Customer intentions toward the adoption of WhatsApp chatbots for restaurant recommendations Open Access

Currently, chatbots are contributing to new forms of interaction between consumers and brands. In the tourism and hospitality sector, their implementation is equally palpable (Paul et al., 2023). Thus, hotels, travel agencies and restaurants offer very diverse customer services through these chatbots, such as responses to messages, quick access for reservations or purchases, takeout orders, customer service and recommendations (Leung and Wen, 2020; Yoon and Yu, 2022). A clear example of success in hospitality are the Ollie (Tripadvisor) and Fliggy and Ctrip chatbots, specialized in recommendations and reservations (Li et al., 2021).

The need for consumers to use a chatbot varies according to the context of the tourism service. This context is defined by the type of service (e.g. hotel, restaurant), the timing of the request (before, during and after the trip), the specific purpose of the interaction (e.g. information search, booking and purchase), and the type of decision inherent to the service (e.g. complex, impulsive and variety-seeking). All these factors influence the pace and perception of waiting time (Noone et al., 2009), consumer engagement (Santos et al., 2022) and the hedonic experience with the service (Maar et al., 2023). For instance, interactions with chatbots for accommodation or airlines are framed within complex shopping behaviors, primarily centered on the trip planning phase. In these cases, the information search is more comprehensive and anticipatory and the decision-making process is more deliberate and prolonged, culminating in booking and purchase (Zhu et al., 2023). In contrast, interactions with chatbots for restaurants tend to be more frequent and rapid, occurring primarily during the trip (Yoon and Yu, 2022), and they are closely aligned with the moment of consumption. This would be a variety-seeking purchase decision (Assael, 1987), characterized by lower personal involvement and risk, where the supply is perceived as differentiated. According to Assael (1987), in such purchases consumers often switch brands frequently without extensively evaluating their decisions, choosing to assess them during the consumption experience. However, the next time, the consumer might seek another alternative, either out of boredom or just to try something different.

This disparity in the use of restaurant chatbots, compared to other tourism services, also affects the factors driving their acceptance by consumers (Figure 1). Despite the significant impact that chatbots are having on restaurant service recommendations (Sano et al., 2018), the literature in this domain remains notably scarce. The limited research on the adoption of chatbots in tourism services has predominantly focused on the pretrip phase, addressing destination selection (Cai et al., 2022) and trip planning activities (Pillai and Sivathanu, 2020), such as their use within travel agencies or hotel booking (Zhu et al., 2023). For this context of complex shopping behavior (Assael, 1987) and prior to the visit, the literature identifies several antecedents for user acceptance, including understandability, reliability, assurance, interactivity, perceived usefulness or performance expectancy (PE), perceived intelligence and anthropomorphism (Lalicic and Weismayer, 2021; Lei et al., 2023; Melián-González et al., 2021). However, it remains unclear whether these factors are equally explanatory for the acceptance of chatbots in variety-seeking shopping situations (Assael, 1987), such as in the restaurant industry, given that they represent a different consumer context. Only a very small group of authors have recently examined the acceptance of chatbots in scenarios such as home ordering (Leung and Wen, 2021), menu selection (Yoon and Yu, 2022) and scheduling reservations (Maar et al., 2023). These studies suggest that aspects such as effort expectancy (EE) and consumption emotions in the online ordering experience (Leung and Wen, 2021), as well as PE (Yoon and Yu, 2022), are relevant for the acceptance of chatbot use in food service. Consequently, not only is there currently limited scientific evidence but also important scenarios, such as the acceptance of chatbots by consumers for restaurant selection through consultation of recommendations, have yet to be thoroughly investigated. This is particularly significant insofar as it represents one of the most frequently used services in interactions with chatbots by tourists.

Given that restaurant recommendation services provided by chatbots are among the most widely used in the sector, it is essential to consider the intention of their continued use as a key aspect of the study. This channel has become well known and increasingly commonplace for many tourism consumers (Maar et al., 2023; Majid et al., 2024). To date, few studies have addressed the use of chatbots in tourism services from this perspective, and the specific literature on restaurants primarily considers occasional use rather than recurrent use.

In response to the identified gaps in the literature, this research aims to identify the factors that drive the continued use of chatbots for information search and restaurant recommendation by tourists. This encompasses mixed search purchase behavior and usage close to the point of consumption.

In a marked change of perspective from previous studies, our research posits that the platform on which the chatbot operates is not a trivial issue, insofar as it conditions the waiting time, the manner in which information is presented and the accessibility and location of the consultation. These aspects are of the essence for tourists’ decision-making processes. However, there is no empirical evidence supporting the influence of the platform on the acceptance of chatbots. To date, the literature has focused more on aspects such as the anthropomorphism of chatbots (Jin and Youn, 2023) or it has conducted experiments with fictitious chatbots designed exclusively for research purposes, without considering the real context of use (Leung and Wen, 2021; Yoon and Yu, 2022), also ignoring the platform on which the chatbot is presented (Melián-González et al., 2021).

To fill this theoretical void, it is crucial to conduct research with samples of consumers using a real and functional chatbot on a given platform. Specifically, studies on the continued use of chatbots on mobile instant messaging (MIM) platforms are required, as these are, a priori, ideally suited to meet the information needs for immediate consumption and variety-seeking behavior due to their accessibility (familiarity with the tool, quick and easy use and consultation in various locations).

This brings us to our main research question:

In the acceptance of the use of chatbots, it is important to consider that hyper-connected “digital natives” appear to differ from other generations of customers, such as “digital immigrants” (Prensky, 2001). If this is true, one cannot suggest a homogeneous behavior by age in their intention to use restaurant services via chatbots (Maar et al., 2023). However, this behavioral heterogeneity has rarely been considered in the uptake of restaurant services, with the exception of the work by Maar et al. (2023), which focuses only on generations X and Z. Along with age, gender is another variable that the classic literature on technology adoption identifies as a moderator of technology uptake (Venkatesh et al., 2012), and which has also been highlighted in studies on the adoption of tourism services. However, in the case of tourism chatbots, its moderating effect has not been tested (Zhang et al., 2023).

Taking the above into account, the second research question is proposed:

To answer these research questions, we will use the unified theory of acceptance and use of technology 2 (UTAUT2) (Venkatesh et al., 2012). This theory is appropriate for two reasons. First, it is the most referenced, tested and validated in the literature for explaining the acceptance of new technologies by end consumers (Gupta and Dogra, 2017), evolving from the previous (Venkatesh et al., 2003) UTAUT (which was primarily oriented toward companies. Second, this theory encompasses factors noted in the limited literature on the acceptance of chatbot use in restaurant services: EE, emotions and PE (Leung and Wen, 2021; Yoon and Yu, 2022), and considers the need to moderate acceptance by age and gender. To confirm the utility of the model and provide robust explanations for the “intention to continue using” this technology, a sample of recurrent users of a chatbot service and MIM, specifically WhatsApp, were surveyed. Furthermore, to deepen our understanding of the main factors influencing this intention to use, this model is enriched with a necessary conditions analysis (NCA), allowing us to assess whether there are any bottlenecks in explaining intention to use (Dul, 2021).

Finally, this study offers four important contributions regarding restaurant recommendation chatbots via MIM:

• the suitability of the UTAUT2 model to explain users’ intention to continue using chatbot services in the context of an information search for an unplanned and varied purchase decision;

• the strength of hedonic motivation (HM) as a predictor of the intention of continued use;

• two predictors, EE and HM, are identified as necessary conditions to explain the behavioral intention (BI) to continue using this technology; and

• there is no significant difference in the acceptance behavior of this technology by age and gender for restaurant search services.

The development of this research approach represents an opportunity for restaurants implementing chatbots to better understand the factors that affect repeat customers’ intention to continue using them. This understanding will enable restaurants to enhance their interaction and loyalty strategies. This is particularly relevant in a context where consumer decisions are increasingly influenced by technology and where restaurants are seeking innovative ways to attract and retain customers.

As a result of the development of artificial intelligence, an alternative to communication with human interlocutors is proliferating in this sector: chatbots. In this field, the existing literature has paid attention to technological aspects (Adamopoulou and Moussiades, 2020), satisfaction with their use (Jiménez-Barreto et al., 2021), attitude toward their use (Maar et al., 2023) and key factors for adoption (Jha et al., 2023). In particular, tourism literature has mainly explored the intention to use chatbots, with much less emphasis on the continuity of their use.

For the study of the reasons that explain the adoption of chatbots in the tourism sector ( Appendix 1), some authors have focused on the anthropomorphic characteristics of chatbots (e.g. Cai et al., 2022 or Pillai and Sivathanu, 2020). Other frequently mentioned factors include perceived usefulness or PE (Majid et al., 2024; McLean et al., 2020; Yoon and Yu, 2022), functionality (Jha et al., 2023; Lalicic and Weismayer, 2021), interactivity (Zhu et al., 2023), perceived intelligence (e.g. Pillai and Sivathanu, 2020) and HM in their use (e.g. Jha et al., 2023). Other less frequent variables include habit of using chatbots (HT), social influences (SI) (Melián-González et al., 2021), trust and ease of use (FC) (Pillai and Sivathanu, 2020) and EE and accessibility (Majid et al., 2024). In the restaurant domain, studies on the causes of intention to use are almost nonexistent. Thus, Yoon and Yu (2022) identify four variables (perceived usefulness, functionality, value and attitude) that influence the intention to use ordering services. On the other hand, Leung and Wen (2020, 2021), approach this intention to use from a reflexive and noncausal model, differentiating the ordering methods (chatbot, mobile and online).

It has been confirmed that the intention of continuity of use with the chatbot for tourism services is associated with customer satisfaction (Dhiman and Jamwal, 2023; Pereira et al., 2022), perceived usefulness or PE (Lei et al., 2023; Zhang et al., 2023), customer satisfaction (Li et al., 2021; Pereira et al., 2022), perceived usefulness or PE (Dhiman and Jamwal, 2023; Zhang et al., 2023). However, each of these studies incorporates other variables, such as predisposition to use self-service technologies, comprehensibility, reliability, security and interactivity (Li et al., 2021); brand attachment (Pereira et al., 2022); ease of use (Lei et al., 2023; Pereira et al., 2022); perceived usefulness (Lei et al., 2023); perceived ease of use, task attractiveness and social attractiveness (Lei et al., 2023); social presence and image processing (Jin and Youn, 2023); and PE, SI, HT, anthropomorphism and personalization (Zhang et al., 2023).

The current literature reveals the heterogeneity of tourist behavior in the acceptance of chatbot use in tourism. Consequently, many studies incorporate moderating variables related to consumption characteristics (e.g. Maar et al., 2023; Zhu et al., 2023), whereas traditional sociodemographic variables are less frequently considered. Only recent work by Maar et al. (2023) and Zhang et al. (2023) has analyzed the influence of age and gender, respectively. It should be noted that the intention to use chatbots is stronger for Generation X than for Generation Z (Maar et al., 2023) and that there are no significant differences in most of the factors influencing the continued intention to use chatbots between men and women (Zhang et al., 2023).

On the other hand, studies on the adoption of chatbots have primarily focused on the following tourism services: travel planning (e.g. Dhiman and Jamwal, 2023), destination activities (Melián-González et al., 2021) and hotel reservations (Jin and Youn, 2023). However, even though initial evidence is found in the context of hotels, travel and tourist destinations, there is a scarcity of research in one of of the most demanded areas in the hospitality industry, namely, restaurants. Only a very limited number of authors, such as Leung and Wen (2020, 2021) and Yoon and Yu (2022), mention service interactions with consumers focusing on take-away orders, while ignoring significant services such as recommendations for restaurant selection. Furthermore, these studies focus solely on the intention to use chatbots rather than on their continued use.

MIM allows synchronous and symmetrical communication, enhancing the personalization of the tourist experience (Buhalis and Amaranggana, 2015). This enables travelers to communicate their contextual needs ubiquitously (Lamsfus et al., 2014). Recent literature highlights relevant contributions regarding customers’ intention to use it in tourism (Lei et al., 2020). The findings indicate that age and perceived usefulness are predictors of continued use intention. Additionally, MIM facilitates value cocreation activities between company and customers (Lei et al., 2020). It is also used by customers before and/or after a stay, primarily for nonurgent issues (Lei et al., 2021). Furthermore, while the literature acknowledges that WhatsApp is among the preferred channels for establishing communications and transactions between employees and customers in the tourism industry (Francis and Jilo, 2021), there are no studies explaining the interest in its adoption.

Although the literature indicates that chatbots and MIM are commonly used communication methods in the hospitality and tourism industry, three key issues arise. First, Lei et al. (2023) reveal that users’ perceptions of recurrent use intention differ between these communication forms. For chatbots, reuse intention is driven by ease of use and perceived usefulness. In contrast, for MIM, while ease of use is significant, social aspects (task attraction and social attraction) are more decisive. Second, despite the importance of these communication methods separately, there are no studies on the joint intention to use both technologies such as chatbots via MIM, even though they exist in business contexts and the hospitality and tourism industry. Examples include chatbots via WhatsApp like “Luzia” (www.luzia.com/), “Carina” (https://carina.chat/) or “Ask Vicente” (www.elmundo.es/f5/descubre/2018/04/21/5ad9fcef268e3ef4088b45f2.html) and the chatbot via Facebook Messenger “Victoria la Malagueña” (www.facebook.com/malagabots/). Examples show that restaurant recommendations are among the most demanded chatbot services via MIM in the sector.

On the other hand, tourist literature that addresses factors affecting the recurrent use intention of chatbots or compares MIM and chatbots is mostly experimental rather than based on real use (Lei et al., 2023; Leung and Wen, 2020).

To discover the factors that motivate frequent users of a restaurant recommendation chatbot via WhatsApp to continue using it, we used the UTAUT2 technology adoption model (Venkatesh et al., 2012).

BI is described as the level at which an individual has consciously made plans to use or not use a specific technology in the future (Venkatesh et al., 2003). Previous research highlights it as the strongest and most immediate predictor of individual behavior (Ajzen, 1991; Davis et al., 1989). Our study aims to understand how BI manifests itself in the context of continued use of our restaurant recommendation chatbot via WhatsApp (Figure 1).

EE is defined as the evaluation of the effort required to complete a task using a specific technology (Venkatesh et al., 2003, 2012). This factor is grounded on Davis et al. (1989) perceived ease of use and has been generally validated to be a strong predictor of BI (Chopdar et al., 2018).

Previous tourism studies show a debate on the influence of EE on the intention to use technology. Some authors assert the presence of a positive relationship (Baydeni̇z et al., 2024), while others find it nonsignificant (Gupta et al., 2018). For instance, Majid et al. (2024) evaluated the relevance of this factor in tourism chatbots. However, no work has specifically addressed its impact on recurrent use, particularly in the field of restaurant services.

As a result, our hypothesis is that:

Facilitating conditions encompass users’ perception of the level of operational and technological support provided by the systems (Venkatesh et al., 2003). The literature indicates that FC significantly impact both the intention to use and user behavior across a wide range of technologies (Macedo, 2017).

However, similar to what occurs with EE, the relationship between FC and technology use intention in the tourism context is not always positive. While most studies, such as those by Jeon et al. (2019) and Chaw et al. (2023), report a positive relationship, some authors question this link (Wu and Lai, 2021). Regarding tourism chatbots, only Pillai and Sivathanu (2020) highlight the relevance of FC for usage intention, though their study focuses solely on travel agency and hotel services.

Therefore, we propose the following hypothesis:

Venkatesh et al. (2012) assert that habit is the result of previous experiences, as well as how those experiences can motivate the use of new technologies (Ajzen, 1991). Subsequent studies have demonstrated the influence of HT on BI toward adopting new technologies (e.g. Wu and Kuo, 2008). In the tourism sector, existing studies on the adoption of smartphone apps by tourists have also highlighted the significant effects of HT on BI and actual usage (Escobar-Rodríguez and Carvajal, 2014; Gupta and Dogra, 2017).

In the context of chatbot literature in tourism, HT appears less frequently compared to other variables and is typically examined in services other than restaurants. Specifically, studies by Melián-González et al. (2021) and Zhang et al. (2023) identify a positive relationship between HT and the intention to use and recurrent use, respectively, in the broader context of travel and tourism.

We postulate the following hypothesis:

HM has been defined by Venkatesh et al. (2003) as the pleasure a user derives from using technology. Researchers consistently indicate that intrinsic motivation is crucial for technology use and it determines users’ continued engagement (Gupta, 2018). Studies also highlight the rewarding experience associated with the use of technology, such as the generation of positive feelings (Chung et al., 2018) and enjoyment during interaction (Gupta, 2018). This driver, as noted by Ashfaq et al. (2020) and Aslam et al. (2022), significantly influences the intention to use these services. Similarly, in tourism related literature, it is a relevant factor in explaining usage intention (Gupta and Dogra, 2017).

Regarding chatbots in tourism, the evidence suggests that HM positively affects users’ intention to use chatbot services, although specific usage and service contexts are not extensively covered (Jha et al., 2023; Melián-González et al., 2021).

Therefore, the resulting hypothesis is:

PE refers to the extent to which technology usage aids users in performing their tasks (Venkatesh et al., 2003). It has consistently been validated as a robust predictor of BI (Macedo, 2017).

In the domain of tourism, the link between PE and BI has been well-documented, particularly in the context of e-commerce websites for both product purchases and tourism services bookings (Chung et al., 2018; Escobar-Rodríguez and Carvajal, 2014). In the context of adopting chatbots for tourism services, there are still few recent studies that explore this positive relationship. This despite the fact that it is the most well-established variable for both examining usage intention (Majid et al., 2024; Yoon and Yu, 2022) and studying continued use (Dhiman and Jamwal, 2023; Lei et al., 2023; Zhang et al., 2023).

We formulate the following hypothesis:

Venkatesh et al. (2012) argue that price value rises as consumers seek greater perceived benefits compared to monetary sacrifice. This has been consistently documented in the literature on technology adoption (Chopdar et al., 2018). In tourism, several authors assert the positive influence of this predictor on the intention to use tourism technologies, such as augmented reality apps for gastronomy (Calderón-Fajardo et al., 2023) and mobile applications such as Airbnb for booking accommodations (Nathan et al., 2020).

In the context of tourism chatbots, this factor has been identified as relevant, particularly for restaurants, in influencing the intention to use and place orders (Yoon and Yu, 2022). Because of this, we formulate the following hypothesis:

SI is the degree to which an individual perceives that important people (e.g. friends and family) believe they should use an innovative technology (Venkatesh et al., 2003), often predicting acceptance. However, some studies, for instance, Ozturk et al. (2021) show that SI is not significant for mobile app usage intentions. Conversely others, such as Kuberkar and Singhal (2020), do find a significant relationship.

In tourism, Melián-González et al. (2021) and Zhang et al. (2023) claim that SI explains the intention to use chatbots and recurrent usage, though not specifically in restaurants. Due to the limited evidence, we propose the following hypothesis:

The UTAUT2 model (Venkatesh et al., 2012) includes three moderator variables: age, gender and experience. Since our study focuses on the recurrent use intention of a real chatbot, the experience variable is irrelevant because users already have prior experience. Therefore, only age and gender are considered. These variables are also noted in the technology acceptance literature in tourism, though not specifically in restaurants. Maar et al. (2023) show that Generation X has stronger usage intentions than Generation Z, while Zhang et al. (2023) find that gender differences do not always explain continued use intentions.

To measure the intention of recurrent use of a restaurant recommendation chatbot service and address the methodological weakness of previous studies with nonrepresentative samples (Leung and Wen, 2020), a sample was drawn from consumers using a real and fully functional chatbot via MIM. The chosen chatbot was AskVicente, a free service available on WhatsApp MIM. The reasons for selecting this chatbot were:

• It is the first gastronomic chatbot in Spain aiding in restaurant recommendations through questions and answers (Bermejo, 2018);

• It had surpassed 30,000 unique users within five months of its launch, indicating a consistent user base; and

• It operates on WhatsApp, the world’s most popular MIM application with two billion active monthly users (Statista, 2024).

A sample of the 537 most frequent users, defined as those with an average usage of twice a month over the past year, was selected. A questionnaire, pretested among three researchers and 30 users, was administered to these 537 individuals. A total of 400 responses were received, 14 of which were discarded due to incomplete data. Thus, 386 valid questionnaires were analyzed, yielding a sample error of 95%. The age distribution was as follows: 27.72% aged 18–25; 31.08% aged 26–35; 18.39% aged 36–45 and 22.79% over 46. Gender distribution was 52.84% female and 47.15% male. Notably, 25.5% of regular consumers did not respond. Nonresponse is a type of nonsampling error, indicating an expected margin of uncertainty (Azorín and Sánchez-Crespo, 1986). Despite the difficulty in addressing this error, our research confirmed that there was no behavioral or sociodemographic bias among nonrespondents. The data collection process was conducted in two waves through the AskVicente chatbot, inviting active users to participate voluntarily to minimize nonresponse.

We utilized partial least squares structural equation modeling (PLS-SEM) to test the proposed hypotheses and NCA to identify potential bottlenecks among the independent variables (Dul, 2021). While other statistical methods could have been applied, PLS-SEM was selected due to its provision of latent variable scores essential for subsequent NCA analysis (Sarstedt et al., 2022). Both PLS-SEM and NCA analysis were conducted using SmartPLS 4.0 (Ringle et al., 2024).

To address measurement bias and following (2015) recommendations, we incorporated the latent variable of common method bias as a dependent variable with specific indicators. This allowed us to evaluate all variance inflation factors, with values consistently below 3.3, confirming the absence of such bias in the model.

To address the hypotheses, we first estimated the measurement model using the PLS-SEM statistical procedure to assess the reliability and validity of the measurement scales. Subsequently, we estimated the structural model to verify the study’s hypotheses and evaluated the overall adequacy of the model, abiding by the criteria established by Hu and Bentler (1999).

Following the guidelines of Roldán and Sánchez-Franco (2012) and Henseler et al. (2015), the loadings of each construct must exceed the threshold of 0.7 to assess the reliability and validity of the measurement model. The results (see  Appendix 2) meet this criterion, demonstrating the discriminant validity of the scales. The model exhibits strong reliability in both constructs and indicators, as well as convergent and discriminant validity, ensuring that the factors are statistically distinct and suitable for evaluating the structural model.

The reliability of the factors was assessed using composite reliability indicators and Cronbach’s alpha. All indicators exceeded the 0.7 threshold proposed by Nunnally (1978). Convergent validity was confirmed by analyzing the average variance extracted (AVE), where all values exceeded 0.5, in accordance with Straub et al. (2004) (Table 1).

To confirm discriminant validity, the heterotrait-monotrait ratio was used, ensuring all results were below 0.9 (Henseler et al., 2015), as shown in Table 2. Additionally, the correlation between variables was calculated, with none yielding significant results.

To assess model fit, we used the standardized square root residual criterion. The obtained value was 0.0624, which is below the 0.08 threshold proposed by Henseler et al. (2015).

Additionally, to ensure a good model fit, we assessed endogeneity. Following the process outlined by Hult et al. (2018) and applying the Cramér-von Mises test, we found that the test conditions were met and that the latent variables are not normally distributed. Subsequently, the Gaussian copulas test was applied to all relationships between independent and dependent variables. All copulas were nonsignificant, indicating that the model lacks endogeneity.

Furthermore, the model explains 56% (R²) of the variance in BI (R²⁾ (Table 3), exceeding the minimum threshold of 0.10 suggested by Falk and Miller (1992).

To evaluate the structural model, we applied a bootstrapping technique with 10,000 subsamples to assess the reliability of the hypothesized relationships. The hypotheses that were rejected (Table 3) involved SI, PE and FC. In contrast, hypotheses related to EE, HT, HM and price value (PV) were accepted, demonstrating a significant influence on the intention to continue using a WhatsApp chatbot for restaurant recommendations.

Regarding the moderator variables (age and gender), the analysis results indicate that they are not significant in predicting the intention for recurrent use. Although age and gender influence FC, PE and HT, they do not moderate the relationship in the context of continued use intention.

Furthermore, the model’s predictive power was assessed using Stone-Geisser Q² values. The obtained Q² values (Table 4) are greater than zero, indicating that the model has predictive capability. This finding supports the model’s ability to make meaningful predictions based on the data.

To assess the results, an importance-performance map analysis (IPMA) was conducted, enabling a detailed examination at the variable and indicator levels based on the model outcomes (Ringle and Sarstedt, 2016). This analysis gauges the importance and performance of constructs and indicators that influence a particular construct. The target construct for this analysis was BI. The IPMA of the model revealed the outcomes depicted in Figure 2.

NCA uses the necessity effect size (d) and its significance to determine whether a variable is a necessary condition. This effect size is calculated by dividing the area without observations by the total area, yielding a range of 0 ≤d ≤ 1. Effect sizes are categorized as small (0 < d < 0.1), medium (0.1 ≤ d < 0.3), large (0.3 ≤ d < 0.5) and very large (d ≥ 0.5) (Dul, 2016). A common threshold for necessity hypotheses is d = 0.1. Statistical significance, assessed through NCA’s permutation test (e.g. p < 0.05), is essential for validating a necessity hypothesis in addition to the practical significance of the effect size. Results for this test are presented in Table 5, showing that EE and HM have minimal and significant effects.

The primary aim of this study was to identify the key determinants of continued usage intention among recurrent users of a restaurant recommendation chatbot service via MIM. This research validates the UTAUT2 model as a relevant theoretical framework within the context of hospitality and tourism services, aligning with its application in other purchase behavior (Assael, 1987) and technology contexts (Gentner et al., 2020). Specifically, it examines a search purchase decision made by tourists during their travel experience.

Our findings confirm that EE, HT, HM and PV positively influence consumers’ intentions to use chatbots. These results are consistent with previous studies highlighting the significant impact of these factors (Aslam et al., 2022; Chung et al., 2018; Baydeni̇z et al., 2024; Jha et al., 2023; Melián-González et al., 2021; Yoon and Yu, 2022). Despite the prevalence of free chatbots, regular users consider PV important due to the premium services offered, indicating a trend toward paid versions that provide more comprehensive services (Fernández, 2023). Notably, HM and, at lesser extent, HT emerged as crucial variables for explaining the intention to continue using chatbots via MIM. Additionally, EE and HM are identified as bottlenecks in explaining recurrent usage intention. The model’s predictive capability is confirmed by a Q² value greater than 0 and a good fit.

Our analysis shows that FC, SI and PE did not significantly impact the continuance intentions in these habitual consumers. FC’s lack of significance may be attributed to the intrinsic ease of use of smartphones that support this technology, as noted by Ho et al. (2021). The familiarity and simplicity of WhatsApp chatbots render FC less relevant for regular users. On the other hand, although previous research identified significant relationships between SI and AI chatbot usage (Kuberkar and Singhal, 2020), as well as in tourism services (Melián-González et al., 2021; Zhang et al., 2023), SI does not appear to predict the continued use of restaurant recommendation chatbots. This discrepancy in our study may be due to the specific context and population of recurrent users who are less influenced by external factors and are familiar with WhatsApp interface. Additionally, while PE is a well-established factor in the continued use of chatbots for travel planning (Lei et al., 2023; Zhang et al., 2023), it is not significant for ongoing use of chatbots for variety-seeking purchases, such as restaurant recommendations, during and before travel.

On the other hand, the adoption of chatbots is influenced by user profiles, as highlighted by the UTAUT2 model, which considers demographic variables such as age and gender. Our results show no significant variation in continued use intention by age or gender, contradicting recent studies addressing these variables (Zhang et al., 2023; Maar et al., 2023).

In summary, answering our research questions, the key factors that influence tourists’ intention to continue using a chatbot for restaurant selection vary by technology and usage stage. For tourism chatbots, particularly restaurant recommendation chatbots via MIM, HM, HT and EE positively influence continued use, highlighting their integration into daily routines and enjoyable interactions. Additionally, EE and HM are necessary conditions to correctly explain the recurrent use intention.

This study contributes to expanding knowledge in the field of tourist technology adoption, particularly by addressing the gap in the literature regarding the differential treatment required for restaurant services via chatbots compared to other tourism services. Specifically, it focuses on the most demanded and used service by tourists: seeking restaurant recommendations. Restaurants require differentiated attention from researchers due to the contextual differences in which they operate, primarily the type of purchase involved and the timing of its use. Our work provides four contributions related to the context of tourists’ behavior concerning the continued adoption of these chatbot services. First, it validates the suitability of the UTAUT2 model to explain users’ intention to continue using chatbot services in the context of information searches for unplanned purchase decisions and variety-seeking behavior. This context is characterized by proximity to the consumption moment and a lower consumer risk perception, such as in restaurant recommendations via MIM. This contribution addresses a gap in the literature that has thus far focused primarily on the use of chatbots in complex purchase decision contexts, particularly during the travel planning phase. Such contexts involve more extensive and advanced information searches, leading to a more deliberate and prolonged decision-making process, which culminates in reservations and purchases. Second, similar to other contexts (Wu and Kuo, 2008), this study confirms the predictive power of the model, showing that, in the restaurant services industry, chatbots have the potential to become integrated into users’ routines. Third, being perhaps the most significant contribution, this research identifies two key predictors – EE and HM – that are necessary to explain the BI to continue using chatbots via MIM. These predictors are critical in understanding BI. Finally, unlike other UTAUT2 models applied to technology adoption in tourism services, the variables of gender and age do not moderate recurring usage behaviors for restaurant selection services.

The results of this research provide valuable guidance for both mobile app developers to develop more attractive apps for customers and for restaurants to acquire more customers and serve them better. In this regard, this work focuses on four key factors to ensure tourists’ continued intention to use chatbots for obtaining recommendations. First, this study proves to developers that they must first minimize the effort required from users, which is achieved by ensuring a high level of familiarity with the platform. Second, they should enhance the enjoyment, social empathy and overall pleasantness of the chatbot experience. HT is just as important as HM in explaining BI but has not performed as well in explaining it, so there is a lot of work to be done in encouraging the habit to achieve this improvement and could help integrate chatbot use into users’ routines, not only before but especially during the tourist’s trip. In summary, our findings can help companies develop more effective design strategies by creating more intuitive chatbots, without the need to differentiate their services by gender or age at this stage, to foster greater interactivity and habituation in user consultations. As acceptance of this technology for recurring use increases with respect to the aforementioned factors (EE, HM and HT), developers or companies might even consider offering premium “paid” versions of recommendation services. This consideration is supported by the study’s findings, which reveal that regular consumers view the PV ratio of restaurant recommendation services as a positive factor influencing their intention to continue using the service.

Considering the previous practical implications and the increasing mediation of consumer decisions by technology, society will gain a new way to enhance confidence in the use of these chatbots via MIM in information search processes for unplanned purchase decisions and variety-seeking behavior, closer to the consumption moment and with lower consumer risk perception. In this regard, considering these four adoption factors – EE, HM, HT and PV – could be essential for enhancing the adoption of chatbots via MIM in the interactions between various industries and their consumers, extending beyond the hospitality and restaurant sectors.

Although this work on chatbot via WhatsApp is a novelty, the focus on a specific MIM channel limits the generalization of its results to other platforms. In this regard, replicating the outcome of the study on different platforms could provide valuable benchmarking across various channels.

Also, the selected sample was limited to recurrent consumers of a specific chatbot via WhatsApp. Expanding the study population to include regular users of other similar chatbots offering restaurant recommendation services could provide a broader understanding of this sector and service across various MIM platforms. Additionally, other moderating factors could be tested.

Finally, this research was quantitative and lacked the nuanced insights that qualitative inquiries can provide. Future research could benefit from mixed methods approaches.

This work was possible thanks to open-access funding provided by the Universidad de Málaga/CBUA.

Table A1 

Table A2