2.1. Technologies and travel planning

While previous similar studies have analyzed the influence of chatbots in various fields and sectors such as commerce [29,30], medicine [31,32] or education [33], among others, studies in tourism are still incipient and no one the influence of this technology in travel planning.

Technological evolution has facilitated travel planning in recent decades, and academic studies have focused on analysing how diverse technologies are used and what functionalities they have in this planning. Initially, studies were carried out on information technologies and social media [34,35]. Fotis et al. [34] demonstrated that social media are essential tools influencing tourists’ travel decisions and that user-generated content is perceived as more trustworthy than official tourism websites. Xiang et al. [36] also demonstrated the growing use of social media in travel planning. Similarly, Huang et al. [37] highlighted the importance of travel websites, social media, and smartphones in travel planning and tourist satisfaction. Subsequently, other types of smart technologies, such as VR and AR or travel recommenders, were studied [38]. Disztinger et al. [39] and Gibson and O’Rawe [40] demonstrated the importance of VR in promoting destinations and generating satisfying pre-experiences that influence travel planning and decisions. Similarly, Ahmad et al., [41] also analyzed AR during Covid and demonstrated its influence on intentions to travel to destinations in the future. More recently, AI in its various forms [chatbots, ChatGPT] has also been studied in travel planning [42]. Shi et al. [42] investigated the impacts of systematic and heuristic cues on travellers’ cognitive and emotional trust and its influence on travel adoption intention toward artificial intelligence AI-based recommendation systems in travel planning. Elizalde-Ramírez et al. [43] demonstrated the importance of AI planning models to offer quality recommendations fully adapted to the needs of tourists when planning trips. Recently, various studies are also emerging that demonstrate the influence of GPT Chat on travel planning [28,44,45]. However, studies on chatbots in travel planning are lacking.

Various studies that have analyzed the use of technologies in travel planning have focused on technological acceptance, applying the Technology Acceptance Model [TAM]. However, the TAM is more common in studies during the trip than exclusively in planning. Some studies analyze the use of user-generated content [46,47] in social media [35,48] in travel planning. Others focus on different technologies such as VR [40] or AR [41] or the use of personalized location-based mobile tourism applications [49], but studies on chatbots that apply TAM are also scarce [50] and not focused on travel planning.

2.2. Hypothesis development

2.3. User satisfaction, usage continuance and destination visit intention

The relationship between information quality and user satisfaction with chatbots has been demonstrated in several studies. For example, in mobile banking, satisfaction with the tool also positively and significantly affects user continuance intention [69].

Ashfaq et al. [70] argue that IQ is essential in developing chatbots for customer service and positively affects user satisfaction and continuity intention. In the same way, Hsiao and Chen [70] analyzed service quality, trust, and satisfaction to predict users’ intention to continue using chatbots to order food in restaurants, and the results showed that satisfaction indeed had significant direct effects on the intention to continue using chatbots. Therefore:

Previous studies have shown the importance of DMOs’ online platforms [websites and social media] in creating the destination image and destination visit intention [71]. Likewise, studies on smart technologies [72,73] and tourism applications like AR [74] or VR experiences [75] or Live streaming [76] have also shown that certain technology factors increase usage intention and destination visit intention. It has been demonstrated that using smart tourism technologies increases destination visit intention [77].

In the field of chatbots, social presence and imagery processing have been shown to influence chatbot continuance intention [78]; and in tourism chatbots, a study on online travel agencies chatbots showed that understandability, reliability, assurance, and interactivity positively influence chatbot continuance usage intention, although technology acceptance moderates this influence [79]. Similarly, Pillai and Sivathanu [50] showed that ease of use, perceived usefulness, perceived trust, perceived intelligence, and anthropomorphism influence usage intention in tourist chatbots, while technological anxiety moderates this effect. A study of food-ordering chatbots showed that anthropomorphism, service quality, and problem-solving are the aspects that influence user satisfaction and usage intention [70].

In destination chatbots, Orden-Mejía and Huertas [80] analyzed the effect of chatbot usage intention on the formation of the destination image, demonstrating that informativeness and empathy are the main attributes that influence user satisfaction and the formation of the destination image and also influence the visit intention. Similarly, Tosyali et al. [15] showed that the informativeness of the chat affects the destination image and the visit intention. However, it is necessary to continue analyzing the factors that motivate the continuance of chatbot usage intention and its relationship with destination visit intention. Therefore:

H9: The destination chatbot’s continued usage intention positively impacts tourists’ visit intention.

3.1. Data collection

The explanatory study relied on a quantitative method because it focuses on objectives and measures them through actions and opinions. A survey strategy was used for data collection via pre-designed questionnaires. Regarding the sample size, there is evidence suggesting that 100 cases are sufficient when there are three factors with three or four indicators each. However, it is also recommended that the sample size should be at least around 200 cases, even with high communalities or uniqueness, or well-determined factors. In our study, we analyzed 284 cases.

In this study, we employed a convenience sampling method for data collection, which involves selecting participants based on their accessibility and proximity to the researchers. In our case, university students attending virtual classes were readily available and easily accessible through the Microsoft Teams platform. Given the hybrid academic environment during the COVID-19 pandemic, this method allowed us to efficiently engage participants without disrupting their regular class schedules. Although the sample was not randomly selected, the students provided valuable information as they could represent potential tourists interacting with the chatbot in a realistic scenario.

The data collection method was conducted in three phases. First, prior to requesting permission from several instructors to take a few minutes of their class to conduct the experiment, we entered the Microsoft Teams platform to contact students remotely because the university at the time of the experiment maintained a hybrid modality [virtual and in-person classes] because of COVID-19. Once in the digital classroom, we present the study without revealing its objectives to avoid bias and explain the characteristics of tourism technologies, including chatbots. Second, we asked the students to enter the link that directed them to the “QuitoGuide” chatbot. Once in the application, they had to exchange messages for 10 minutes, enough time for this type of experiment [66]. Participants were free to interact with the application, imagining their intention to visit the city of Quito. We emphasize that the university is located on the coast of Ecuador, 420 km away from Quito. With this, we consider having obtained a representative sample of the tourist population. Third, after the human chatbot conversational session, students had to access a link to complete an online questionnaire about their experience with the chatbot. The questionnaire was developed using Google Forms, and quality mechanisms were considered to ensure that each user only submitted the form once. Finally, the participants were thanked for their collaboration, and we said goodbye. The entire procedure lasted approximately 30 minutes on average. In the end, we obtained 312 responses between men and women. However, 28 were discarded for having missing data, outliers or being an outliner.

3.2. Measurement

The questionnaire design comprised multiple items and constructs adapted from previous studies. To measure the indicators, we used seven-point Likert-type scales. We measured information quality using seven items adapted from [85]. The scale developed by [69] was used to measure perceived usefulness. Perceived enjoyment was examined with four items used in the instrument developed by [86]. User satisfaction was measured using a five-item scale from [63,86]. The continuance intention dimension was examined with five items from [67]. Finally, the intention to visit destination [IVD] factor was evaluated based on five items adapted from [87] (see Appendix 1).

3.3. Data analysis

A partial least squares structural equation modeling PLS-SEM model was used. Variance-based partial least squares structural equation modeling [PLS-SEM] is considered the ideal data analysis method to simultaneously evaluate all structural paths between variables in a conceptual model. In addition, it can achieve a higher statistical analysis level than others based on covariance [88]. For example, it maximizes the variance of endogenous constructs [89] and can handle non-normally distributed data [90]. Additionally, it is effective in estimating complex models with a small sample size while ensuring adequate statistical power.

The study strictly followed the two-step approach of PLS-SEM, inspecting the measurement model before examining the structural model [88]. Additionally, common method bias, mediation, and the predictive power of the model are reported [89]. Finally, model fit indicators and the coefficient of determination R² are presented.

In the first step, the measurement model is created, where the scores of latent variables and the loadings of indicators are estimated through iterative steps [63]. Furthermore, reliability and internal consistency indices Cronbach’s Alpha and Composite Reliability, as well as convergent validity AVE: Average Variance Extracted] and discriminant validity [Fornell-Larcker and HTMT: Heterotrait-Monotrait Ratio, are reviewed. In the second step, the bootstrapping technique is applied to evaluate the significance of the structural path coefficients of the model. Additionally, to determine the effect size or the magnitude of the impact of a predictor on a dependent variable within a structural model, the f² is reported.

Common method bias is a drawback to the extent that exogenous and endogenous variables are collected from the same respondent [91]. To ensure that CMB is not a severe concern in this study, we ensured that the questionnaire items had simple and concise language. In addition, we informed the participants that their anonymity was guaranteed and that there was no right or wrong answer to each statement they had to answer. Finally, we used the maximum likelihood method in Harman’s single factor test [92] using SPSS. The results reported that the common variance explained by the single factor is 48.58% below the threshold of 50% of the variance [93]. Hence, this type of bias is not a serious concern.

WarpPLS 8.0 was used for structural model analysis, which allowed us to evaluate the relationships between constructs and determine the validity and reliability of our model. WarpPLS 8.0 is a robust tool for structural equation modelling SEM, particularly useful for handling small samples and non-normal data, ensuring precise estimation of model coefficients.

4.1. Sample description

The general sample [N =  284] included mainly women [n =  204; 71.8%] and a smaller number of men [n =  80; 28.2%]. Most participants [n =  259; 91.2%] were between 18 and 25 years old. In the importance ranking, students exchanged messages with the chatbot on restaurants, tourist attractions, hospitality, and activities to do and visit.

4.2. Inspecting the measurement model

Six constructs were involved in the model that used partial least squares structural equations. The evaluation of the measurement model must meet both reliability and convergent and discriminant validity [94]. The indicator loading must be greater than 0.7 for the reliability to be considered acceptable [91]. The results in Table 1 show that all items are above the suggested threshold; Hence, the criterion is met. Cronbach’s alpha was used to measure the reliability of the constructs with a cut-off point of 0.7 [92] The values in this research range between 0.916 and 0.973, higher than what was proposed by Hair (see Table 1).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Indicator loadings, reliability and convergent validity.

https://doi.org/10.1371/journal.pone.0319463.t001

Two widely accepted types of validity must be established: convergent validity and discriminant validity [90]. Based on its description, the average variance extracted [AVE] [89] and composite reliability [CR] [93] were used to evaluate convergent validity. The results in Table 1 show that the values of each construct were higher than the cut-off point of 0.5 for AVE and 0.7 for CR; Hence, the assumption of convergent validity is met.

While for discriminant validity, the Fornell-Larker and heterotrait-monotrait [HMTM] criteria were used [95]. The HMTM ratio of correlations to be significant must be less than one [96]. Table 2 indicates that this criterion is met. Further, we used the square root of the AVE to measure discriminant validity [93]. In this sense, the square root of the AVE for each construct must be greater than the correlations between constructs [89]. Table 3 reports that the assumed discriminant validity is met because the square root of the AVE of each construct [in bold diagonal] is greater than the correlations between other constructs.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Result of discriminant validity [HTMT2].

https://doi.org/10.1371/journal.pone.0319463.t002

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 3. Result of discriminant validity [Fornell-Larcker].

https://doi.org/10.1371/journal.pone.0319463.t003

Finally, before analyzing the structural model, we evaluate the general model. The model’s fit can be assessed by applying the standardized root mean square residual [SRMR], which must be below 0.08 [97]. In our research, the value of SRMR is 0.052, which suggests a good fit for the measurement model.

4.2. Examining the structural model

Upon confirming a measurement model that meets the reliability and validity assumptions, the next step was to examine the structural model using the sign, size, and significance of the path coefficients [91] to statistically verify the theoretically established paths and validate the developed hypotheses. Additionally, to indicate the explanatory power of the model, R2. Values were calculated for the five endogenous [dependent] constructs. Table 4 and Fig 1 show the results of the structural model, including the path coefficients and R2.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 4. Result: Structural model.

https://doi.org/10.1371/journal.pone.0319463.t004

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. Structural model.

https://doi.org/10.1371/journal.pone.0319463.g001

The model fit indicators obtained from WarpPLS 8.0 demonstrate a robust model with strong predictive validity and reliability. The Average Path Coefficient (APC) is 0.309 (P <  0.001), and the Average R-squared (ARS) is 0.468 (P <  0.001), with an Average Adjusted R-squared (AARS) of 0.465 (P <  0.001). The Average Block VIF (AVIF) is 2.084, and the Average Full Collinearity VIF (AFVIF) is 2.343, both well within acceptable limits (<= 5, ideally <= 3.3). The Tenenhaus GoF (GoF) is 0.611, indicating a large effect size. Additionally, the Simpson’s Paradox Ratio (SPR) is 0.857, the R-squared Contribution Ratio (RSCR) is 0.994, the Statistical Suppression Ratio (SSR) is 1.000, and the Nonlinear Bivariate Causality Direction Ratio (NLBCDR) is 1.000, all meeting or exceeding their respective acceptable thresholds. These indicators collectively affirm the robustness and reliability of the structural model.

Concerning the influences of the different constructs in chatbot technology, first of all, the information quality influences the perceived usefulness of tourists [β=0.77; p˂0.001], user satisfaction [β=0.24; p ˂0.001] and perceived enjoyment [β=0.71; p˂0.001]. These results are statistically significant and favor H1, H2, and H3. Therefore, the first three hypotheses are corroborated. Secondly, the effect of perceived usefulness [β=0.27; p˂0.001] and perceived enjoyment on satisfaction [β=0.39; p˂0.001] has a positive impact; therefore, H4 and H6 are accepted. Likewise, the effect of PU [β=0.35; p˂0.001] and PE [β=0.27; p˂0.001] on the intention to continue using the chatbot tool are positive and significant, which corroborates H5 and H7. Also, we demonstrate that user satisfaction is an antecedent in the intention to continue using a chatbot [β=0.26; p˂0.001], corroborating H8. Finally, the intention to continue is an antecedent that will influence the tourist to visit a destination [β=0.57; p˂0.001]; therefore, H9 is supported (see Table 4).

5.1. Theoretical implications

This study’s theoretical and methodological contribution is to demonstrate the suitability of certain theoretical models [TAM, ECM, and ISSM] for research on chatbots of tourist destinations. Previous studies have applied TAM, ECM, and ISSM models to the study of various tourism technologies [35,40,41,47,48], but our study has shown that these models are very useful to analyze destination chatbots in the context of travel planning, to examine the crucial factors that predict a satisfactory experience, continued behavior in the use of technology and the tourist’s subsequent intention to visit.

Firstly, our research confirms that the informativeness of chatbots, characterized by the accuracy and relevance of the information they provide, is a key driver of user satisfaction, enjoyment, and perceived usefulness, in line with previous studies [65,72,73]. This aligns with the Technology Acceptance Model [TAM], which posits that the quality of information directly influences users’ perceptions of technology. Our results highlighted the importance of information quality [IQ] as a critical factor in creating positive user experiences with chatbots. The improved perception of usefulness and enjoyment due to high-quality information supports the theoretical constructs of TAM and highlights the role of informativeness in enhancing user engagement.

Secondly, the study reveals that perceived enjoyment and usefulness are significant psychological factors affecting users’ intention to continue using chatbots. This finding reinforces the theoretical framework of the Expectancy Confirmation Model (ECM), which suggests that perceived usefulness and enjoyment contribute to user satisfaction and continued use. Although prior research has focused on general technological tools, our study explicitly demonstrates how these constructs apply to chatbots, thereby expanding the theoretical application of ECM in travel planning.

Thirdly, our findings show that an engaging and enjoyable interaction with chatbots increases users’ likelihood of continued use, as previous studies have shown [74]. This outcome aligns with the Information Systems Success Model (ISSM), which emphasizes the importance of user satisfaction in predicting future use. The study extends the ISSM framework by demonstrating that perceived enjoyment significantly predicts continued use, even in specific contexts like travel planning. This underscores the value of creating positive user experiences to encourage sustained engagement with chatbots.

The other significant theoretical contribution is covering a gap in studies on the subject. Since the emergence of ChatGPT and the development of AI, technologies have increasingly been used in travel planning. But even though studies on chatbots have increased [17,18,98], there is practically no research on chatbots in travel planning and their influence on decisions to visit a tourist destination. This study contributes to filling this gap.

5.2. Managerial implications

The study analyzes how potential tourists perceive the conversational style of a chatbot in terms of quality of information, perceived usefulness, enjoyment of the tool, and their subsequent satisfaction and intention to continue using it. By exploring a diversity of cognitive and affective constructs, scholars and designers of these technologies can understand how tourists experience psychological changes and decide to select and visit a destination based on their experience interacting with the chatbot.

Thus, destinations and chatbot designers can improve the AI capabilities of this tool in three ways: First, chatbots must take care of and improve their contingency capacity so that the conversation is compelling, dynamic, and interesting, provides them with fair and quality information they need at every moment of trip planning. Secondly, the chatbot must be perceived as easy to use and very useful so that potential tourists continue to use it during planning.

Furthermore, improving the chatbot’s conversational quality to make interactions enjoyable and engaging can significantly impact user retention. A chatbot that combines effective information delivery with a pleasant user experience is more likely to foster trust and satisfaction among tourists. This approach can lead to increased usage and positive word-of-mouth, ultimately influencing travel decisions.

The strong relationship between the intention to continue using chatbots and the decision to visit a destination highlights the strategic importance of chatbots in the tourism industry. By leveraging chatbots as tools to facilitate travel planning, destinations can enhance their attractiveness and encourage visits. The ability of chatbots to provide personalized and engaging interactions can be a key differentiator in the competitive travel market.

Finally, the friendly, warm, empathetic, and emotional tone in conversation styles, which generate fun conversations, make chatbots emulate humans and establish emotional connections with users, facilitating communication. Consequently, taking advantage of the digital footprint of user interactions is crucial to offering personalized services based on extensive data analysis of conversations.

6.1. Limitations

One of the main limitations of this study is the homogeneity of the sample, which is predominantly composed of young university students under 25 years old, many of whom had prior experience using chatbots. This may limit the generalization of the findings to broader populations, particularly to older individuals or those who have not previously interacted with chatbots. Future studies should aim to include a more diverse range of participants, both in terms of age and experience, to assess whether the results hold across different demographic groups. It could also be applied to users with varying sociocultural characteristics, including those resistant to innovation or with technological asymmetries, in order to compare tourist behavior based on cultural background.

Another limitation relates to the functionality of the chatbot used in this study. While it offers several hotel options and integrates links to online travel agencies like TripAdvisor, it does not create personalized travel packages nor can it make reservations for events or tourist attractions. This limited functionality may have impacted user satisfaction and the intention to continue using the chatbot. More research is needed to evaluate the impact of more advanced chatbot features, such as personalized recommendations, on tourist behavior and satisfaction.

Indicators for constructs

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image