: H3: Webcare messages with multiple linguistic elements of all three main categories of CHV will lead to higher levels of perceived CHV compared to webcare messages containing elements of only one main category of CHV.

Materials and design
The experimental materials consisted of conversations between a consumer and a fictitious brand (i.e., reactive webcare, van Noort & Willemsen, 2012). Basic webcare responses were formulated containing an average amount of CHV (which was determined in a pretest 3 ). An example of such a basic webcare message is shown in Table 4 (condition  1).
Subsequently, the basic webcare responses were adjusted by adding a linguistic element to the responses. Following Liebrecht and Hooijdonk (2020), nine CHV subcategories were included in the experiment (i.e., three subcategories for Message Personalization, Informal Speech, and Invitational Rhetoric). For Message Personalization, a personal greeting, addressing consumer, and signature were selected. For Informal Speech, contractions, visual elements, and interjections were included. Finally, showing sympathy/empathy, stimulating dialogue, and well-wishing were selected from the Invitational Rhetoric category. In Table 4, conditions 2 to 10 represent the experimental webcare messages containing one linguistic element of CHV. Next to webcare responses with one CHV element, we also created messages that contained all three subcategories per main category, and one message in which all nine subcategories were present (see Table 4, condition 11 to 14).
To increase the generalizability of the perception experiment compared to Liebrecht and Hooijdonk (2020) study, we varied the type of consumer messages (electronic Word-of-Mouth, eWOM) the brand responded to. Next to neutral consumers' questions the brand could also respond to consumers' negative comments. The type of eWOM (neutral versus negative) was included as a second factor in the experiment. To generate a difference between the consumers' messages, the sentiment of the message with a question was slightly adjusted to a message with a complaint by adding negative adjectives and/or adverbs, and exclamation marks. A pretest 4 revealed this manipulation was successful: consumers' complaints were perceived as more negative than consumers' questions.
In sum, the experiment had a 14 (type and amount of linguistic elements of CHV) × 2 (type of eWOM) mixed-design where the former factor was within-subjects, and the latter factor was between-subjects. Participants were thus exposed to all 14 CHV conditions, but were only exposed to the webcare messages in response to neutral or negative eWOM. In order to prevent repetition of the eWOM topics, 14 customer service topics (such as not working websites and unwanted newsletters)

Table 4
Manipulations of CHV in webcare messages in Study 2.

CHV subcategory Example Condition
Basic response Thanks for the notification. This part of the website is under construction until tonight, after which the Projects page is completely up to date. Sufficient information will be available soon! 3 Participants rated the perceived CHV of twenty webcare responses on three items on 7-point Likert scales (e.g., 'The webcare response is informal'). For the main experiment, we selected ten webcare responses with an average CHV score between 3.5 and 4.0. Participants' ratings indicated these webcare responses were also considered natural (three items on 7-point Likert scales (e.g., 'The webcare response is realistic'), M = 4.8-5.9). 4 Participants rated the sentiment of either the neutral or negative consumers' messages (incl. four positive messages to avoid bias) on five items on 7point Likert scales (e.g., 'I consider the expression of the consumer to be satisfied'). An independent-samples t-test revealed that, on average, consumers' complaints were perceived as more negative (M = 1.88, SD = 0.34) than consumers' questions (M = 3.38, SD = 0.80, Mdif = 1.50, t(18), p<.001).
and accompanying webcare messages were created for the two types of eWOM. Each participant assessed one experimental condition per topic for one type of eWOM, which generated 14 responses per participant in total.

Instrumentation and procedure
Participants assessed the perceived CHV of each webcare message on a 7-point Likert scale. The perceived CHV was operationalized with three items that covered the perception of the three main categories of CHV: 'The webcare message is personal/informal/detached (reversed item)'. An explanatory factor analysis extracted one component that explained 51.31% of the variance. However, our reliability analysis of the scales showed a moderate internal consistency of the items (α = 0.51, M = 4.52, p = .182). Since deleting one item (i.e., the item regarding the perception of informality of the webcare response) would only slightly enhance Cronbach's α to 0.60, we decided to analyze the three items separately.
Participants were approached by convenience sampling by sharing the link to the questionnaire in Qualtrics. After clicking the link and participant's consent to participate, they were randomly assigned to one of the questionnaires in which the 14 webcare messages to either neutral or negative eWOM were shown one by one. Subsequently, participants answered questions concerning their demographics. The experimental questionnaire took approximately 20 min and participants did not receive compensation. The entire study was approved by the Research Ethics and Data Management Committee of our university (REDC #2019/130). Research data an experimental materials are online available via Open Science Framework (OSF), https://osf.io/sxp8u/.

Results
Mixed ANOVA's with planned contrasts were performed to investigate the participants' perceptions of the three separate CHV items after reading webcare messages 5 with different linguistic elements. We will first report the perceived CHV of the individual CHV manipulations, per main category and between subcategories. Findings with regard to the perception of CHV are shown in Table 6.

Contribution of individual CHV elements to perceived personalization
As shown in Table 6, the basic webcare message received an average score (M = 3.14). The first analysis, concerning the effects of CHV manipulations on perceived personalization, showed a significant main effect for the main CHV categories (F (3, 405) = 50.18, p < .001, η p 2 = 0.27). Within-subject contrasts revealed that each main category differed significantly from the basic webcare message (all p's < 0.001). Post-hoc comparisons using Bonferroni correction showed Message Personalization contributed most to the perception of personalization, followed by Invitational Rhetoric, and consequently Informal Speech (all p's ≤ 0.003).
Focusing on the different subcategories, we also found a significant main effect of the CHV elements with regard to the perceived personalization (F (8.71, 1176.31) = 32.40, p < .001, η p 2 = 0.19). 6 All subcategories of CHV elements enhanced the perception of CHV compared to the basic webcare message (all p's < 0.001), except for Contractions (p = .20). Next to differences with regard to basic webcare message, we also found differences between the CHV subcategories. Notable is that a personal greeting contributed most to the perception of personalization in comparison to all other CHV elements (all p's < 0.001). Within the main category Personalization and the main category Informal Speech, all subcategories differed significantly from each other (all p's < 0.02), but no differences were found between the subcategories of Invitational Rhetoric (all p's ≥ 0.92).

Contribution of individual CHV elements to perceived informality
Similar to the analysis with regard to perceived personalization, the first analysis concerning the effect of the main CHV categories on perceived informality showed a significant main effect for the main CHV categories (F (2.31, 311.24) = 4.09, p = .01, η p 2 = 0.03). 7 Within-subject contrasts revealed that compared to the basic webcare message (M = 4.00), Message Personalization (p = .02) and Informal Speech (p = .008) contributed significantly to the perception of informality. However, Invitational Rhetoric did not differ from the basic webcare message on perceived informality (p = .11). Furthermore, post-hoc comparisons showed the three main categories did not differ from each other (all p's ≥ 0.45).
Focusing on the different subcategories, we found a significant main effect of the CHV elements with regard to the perception of informality (F (9, 1215) = 7.10, p < .001, η p 2 = 0.05), but the pattern was less clear compared to the perception of personalization. In comparison to the basic webcare message, significant differences were found for personal greeting (p < .001), signature (p = .004), visual elements (p = .008), interjections (p < .001), and well-wishing (p = .01), whereas no differences were found for personal addressing consumer (p = .26), contractions (p = .99), stimulating dialogues (p = .32), and showing sympathy or empathy (p = .65). Thus, per main category some sub categories enhanced the perception of informality of a webcare message, while others did not.

Contribution of individual CHV elements to perceived engagement
With regard to the perceived engagement, the basic webcare message received an average score (M = 3.56). Similar to the perceived personalization and perceived informality, the first analysis showed a significant main effect for the main categories of CHV (F (1.92, 258.63) = 20.29, p < .001, η p 2 = 0.13), 8 on perceived engagement. Within-subject contrasts revealed that each main category differed significantly from the basic webcare message (all p's < 0.001). Post-hoc comparisons using Bonferroni correction showed Message Personalization and Invitational 5 Since eWOM type hardly showed any significant results, we report the findings of this additional variable here. We performed mixed ANOVA's per main CHV category with eWOM type as a factor (neutral vs. negative) for perceived personalization, perceived informality, and perceived engagement. No significant main effects of eWOM type were found for all three CHV measures (all p's ≥ 0.05), nor interaction effects between eWOM type and CHV main categories (all p's ≥ 0.05). We also performed mixed ANOVA's between CHV subcategories with eWOM type as a factor for perceived personalization, perceived informality, and perceived engagement. Only one main effect of eWOM type was found for perceived engagement, F (1, 135) = 4.36, p = .04, η p 2 = 0.03). Participants assessed the perceived engagement of webcare messages to neutral eWOM (M = 4.36, SD = 0.72) higher than to negative eWOM (M = 4.12, SD = 0.65). No other effects were found of eWOM type (all p's ≥ 0.05). 6 Mauchly's Test of Sphericity was not assumed (χ 2 (44) = 61.97, p = .04) and Greenhouse-Geisser was ε 0.90. Therefore degrees of freedom were corrected using Huynh-Feldt estimates of sphericity (ε 0.97). 7 Mauchly's Test of Sphericity was not assumed (χ 2 (5) = 63.66, p < .001) and Greenhouse-Geisser was ε 0.75. Therefore degrees of freedom were corrected using Huynh-Feldt estimates of sphericity (ε 0.77). 8 Mauchly's Test of Sphericity was not assumed (χ 2 (5) = 115.13, p < .001) and Greenhouse-Geisser was ε 0.63. Therefore degrees of freedom were corrected using Huynh-Feldt estimates of sphericity (ε 0.64).
Rhetoric did not contribute differently to the perceived engagement (p = .41), but both conditions did contribute more to this perception than Informal Speech (p personalization < 0.001, p rhetoric = 0.04). The analysis with regard to the different subcategories revealed a significant main effect for the CHV elements (F (9, 1215) = 15.10, p < .001, η p 2 = 0.10). All subcategories of CHV elements enhanced the perception of engagement compared to the basic webcare message (all p's ≤ 0.001), except for Personal addressing consumer (p = .22) and Contractions (p = .24). Next to differences with the basic webcare message, we also found differences between the CHV subcategories. Notably, a personal greeting contributed most to the perceived engagement in comparison to all other CHV elements (all p's < 0.001). This finding is similar to findings for perceived personalization.

Contribution of multiple CHV elements to perceived CHV
Next to the contribution of separate subcategories to the perceived CHV, we also investigated the impact when linguistic elements are combined in a webcare message. Four experimental conditions were tested: three conditions in which the webcare messages contained the three subcategories per main category, and one condition in which the webcare messages contained all nine subcategories. The findings are presented in Table 7.
With regard to the first CHV item, perceived personalization, a significant main effect of CHV elements was found (F (2.31, 311.52) = 77.04, p < .001, η p 2 = 0.36). 9 Post-hoc comparisons showed all four conditions differed significantly from each other (all p's < 0.05) with the highest perceived personalization for the webcare message containing all linguistic element categories, followed by all Message Personalization subcategories, and all Invitational Rhetoric subcategories. The webcare message containing all Informal Speech subcategories contributed least to the perception of personalization.
The second CHV item, perceived informality, also showed a significant main effect of CHV elements (F (3, 405) = 27.90, p < .001, η p 2 = 0.17). Post-hoc comparisons revealed the webcare message containing all elements differed significantly from the other three conditions (all p's < 0.001). Furthermore, the condition containing all Message Personalization categories scored higher on perceived informality than the condition containing all Invitational Rhetoric categories (p < .001), but no difference was found with the webcare message containing all Informal Speech categories (p = .08). Lastly, the Informal Speech condition and Invitational Rhetoric condition did not differ in perceived informality (p = .81). The third CHV item, perceived engagement, revealed a significant main effect of CHV elements as well (F (2.81, 379.44) = 34.58, p < .001, η p 2 = 0.20). 10 Post-hoc comparisons showed all four conditions differed significantly from each other (all p's < 0.01), except for the webcare message containing all elements of Informal Speech and the webcare message containing all elements of Invitational Rhetoric (p = 1.00). The finding that these two conditions scored lowest corresponds to the findings with regard to the perceptions of informality of the webcare messages.

Conclusion
It can be concluded that linguistic elements in webcare messages differ in their contribution to the perceived CHV, which was assessed with the perception of the message being personal, informal, and engaging. We first investigated the effects of single linguistic elements in a response and found that all categories of CHV were perceived as more personal and more engaging than the basic webcare message without Table 5 Characteristics of participants per between-subjects condition (type of eWOM).   CHV elements, with the exception of contractions and personal addressing the consumer. The data therefore confirm hypothesis 1 that predicted webcare messages containing one linguistic element of CHV will lead to higher levels of perceived CHV compared to webcare messages without linguistic elements of CHV. Moreover, a personal greeting contributed most to the perception of personalization and engagement compared to the other subcategories. Generally, we can conclude that Message Personalization enhanced the perceived personalization and engagement the most, supporting hypothesis 2. However, the contribution of single CHV elements to the perceived informality showed a less clear pattern as some subcategories were perceived as more informal compared to the basis webcare message (i.e., personal greeting and signature, visual elements and interjections, and well-wishing) while others did not (i.e., personal addressing consumer, contractions, and stimulating dialogues and showing sympathy and empathy). Furthermore, we showed that adopting multiple linguistic elements in a webcare message is fruitful: as expected in hypothesis 3, participants evaluated the perceived CHV the highest after reading webcare message that contained all CHV subcategories, followed by responses that contained only Message Personalization subcategories. Webcare messages that contained all subcategories of Invitational Rhetoric were seen as more personal that messages containing all subcategories of Informal Speech, but they did not differ in the perceived informality and engagement.

General discussion
The conversational human voice (CHV) is considered to be a key concept in improving relationships between brands and consumers via online communication (Kelleher, 2009;Kelleher & Miller, 2006) and has been studied intensively by scholars in the fields of communication science, public relations, and marketing communications. Previous research shows mixed results on the use of CHV in online brand messages which might be explained by Kelleher's definition of CHV allowing multiple ways to operationalize the concept in online brand messages. The present research addressed this issue by conducting two studies: an integrative literature review to investigate which linguistic elements are used to establish a sense of CHV in online brand messages and a perception experiment to investigate the contribution of linguistic elements in online brand responses to the perception of CHV.
Our integrative literature review shows CHV is operationalized in various ways in previous research, and reflection is missing on how the type and number of elements used in experimental materials relate to the perception of CHV. Based on the literature review, a taxonomy is created consisting of seventeen subcategories associated with three main CHV tactics: Message Personalization, Informal Speech, and Invitational Rhetoric. Furthermore, the perception experiment showed that the contribution of linguistic elements to the perception of CHV in online brand messages differs. Message Personalization seems to be key in creating CHV: linguistic elements in this category were most frequent in prior experimental research, and the perception experiment showed that their contribution to perceived CHV was the highest. Also, linguistic elements of Invitational Rhetoric were frequently used in prior research and the results of the experiment showed that they were second relevant contributors to the perceived CHV. Informal Speech, on the other hand, was less commonly present in previous research and the associated linguistic elements had a minor impact on participants' perceptions of CHV. Finally, the perception experiment showed adopting multiple elements of CHV in an online brand message results in a higher level of perceived CHV: it is thus beneficial for brands to include many linguistic elements of CHV in brand messages if they want to come across as an engaging conversation partner.

Theoretical implications
Prior studies on CHV differed in the use of linguistic element types to operationalize CHV as well as the number of linguistic elements in online brand messages. This variety could negatively impact the comparability, validity, and generalizability of experimental research (cf., Gretry et al., 2017;van Hooijdonk & Liebrecht, 2018). The current study's taxonomy of linguistic elements as well as their weighted contribution to the perception of CHV is a valuable resource that can be used to operationalize CHV and compare corresponding effects more systematically in future research.
Furthermore, the present research provides valuable insights into consumers' perceptions and expectations of a brand's communication style in online brand communication (e.g., Fournier & Avery, 2011;Kaplan & Haenlein, 2010) that can be explained with the Language Expectancy Theory (LET, Burgoon & Miller, 1985). LET is developed in the field of persuasion and assumes people have expectations and preferences concerning the language and message strategies of others in persuasive attempts. These communicative expectations are derived from three factorsthe communicator, the relation between communicator and receiver, and contextand based on cultural and sociological norms, leading to preferences for i.e., competent communication performance.
Our research shows consumers prefer message personalization in online brand communication in response to consumers' messages. By adopting personalization elements in their responses, brands are able to transform the online public environment into a more personal one: consumers could experience a more personal conversation when the brand includes a personal greeting and a signature to the message. Elements of invitational rhetoric could also tighten the brand-consumer relation. These elements are valuable contributors to the 'markets as conversations' approach (Searls & Weinberger, 2000) in which brands have to engage in a dialogue with their consumers. Apparently, these elements match consumers' preferences for online brand communication as well.
Informal speech, on the other hand, seems to correspond less with consumers' preferences for competent communication performance of brands. Although informal speech is frequently used by consumers on social media, it is less desirable for brands to adopt this communication style in their online communication. Building on LET, we suggest the adoption of informal speech desires customization. This can be explained with Role Theory (Sarbin & Allen, 1968) that states the evaluation and success of interactions depend on the appropriateness of the behavior of the interaction partner in regard to their social roles. An informal communication style is considered appropriate if communication partners know each other well, while a formal communication style is more suitable in case of unfamiliarity. This also applies for online brand communication (Gretry et al., 2017). Brands should therefore be aware of consumers' expectations and preferences regarding the brand as communicator, the relation between brand and consumer, and the accompanying context.

Managerial implications
Our research informs brands of the importance of linguistic elements used in online brand communication. In order to seamlessly participate in the interactive online environment of web 2.0 without being seen as party crashers (Fournier & Avery, 2011), they need to be aware of the impact of their communication style. The current study's taxonomy of linguistic elements as well as their contribution to the perceived CHV can be seen as a useful guide for marketers who aim to adopt a 'markets as conversations' approach that matches the preferences of their audience. Generally, it can be advised to use linguistic elements of message personalization and invitational rhetoric as these categories contribute most to the perception of CHV. Besides the quality of the chosen linguistic elements, quantity matters as well. Our study revealed that brands should include multiple elements of the main CHV categories in brand messages since this yields the highest perception of CHV.
Our research provides marketers concrete guidance in determining and implementing the communication style that is preferable for the brand. For example, they can discuss which amount of CHV and which corresponding linguistic elements matches the brand's values and the consumers' expectations. Consequently, once the preferred communication style and corresponding desirable CHV elements are determined, brands can make internal agreements on the strategic use of CHV amongst employees who communicate on behalf of the brand. Often multiple employees are involved in webcare communication and clear agreements help them to justify their choices and thereby contribute to a uniform brand image.

Limitations and directions for future research
This research has certain limitations that calls for future investigations. With regard to the integrative literature review, the study's dataset was biased toward international peer-reviewed papers written in English. Although this approach enhanced the comparability of previous research, potential relevant insights on CHV operationalizations from research papers written in other languages are absent in our taxonomy. Such a follow-up investigation could not only yield additional categories for our taxonomy (for example, the existence of formal versus informal personal pronouns, which is absent in English), it could also show cultural differences in, for example, operationalizations of informality (for example, Dutch and Flemish citizens differ in their perceptions of the formality of the possessive pronoun uw). Cultural differences could also affect CHV perceptions and subsequent effects. The Austrian study of Kniesel, Waiguny, and Diehl (2016), for example, revealed a corporate tone of voice positively affected attitudes compared to CHV, which they explained by the more 'content, goal and truth-oriented and less relationship-oriented' nature of Germans in comparison to English speaking people (p. 6).
Moreover, the taxonomy of linguistic elements derived from the three main CHV categories is rather strict while it is reasonable that CHV elements used in prior studies could be classified in multiple main CHV categories. For personal greeting, for example, it is shown both Dear Mr. Smith and Hi John were used to operationalize CHV. However, these personal greetings vary in informality; Hi John presumably contributes more to the perceived CHV than Mr. Smith, which in turn is more personal than a message without a greeting. The same is true for the personal signatures of the brand's representative, which vary from only first names and initials (Thomas, ^TO) to full names (Morgan Smith) and professional titles (Morgan Smith General Manager). Arguably, the perception of personalization and informality differs when the brand, the CEO or a staff member signs the message (e.g., Ghosh & Amar, 2018;Kim & Park, 2017;Kniesel, Waiguny, & Diehl, 2016;Sparks et al., 2016).
With regard to the perception experiment, the internal consistency of the scale used to measure perceived CHV was insufficient. This was somewhat surprising, as the scale was also used in the study of Liebrecht and Hooijdonk (2020) and turned out to be reliable. Although the outcomes of our research on the perception of CHV elements were comparable with the findings of Liebrecht and Hooijdonk (2020), it is advisable to adopt validated scales in future research, such as Kelleher's instrument with 11 items, to examine the perception of CHV (Kelleher, 2009;Kelleher & Miller, 2006), or a subset of these items as performed in other CHV studies (e.g., Dijkmans et al., 2015;Schamari & Schaefers, 2015;Sparks et al., 2016).
What is more, the present research did not examine the underlying assumption that the great variety in CHV operationalizations in prior research could have caused different effects on consumer-and brandrelated outcomes, such as evaluations of the interaction and the brand, trust, and reputation (e.g., Jahng & Hong, 2017;Jakic et al., 2017;Kelleher, 2009;Kerkhof et al., 2011;Park & Cameron, 2014;Yang et al., 2010). A meta-analysis could provide more insight into the relation between CHV operationalizations and their subsequent effects. Moreover, several factors could impact on these results, like demographic characteristics (e.g., age, gender, culture), brand characteristics (e.g., type of industry, brand image), and channel characteristics (e.g., platform type). With regard to the latter, CHV research was initially focused on corporate blogs (e.g., Kelleher, 2009;Kelleher & Miller, 2006) but scholars also examined CHV on review websites (e.g., Sparks et al., 2016;Zhang & Vásquez, 2014), and on social media (e.g., Barcelos et al., 2018;Crijns et al., 2017;Gretry et al., 2017;Kwon & Sung, 2011). These platforms can be categorized in platforms that are initiated, managed, and controlled by the brand (i.e. brand-generated platforms, such as corporate blogs), and platforms that are initiated by consumers (i.e. consumer-generated platforms, such as review websites). The perception and effects of CHV in online brand messages differ between these platform types, which could be explained by the extent to which consumers feel their privacy is violated by a webcare response (compare Schamari and Schaefers (2015) and van Noort and Willemsen (2012)).
For both scientific scholars and marketing communication practitioners, it is thus relevant to be aware of the impact of linguistic elements of CHV in online brand messages. Our taxonomy could assist them to adopt CHV more systematically in order to achieve an appropriate communication style that matches or even exceeds both the desires of consumers and the brand in order to truly engage in a conversation with the market, as originally intended by Searls and Weinberger (2000) and Kelleher (2009).

Autobiographical Note
Christine Liebrecht is assistant professor at the Department of Communication and Cognition, Tilburg University. Christina Tsaousi is student in the Research Master Language and Communication of Tilburg University and Radboud University Nijmegen. Charlotte van Hooijdonk is assistant professor at the Department of Languages, Literature and Communication, Utrecht University.