2.1. AI-driven chatbot determinants (Personalization, Interactivity, and Chatbot ease of use)

Personalization involves the ability of AI chatbots to analyze content about customer needs and behaviors, providing customized recommendations that improve customer satisfaction through virtual interactions [13]. Relationship marketing theory (RMT) asserts that trust is a core factor that strengthens customer-firm relationships [18], which may foster long-term commitment, engagement, and emotional connection. Extending the RMT to AI-related travel service experiences posits that machine-human interactions can foster trust, as human-to-human interactions do in traditional service delivery. In this context, personalization serves as a crucial relational instrument through which firms can demonstrate recognition, responsiveness, and individualized value to customers, which cultivates trust [13,18]. According to RMT, trust is a key component of long-term relationships that grows over time through personalized communication that exceeds customer expectations [23]. AI-driven chatbots that create personalized content or recommendations following customers’ behavior, likes, or localized needs promote a perception of personalized attention and affection. Such personalized interactions foster relational trust and boost perceived service quality, indicating that the AI-driven chatbots are dependable and prioritize the customer’s best interests [14,24]. Therefore, personalization may boost the firm’s relational value and benevolence, which can increase trust in AI-driven chatbots. It is proposed:

H1a: Personalization significantly influences chatbot trust.

Interactivity has become a foundational concept in service quality and relationship marketing research. It can be defined as the chatbot’s ability to engage in real-time, reciprocal communication with customers while maintaining conversational consistency [16,25]. Interactivity improves the functionality of AI technology, resulting in better responses, increased trust, and greater user control over the communication process [26]. Within the context of relationship marketing theory, interactivity facilitates two-way communication and responsiveness in relational exchanges, which may assist in building customer trust [18,24]. Previous studies demonstrated that chatbot interactivity enhances communication service quality, customer satisfaction, and service co-creation [14,16,27], which may assist in developing customer trust. AI-driven chatbots mimic human-like communication, acting as relationship agents that can improve through reciprocal communication flow, fostering relationship development and mutual trust [26]. Interactive chatbots reduce cognitive load and support dialogic planning adjustments. For example, customers who customize trip planning to meet their immediate needs can improve their trust in AI chatbots. The rising expectations of Chinese customers for responsive, intelligent service agents in digital ecosystems accentuate the strategic importance of interactivity in relationship quality. According to RMT, chatbot interactions may promote customer participation and develop trust. It is proposed:

H1b: Interactivity significantly influences chatbot trust.

Ease of use refers to “the degree to which a person believes that using a particular system would be free of effort” [17, p. 320]. Chatbot ease of use can be defined as the extent to which customers believe that interacting with a chatbot would be effortless, easy, and friendly [28]. Previous studies investigated the ease of use to understand user satisfaction and chatbot usage intentions [11,28]. However, little is known about the influence of chatbot ease of use on customer trust in chatbots, particularly in the travel industry. Clear, understandable, and easy-to-use customer-chatbot interactions can assist in building initial trust in chatbots [29]. Relationship marketing theory asserts that supporting easy interactions and reducing customer effort improve the relational exchange process, enhancing customer trust and satisfaction in services where ambiguity and complexity may undermine it [23,24]. Ease of use has emerged as a key relational factor in China’s travel industry, where AI is prevalent, and tech-savvy Chinese travelers demand easy, fast, and effective interactions. As a result, chatbot ease of use can be an effective measure to build trust and relationships with customers.

H1c: Chatbot ease of use significantly influences chatbot trust.

2.2. Chatbot trust

Trust can be a critical factor in developing dependable, successful interactions between humans and chatbots [1]. Previous research highlighted the importance of trust for developing customer interactions with AI chatbots [29]. However, prior research has paid little attention to investigating customer trust in AI chatbots. Previous studies have focused on evaluating the effects of chatbot trust on user intentions to use chatbots and e-brand loyalty [29–31]. According to commitment-trust theory (CTT), trust serves as the bedrock of long-lasting relationship interactions [18]. Previous research indicates that CTT plays an important role in enhancing consumer-brand interactions and other key stakeholders in the social media context [32]. Therefore, chatbot trust may reduce perceived risk and uncertainty, which can motivate customers to interact with chatbots more frequently. Similarly, relationship marketing theory supports that when customers feel relational trust, they reciprocate with proactive behavior, including interactive questions, co-creation, and service advocacy [24]. Chinese travel platforms offer a wide range of service quality choices, and travelers can rely on relational cues to assess chatbot trust. As a result, when chatbots respond quickly, precisely, and empathically, customers may be more inclined to interact with them more deeply. It is proposed:

H2a: Chatbot trust significantly influences engagement.

RMT posits that personalized content and recommendations can indicate relational value and responsiveness, reflecting care and trust at the individual level [23,24]. Interactivity accentuates two-way communication through real-time conversation and receptive follow-ups [16]. It represents relational responsiveness, which can increase trust. CTT postulates that such responsive communication decreases perceived risk and uncertainty and fosters trust in the service agent over time [18]. Chatbot ease of use signifies simple service accessibility and decreases relational challenges, indicating care for customers’ comfort, which can enhance trust and satisfaction [28]. Previous studies have shown that personalization and interactivity can positively influence user satisfaction through communication quality and the virtual flow experience [13,16]. Likewise, research revealed that ease of use can enhance actual use of AI chatbots by shaping users’ intentions to use chatbots [11]. However, research examining the determinants of AI-driven chatbots, such as personalization, interactivity, and ease of use on customer engagement via chatbot trust, particularly in the Chinese travel industry, remains scarce. Therefore, this study intends to offer new insights into literature. It is proposed:

H2b: Chatbot trust positively mediates the relationship between personalization and engagement.

H2c: Chatbot trust positively mediates the relationship between interactivity and engagement.

H2d: Chatbot trust positively mediates the relationship between chatbot ease of use and engagement.

Brand success in the service sector, specifically travel, depends heavily on building trust and cultivating long-term customer relationships. CTT suggests that trust is a key antecedent of relational exchanges and emotional attachment with a brand [18]. RMT postulates that trust develops through continued value-generating relationships and provides the basis for attachment, which can boost emotional connection [23,24]. These theories provide a strong foundation for understanding how AI-driven chatbot trust influences customers’ relational and emotional responses, including emotional connection. Emotional connection refers to the feelings that customers experience when they are emotionally attached to a brand [33]. This study defines emotional connection as a customer’s sense of psychological empathy, intimacy, and emotional association with the AI chatbot. AI chatbots are often characterized as AI companions for their ability to foster relational closeness and empathy, which enhances deeper relationships with users [34]. RMT emphasizes that effective relationships extend beyond transactional interactions, integrating trust and emotional connections to strengthen relational strength [18]. Previous studies have emphasized the outcomes, such as chatbot usage intentions and customer satisfaction [13,16,19]. Research focusing on the associations between AI chatbot trust and customer emotional connection is rare. It is hypothesized:

H3a: Chatbot trust significantly influences emotional connection.

AI chatbot determinants, such as personalization, interactivity, and ease of use, may serve as building blocks for developing emotional connections with customers via trust. Relationship marketing and trust commitment theories support the notion that favorable relationships enhance trust, resulting in higher customer attachment to the brand [18,23,24]. Previous studies have revealed that personalization, interactivity, and ease of use can improve customer satisfaction and their intentions to use chatbots [11,13,28]. These factors may motivate customers to build trust in chatbots [29,31], potentially leading to greater attachment to chatbots [1]. In the Chinese travel industry, where AI implementation and adoption rates are rapid and customer digital literacy is high, personalization, interactivity, and ease of use can be factors that influence customer trust in AI travel chatbots, which may deepen their relationship with chatbots. In line with RMT and CTT, it can be proposed:

H3b: Chatbot trust positively mediates the relationship between personalization and emotional connection.

H3c: Chatbot trust positively mediates the relationship between interactivity and emotional connection.

H3d: Chatbot trust positively mediates the relationship between chatbot ease of use and emotional connection.

2.3. Engagement and emotional Connection

RMT posits that engagement is imperative for developing relational mechanisms, which connect transactional exchanges and long-term relationships [24]. Previous studies have argued that engagement serves as a key component of relationships and contributes to customer satisfaction, purchase intentions, and price premium [29,35,36]. Research investigating the impact of engagement on emotional connection with AI chatbots in the Chinese travel industry remains untapped. CTT reiterates its argument by demonstrating that consumers are prone to form emotional relationships when they engage in relational commitments, as reflected in their behaviors [18]. In the Chinese travel industry, where AI-driven travel chatbots interact with different customers, engagement is important for building relationships through information exchange and continuous collaborative dialogues, which may foster customer emotional resonance. In line with RMT and CTT, it is hypothesized:

H4a: Engagement significantly influences emotional connection.

Service relationship models anchored in RMT and CTT emphasize that trust is an integral component in fostering deeper cognitive and affective relationships [18,24]. The proposed hypothesis might extend relationship theories by demonstrating that engagement positively impacts the relationship between chatbot trust and emotional connection in contemporary AI-driven travel chatbot contexts. Previous research supports the premise that trust can enhance customer engagement [29], and engagement may foster customer satisfaction [36]. Emotional connections are more likely to form when customers trust AI chatbots, which may lead to meaningful relationships between customers and AI chatbots. Therefore, engagement can be a central construct that meaningfully translates trust into emotional connections. It is hypothesized:

H4b: Engagement positively mediates the relationship between chatbot trust and emotional connection.

2.4. Chatbot social presence

Social presence refers to the degree to which a channel facilitates individuals to experience other individuals as psychologically present [21]. Social presence theory [37] was historically applied to human-to-human interactions, but its importance has grown in the context of AI-human communication. This study postulates that chatbots with a high social presence, such as warm, anthropomorphic, and empathetic, as well as socially driven conversational clues, foster relational significance of trust and behavioral consequences. Social presence boosts parasocial interaction, perceived dialogue quality, and psychological involvement, which in turn may increase engagement [38]. CTT suggests that trust acts as a cognitive driver for relational behavior and engagement [18], but without relational signals, trust may be considered as utilitarian or abstract. Prior studies have emphasized the role of chatbot social presence to understand users’ continuous intentions, trust, and social-oriented communication [20,22]. However, research on the moderating effects of chatbot social presence is rare. With the support of CTT and social presence theory, it hypothesized:

H5a: Chatbot social presence positively moderates the relationship between chatbot trust and engagement; trust has a greater effect on engagement when chatbot social presence is high and vice versa.

RMT highlights that relationship warmth and commitment are vital for cultivating a deeper emotional attachment [23]. Additionally, CTT demonstrates that trust is a key factor for developing relationships; hence, social cues enhance its impact by reinforcing relational value [18]. Emotional connection can be described as a feeling of psychological intimacy, love, and affective relationships with the AI chatbot, which serves as a brand ambassador [39]. Social presence theory posits the significance of interactive communication across different media in conveying verbal and visual information, and for influencing individuals’ ability to communicate their perceptions of others’ physical presence [40]. Previous studies have investigated the role of social presence in shaping dialogues and parasocial interactions with chatbots, and chatbot continuous intentions [20,38]. Previous research supports the argument that continuous dialogues and parasocial interactions may enhance closeness, love, and affective relationships with chatbots [38,39]. Therefore, it can be proposed:

H5b: Chatbot social presence positively moderates the relationship between chatbot trust and emotional connection; trust has a greater effect on emotional connection when chatbot social presence is high and vice versa.

3.1. Sampling and data collection procedures

This study targeted Chinese adult tourists who primarily interacted with travel-related AI chatbots, such as Fliggy, Ctrip, Dianping, WeChat, Alipay, Meituan, Qunar, and trip.com. These digital platforms are China’s leading travel service providers, offering real-time trip-planning information and active customer care via AI chatbots. Consistent with previous research, this study used purposive sampling to collect data from respondents [42]. Purposive sampling was deemed suitable to verify that participants had adequate knowledge and experience to effectively assess the study constructs. The respondents were recruited by sharing an online survey QR code via WeChat (a leading platform) and contacting them personally. The survey was posted in Chinese on a leading survey platform (i.e., www.wjx.cn) from February 2, 2025, to March 3, 2025. Respondents had to rate each question on a scale of seven points. For example, they had to provide a higher rating (i.e., 7) to questions on which they were extremely in agreement and a lower rating (i.e., 1) to those on which they were extremely in disagreement. The respondents who completed the survey were rewarded with 2 yuan as a token of appreciation. The modest financial incentive helps reduce potential risks and fosters respondents’ engagement in providing careful responses to survey questions. Each participant was allowed to respond to the survey once. We applied several measures to enhance data quality, such as IP address tracking to prevent duplicate submissions, screening questions to confirm eligibility, and removing straight-lined responses. This study strictly adhered to ethical principles about the confidentiality and anonymity of respondents’ data.

Initially, 90 respondents were recruited for a pretest to evaluate tourists’ perceptions of the survey questions and to validate the scales. After data screening, 80 responses were considered for pretest analysis. The results revealed that all outer loading values were greater than 0.70, except for chatbot trust item 2 (i.e., 0.65). Likewise, Cronbach’s alpha and composite reliability values exceeded 0.70, and convergent validity (AVE) values surpassed 0.50, demonstrating the reliability and validity of the survey items [43]. After receiving feedback from some respondents, a few initial check questions were revised, and the survey was distributed to the mass population, resulting in 750 new responses received. After data filtration, 690 responses were included in the analysis. Table 1 illustrates the demographic information of the respondents. This study mainly emphasized the younger demographic (i.e., aged 18–33 years) because they are technology-savvy, better educated, and an important market segment for many companies [10,41,44]. The data was accessed on March 4, 2025, for analysis purposes. However, the authors did not have access to identifying information about respondents.

Download:

• PNG
  larger image
• TIFF
  original image

Table 1. Respondent demographic information.

https://doi.org/10.1371/journal.pone.0356073.t001

This study involves human participation and was approved under Ref. No. HGNU/ERC/25/0102 dated 2025-01-10 by the Ethical Review Committee at the Business School, Huanggang Normal University, Huanggang, China. This study was conducted in accordance with the local legislation and institutional requirements. The participants were provided with written informed, voluntary consent to participate in this study. The minors were excluded from participating in the study. Participants had the option to withdraw from the survey at any time during its completion.

3.2. Measures

This study measured 25 items to discover novel insights in a relational intelligence framework (See Fig 1). The scales were adjusted from previous research published in top-tier journals. The determinants of AI-driven chatbots, such as personalization and interactivity, were measured using three items each, and chatbot ease of use was measured with four items [13,16,45]. Relational outcomes, such as chatbot trust and emotional connection, were assessed using three items each, and engagement was assessed with four items [29,33,46]. Chatbot social presence was assessed with five items [21].

3.3. Common method bias

It is critical to assess data bias, as survey methodologies may induce self-selection bias. This study employed two established tests to identify bias in the data. First, we performed Harman’s single-factor as a diagnostic test for bias [47,48]. The findings indicated that a single factor accounted for 48.95% of the total variance, which was below the 50% threshold, suggesting that the bias does not pose a significant threat to the data. However, the Harman single factor test has limitations and is inadequate for demonstrating the absence of bias in the data. Therefore, we employed a full collinearity test using the variance inflation factor (VIF) [49]. The findings revealed that all VIF values are below the threshold of 3.30, indicating no potential threat of bias in the data (see Table 2).

Download:

• PNG
  larger image
• TIFF
  original image

Table 2. Construct reliability, convergent validity, and collinearity statistics (VIF).

https://doi.org/10.1371/journal.pone.0356073.t002

Assessing endogeneity is essential since its absence shows that the data is unbiased, ensuring the validity of the findings and their significant contribution to literature [43]. The endogeneity issue arises when exogenous variables are linked to the error term of endogenous variables. We used the Gaussian copula approach to evaluate the potential existence of endogeneity [50]. The results indicate that all Gaussian copula conditions are statistically non-significant (p > 0.05), suggesting that endogeneity does not pose a serious threat to the structural model relationships [50,51]. As a result, the findings validate the robustness and reliability of the structural model estimates.

4.1. Measurement model evaluation

Measurement of the outer model evaluation is critical for the constructs’ reliability and validity. Current literature demonstrates that scholars must follow standard procedures to ensure the robustness of the measurement model. Therefore, robust criteria for indicator loadings, Cronbach’s alpha, composite reliability, AVE, and HTMT must be followed to ensure the accurate reliability and validity of the proposed model [43,53]. Table 2 shows that all the values of internal consistency reliability, such as indicator loading, Cronbach’s alpha, and composite reliability, are greater than 0.70, and convergent validity (i.e., AVE) values surpassing 0.50, demonstrating that the measurement model has satisfied the threshold [43]. The findings revealed that all constructs, including AI chatbot determinants (personalization, interactivity, chatbot ease of use) and AI chatbot relational aspects (chatbot trust, engagement, emotional connection, and chatbot social presence), demonstrating construct reliability and validity.

Discriminant validity was calculated using the well-renowned HTMT ratio criterion [53]. Table 3 shows that all HTMT values are below 0.85, indicating that the measurement model has achieved sufficient discriminant validity [53]. The findings demonstrated that AI chatbot determinants (personalization, interactivity, and chatbot ease of use) and AI chatbot relational aspects (chatbot trust, engagement, emotional connection, and chatbot social presence) are significantly distinct from each other, thereby validating discriminant validity. Furthermore, we assessed discriminant validity by analyzing item cross-loadings. Table 4 indicates that all items exhibit the highest loading values on their corresponding constructs, which consistently exceed the cross-loading values of other constructs. These findings demonstrate that each construct effectively represents a unique concept, which supports discriminant validity [43].

Download:

• PNG
  larger image
• TIFF
  original image

Table 3. Discriminant validity; Heterotrait-monotrait ratio.

https://doi.org/10.1371/journal.pone.0356073.t003

Download:

• PNG
  larger image
• TIFF
  original image

Table 4. Cross loadings.

https://doi.org/10.1371/journal.pone.0356073.t004

4.2. Structural model evaluation

The structural model evaluates the inner model. It examines multicollinearity, R², model fit, Q², effect size, and the relationships of proposed hypotheses [43]. First, we examined multicollinearity using the variance inflation factor (VIF) metric. Table 2 shows that all VIF values are below 3, demonstrating that the data are free from multicollinearity and bias [54]. Second, the model’s explanatory power was assessed via the R² metric. Fig 2 illustrates R² values of 0.64 for chatbot trust, 0.70 for engagement, and 0.78 for emotional connection, demonstrating that the proposed model sufficiently explains the variance in all endogenous constructs [55]. Third, model fit was determined through SRMR metrics. The results showed that the SRMR value was 0.05, which is less than 0.08, explaining an excellent model fit [54]. Fourth, the model’s predictive relevance (Q²) was calculated using blindfolding procedures. The results showed that Q² values of 0.36 for chatbot trust, 0.39 for engagement, and 0.44 for emotional connection, demonstrating sufficient predictive relevance of the proposed model [43]. The effect size (f²) was estimated to assess the practical significance of structural interactions [43]. Research indicates that f² values of 0.35, 0.15, and 0.02 correspond to large, moderate, and small effects, respectively [56]. The study found (see Table 6) that personalization had a moderate influence on chatbot trust (f² = 0.235), whereas interactivity and ease of use had small effects (f² = 0.103). Chatbot trust had strong effects on engagement (f² = 0.420) but moderate effects on emotional connection (f² = 0.149). Likewise, engagement had a moderate effect on emotional connection (f² = 0.163). Chatbot social presence had a small but meaningful moderating effects on the association between chatbot trust and engagement (f² = 0.078) [43,57], but little impact on the relationship between chatbot trust and emotional connection (f² = 0.017).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 2. Structural Model Results Source: “Authors’ own work”.

https://doi.org/10.1371/journal.pone.0356073.g002

This study applied bias-corrected accelerated bootstrap procedures with 10,000 subsamples (two-tailed, α = 0.05) to estimate path models and test hypotheses [43]. Table 5 presents a sequential path analysis that includes various model types and combinations to validate the findings. Model 1 shows the impact of AI chatbot determinants, such as personalization, interactivity, and ease of use, on trust in AI chatbots. Model 2 demonstrates the impact of AI chatbot determinants, such as personalization, interactivity, and ease of use, on user engagement and emotional connection. Model 3 presents results with mediation, whereas Model 4 includes comprehensive results, outlining direct, indirect, and interactive effects. These findings reinforce the validation of the results. Table 6 and Fig 2 explicitly demonstrate the structural model results of hypothesized relationships. The findings revealed that AI-driven chatbot determinants, such as personalization, interactivity, and chatbot ease of use, significantly improved chatbot trust, supporting the H1a-1c. Likewise, chatbot trust significantly enhanced engagement and emotional connection. Therefore, H2a and H3a were supported. The findings also revealed the positive effects of engagement on emotional connection, supporting H4a.

Download:

• PNG
  larger image
• TIFF
  original image

Table 5. Sequential path analysis.

https://doi.org/10.1371/journal.pone.0356073.t005

Download:

• PNG
  larger image
• TIFF
  original image

Table 6. Hypothesis assessment results.

https://doi.org/10.1371/journal.pone.0356073.t006

We conducted an analysis using bias-corrected accelerated bootstrapping procedures with 10,000 subsamples on the proposed model [43]. We identified the mediation effects by assessing the significance of indirect effects [58,59]. Table 7 presents the significant indirect effects of personalization (β = 0.247, p < 0.001; β = 0.151, p < 0.001), interactivity (β = 0.174, p < 0.001; β = 0.106, p < 0.01), and chatbot ease of use (β = 0.174, p < 0.001; β = 0.106, p < 0.01) on engagement and emotional connection, respectively, thereby supporting H2b-2d and H3b-3d. In contrast, engagement was identified as a complementary mediator (β = 0.222, p < 0.001), thus supporting H4b. These findings indicate that trust in chatbots is highly essential, while engagement serves as a significant mediator in enhancing relationships among tourists in the Chinese travel industry.

Download:

• PNG
  larger image
• TIFF
  original image

Table 7. Mediation Results.

https://doi.org/10.1371/journal.pone.0356073.t007

The moderating results were calculated using the interactive effects [43,54]. The findings revealed that chatbot social presence x chatbot trust significantly boosted engagement, supporting H5a as (β = 0.206, p < 0.01). Fig 3 shows that higher chatbot social presence and trust significantly enhances tourist engagement toward AI chatbots. In contrast, the interactive effects of chatbot social presence x chatbot trust did not influence emotional connection, which did not support H5b (β = 0.086, p > 0.05).

Download:

• PNG
  larger image
• TIFF
  original image

Fig 3. Moderation Results Source: “Authors’ own work”.

https://doi.org/10.1371/journal.pone.0356073.g003

4.3. Asymmetric analysis (fsQCA)

Symmetric (PLS-SEM) analysis evaluates the effects of exogenous variables on endogenous variables [43]. Asymmetric analysis is based on causation, with different combinations of factors contributing to the outcomes, and fsQCA (Fuzzy Set Qualitative Comparative Analysis) is specifically designed to demonstrate equifinal, conjunctural, and asymmetric causation [60]. Previous research shows that fsQCA is important for theoretical advancement and is appropriate for large samples and regression-based frameworks [61].

4.3.2. Necessary conditions analysis.

A necessary condition analysis was undertaken to determine which conditions are required to generate a strong emotional connection. The proposed fsQCA guidelines describe a condition as necessary if its consistency exceeds 0.90 [65,66]. Table 8 demonstrates that both the presence and the negated condition are insufficient to meet the required threshold. Despite a high consistency of 0.818, engagement remained below the established threshold. These findings suggest that no single condition is necessary for developing emotional connections; instead, various combinations of conditions may affect the emotional connection of tourists. These findings support the configurational mechanism that underlies fsQCA [67]. The findings serve as a roadmap for further sufficiency analysis, which explores how various combinations of conditions foster emotional connection.

Download:

• PNG
  larger image
• TIFF
  original image

Table 8. Necessary conditions analysis.

https://doi.org/10.1371/journal.pone.0356073.t008

4.3.3. Truth table and configural outcomes.

After conducting the necessary condition analysis, a truth table was developed to identify the mechanisms for generating configurations that promote emotional connection (see Table 9). In line with existing literature and considering the larger sample size (n = 690), a frequency threshold of five cases and a consistency threshold of 0.900 were used to retain configurations that are both empirically meaningful and theoretically relevant [66,67]. Following that, a sufficiency analysis was conducted to identify various configurations that exhibit a strong association with emotional connection. We analyzed both parsimonious and intermediate solutions to differentiate between core and peripheral conditions. The existing literature indicates that conditions present in both the parsimonious and intermediate solutions are categorized as core conditions, while those found solely in the intermediate solution are categorized as peripheral conditions [62,66,67]. The results indicate that parsimonious solutions, including PRS, CEU, and ENG, are identified as core conditions that serve as key factors of emotional connection, whereas INT, CBT, and CSP have been classified as peripheral conditions, suggesting that their contributions differ across various alternative configurations.

Download:

• PNG
  larger image
• TIFF
  original image

Table 9. Truth table.

https://doi.org/10.1371/journal.pone.0356073.t009

Table 10 shows the intermediate solutions, revealing various configurations that foster a strong emotional connection. Some configurations greatly strengthen tourists’ emotional connection toward AI chatbots. For example, PRS*INT*CBT*ENG demonstrated the optimum raw coverage = 0.513 and consistency = 0.940, indicating that the combined effect of personalization, interactivity, chatbot trust, and engagement contributes to emotional connection. The PRS*INT*CEU*ENG configuration yields a raw coverage = 0.487 and consistency = 0.925, suggesting that the interplay of personalization, interactivity, chatbot ease of use, and engagement effectively fosters emotional connection. The PRS*CEU*CBT*ENG*CSP configuration shows a raw coverage = 0.469, and consistency = 0.956, highlighting the complementary role of chatbot trust and social presence as peripheral conditions in addition to personalization, chatbot ease of use, and engagement as core conditions, in strengthening emotional connections. The INT*CEU*CBT*ENG*CSP configuration has a raw coverage of 0.473 and a consistency of 0.949, highlighting the roles of interactivity, chatbot trust, and social presence as peripheral conditions, along with chatbot ease of use and engagement, in promoting emotional connection. These configurations illustrate the principle of equifinality by showcasing various combinations of chatbot determinants and relational mechanisms that foster emotional connection [66,67]. The overall solution coverage (0.687) demonstrates that reported configurations offer substantial evidence to foster emotional connection, while the overall solution consistency (0.879) exceeds the minimum threshold of 0.75, which increases the reliability and validity of the discovered configurations [66]. These findings collectively demonstrate that both core and peripheral configurations significantly enhance emotional connection. Particularly, engagement emerged as a core condition in all configurations, underscoring its vital role in strengthening tourists’ emotional connection with AI chatbots.

Download:

• PNG
  larger image
• TIFF
  original image

Table 10. Configuration results.

https://doi.org/10.1371/journal.pone.0356073.t010

5.1. Theoretical contributions

This study significantly contributes to relationship marketing theory (RMT), commitment-trust theory (CTT), and social presence theory in the context of AI chatbots in China’s travel industry. Traditional RMT emphasized human-to-human interactions in enhancing trust [18,24]. Our findings extend these interactions from machine-to-human to improve individuals’ trust in machines. For example, AI chatbot determinants, including personalization, interactivity, and chatbot ease of use, can help develop trust. The findings derived from the perspectives of relationship marketing, trust–commitment, and social presence theories enhance the relationship between AI chatbots and tourists within the context of China’s emerging travel industry [18,23,40]. The findings strengthen CTT in the AI context, indicating that chatbot trust can provide relational support related to human-machine interactions. The findings also contribute to RMT by fostering AI engagement and emotional connections, supporting AI-traveler relationships, and integrating another layer to prevailing relationship theories. The findings contribute to CTT and RMT [18,23,24] by indicating that engagement is a relational driver that enhances emotional connections toward AI chatbots. In addition, engagement as a positive mediator bridges the gap between chatbot trust and emotional connection in the AI context. The findings contribute to the social presence theory in the context of the burgeoning domain of human-AI interactions [37,40]. Social presence theory offers a robust framework for predicting chatbot social presence and its effects on cognitive and behavioral outcomes (i.e., engagement).

5.2. Managerial implications

This study offers several guidelines for digital marketing managers. First, travel platforms can improve tourist trust by providing customized AI chatbot services that enable effortless interactions with minimal input and smooth navigation. They can implement deep personalization via tourist data and real-time behavioral tendencies. Improved interactions should reflect consciousness, awareness, and empathy, aligning with travelers’ expectations through harmonious, respectful language. In addition, measuring relational key performance indicators (KPIs) like trust and commitment is essential to sustain relational value. Second, the findings showed that chatbot trust is highly relevant in the Chinese travel industry. All online travel agencies (OTAs) should enhance trust by ensuring data privacy, transparency, consistency, and response accuracy, which will help improve interactions and foster emotional ties with customers. Third, the findings showed that along with chatbot trust, OTAs should plan campaigns to enhance customer engagement, as it leads to emotional connections. In addition, managers must prioritize building emotional relationships with customers, which can lead to positive WOM, loyalty, and advocacy. Instead of focusing on chatbot functions, managers should emphasize chatbots as relational agents to gain a competitive advantage in China’s dynamic OTA market. Finally, considering the significant effects of chatbot social presence, OTAs must integrate it with consistent personalized narratives, service quality, and cultural relevance to promote deeper engagement. Likewise, the dual-layered strategy, such as social presence and trust, can elevate customer-chatbot relational dynamics in the Chinese AI-driven travel market.

5.3. Limitations and future research agenda

This study has several limitations. For example, this study was conducted exclusively in the context of China. Future research may evaluate the role of AI-driven chatbots in developing relational outcomes in other countries, specifically in Europe, or conduct cross-cultural research to deepen insights into AI-driven chatbots. This study used a cross-sectional methodology, with all data taken from a single source. While proposed relationships are theoretically valid and supported by empirical evidence, the cross-sectional design is insufficient to facilitate definitive causal findings. Future research should employ longitudinal or experimental methods to provide additional evidence regarding the causal relationships among chatbot determinants, trust, engagement, and emotional connection. This study integrated a single mediator (i.e., chatbot trust) into the AI relational framework. Future research may explore other mediators, such as service or relationship quality. This study incorporated a single moderator (i.e., chatbot social presence); future research may incorporate other relational variables, such as attachment or evangelism, to strengthen the relationship theories. This study employed purposive sampling and primarily focused on digitally oriented respondents aged 18–33 years old, which may limit the generalizability of the findings. Future research can utilize other sampling techniques, such as random sampling, and gather data from a diverse range of age groups to generalize the findings to the entire population of Chinese tourists. Finally, this study did not consider control variables. Researchers may integrate control variables, such as gender, age, income, and education, into the proposed framework to reveal more compelling insights.