Extending UTAUT2 to explore intention to use ChatGPT for travel planning: a hybrid PLS-ANN approach Open Access

Technology advancement has an undeniable impact on every aspect of human life, including travel planning. The rapid rise of “Artificial Intelligence” (AI) and “Machine Learning” (ML) presents a transformative shift in technology expected to redefine and enhance hospitality and tourism technology solutions (Ivanov and Soliman, 2023). ChatGPT is an AI-powered chatbot developed by OpenAI that leverages Natural Language Processing (NLP) to create responses that mimic human conversation. For travelers, this means they can interact with a system that understands their queries, provides detailed responses, and even offers personalized recommendations related to their journey. This includes insights into destinations, notable attractions, events, and various travel services. With ChatGPT, one can find suggestions for unique local experiences or find hotel and restaurant recommendations in real time (Mich and Garigliano, 2023). Furthermore, ChatGPT can offer guidance on cultural nuances, visa prerequisites, and general travel advice, adding significant value to the overall travel experience. It is not just about providing information–this AI tool can also assist travelers in making bookings and reservations, sharing pertinent details such as pricing and availability to streamline the process (Gursoy et al., 2023). Beyond just assistance, the true power of ChatGPT lies in its ability to offer personalized experiences (Carvalho and Ivanov, 2024).

The benefits of ChatGPT in the tourism sector are substantial for customers. It enhances the travel planning process by providing real-time, accurate, and comprehensive information (Wong et al., 2023). By allowing users to quickly find answers to their travel-related questions, ChatGPT saves time and eliminates the need to search through multiple websites or contact various service providers (Gursoy et al., 2023). Furthermore, with ChatGPT’s personalization capabilities, travelers are guaranteed recommendations that are specifically tailored to their preferences and past behaviors (Demir and Demir, 2023). This ultimately leads to more enjoyable and fulfilling travel experiences. Furthermore, ChatGPT offers convenience by providing assistance with booking services, suggesting itineraries, and providing updates on travel conditions or changes (Carvalho and Ivanov, 2024).

Also, service providers in the tourism industry, such as hotels and restaurants, can greatly benefit from integrating ChatGPT. Hotels, for instance, can rely on ChatGPT to handle routine inquiries, which allows the staff to dedicate their time to more complex tasks (Dwivedi et al., 2023). Moreover, it can assist with managing bookings, provide guests with information about hotel amenities, and even offer personalized recommendations for local attractions, thereby enhancing the overall guest experience (Wong et al., 2023). In the case of restaurants, ChatGPT can be used to manage reservations, suggest menu items based on customer preferences, and provide details about special promotions or events (Carvalho and Ivanov, 2024). Besides, ChatGPT has the capability to collect and analyze customer feedback. This valuable information can be leveraged by service providers to enhance their offerings (Haleem et al., 2022). By using ChatGPT, service providers can achieve increased operational efficiency, higher customer satisfaction, and ultimately, greater loyalty and revenue (Kumar et al., 2024).

Despite being a relatively new technology, ChatGPT has rapidly gained the attention of scholars within the hospitality and tourism sectors due to its significant potential (Carvalho and Ivanov, 2024). Researchers have acknowledged its ability to revolutionize various aspects of these industries, including customer service, personalization of experiences, and operational efficiency (Iskender, 2023). The rapid adoption of ChatGPT stimulates academic interest in exploring its implications, challenges, and potential benefits. For example, Gursoy et al. (2023) and Carvalho and Ivanov (2024) investigated the risks and benefits associated with ChatGPT in the hospitality and tourism industry. However, despite early contributions, there is a scarcity of empirical studies investigating the factors contributing to ChatGPT adoption by travelers/tourists for travel-related purposes (Gursoy et al., 2023). Most existing studies focus on the technological capabilities and potential applications of ChatGPT, but they do not explore the behavioral drivers that influence its adoption in the context of travel planning (Iskender, 2023; Mich and Garigliano, 2023). Additionally, there is limited understanding of how personal characteristics such as innovativeness and risk aversion affect the adoption of ChatGPT, especially in the tourism sector. These gaps highlight the need for a comprehensive study that examines both the direct and moderating effects of these factors on the intention to use ChatGPT for travel planning. To address this research deficiency, the study employs the “extended unified theory of acceptance and use of technology” (UTAUT2) to investigate the influential drivers of travelers' adoption of ChatGPT. UTAUT2 is recognized for its high explanatory power in explaining technology adoption and usage (Beh et al., 2021; Foroughi et al., 2023). This framework is particularly appropriate for our study because it incorporates additional constructs such as hedonic motivation, price value, and habit, which are relevant in the context of AI-driven applications like ChatGPT. Moreover, UTAUT2 allows for the inclusion of personal characteristics such as innovativeness and risk aversion, which can significantly influence technology adoption behaviors in the dynamic and personalized context of travel planning. By incorporating these factors, UTAUT2 provides a comprehensive model that captures both the utilitarian and hedonic aspects of technology use. This makes it well-suited to explore the adoption of ChatGPT in the tourism industry.

From the methodological lenses, the present research combines the findings derived from two distinct methods, namely, “partial least squares” (PLS) and “artificial neural network” (ANN). The SEM-ANN technique is suitable for this study due to its ability to integrate the strengths of both PLS-SEM and ANN. PLS-SEM is an effective method for testing theoretical frameworks and gaining a deeper understanding of the relationships between constructs. This makes it particularly suitable for hypothesis testing (Guenther et al., 2023). However, it often falls short of capturing the complexities and non-linear relationships inherent in human behavior and technology adoption (Yin et al., 2023). ANN, on the other hand, excels in handling non-linear patterns and provides robust predictive power (Dadhich and Hiran, 2022). By combining PLS-SEM and ANN, we can leverage the explanatory power of PLS-SEM and the predictive accuracy of ANN to obtain a comprehensive understanding of the factors that influence ChatGPT adoption. Previous studies in the tourism context that have examined the adoption of ChatGPT have mostly used PLS-SEM (e.g. Ali et al., 2023; Xu et al., 2024). This is because PLS-SEM is known for its capability to handle complex models and non-normal data (Hair et al., 2019). However, there is an increasing recognition of the necessity for techniques that can capture non-linear relationships (Richter and Tudoran, 2024). Consequently, recent research has seen the adoption of ANN in combination with PLS-SEM (e.g. Giovanis et al., 2022; Mishra et al., 2023). This hybrid approach provides a more nuanced understanding of the factors influencing technology adoption by addressing the limitations of using either method alone. Inspired by this rationale, the present study sets out to accomplish the following objectives:

• (1)

  To investigate the influential drivers of using ChatGPT for travel planning.

• (2)

  To extend the UTAUT2 framework in the ChatGPT context by assessing the direct and moderating effects of personal innovativeness and risk aversion.

• (3)

  To explore and rank the influential drivers of travelers’ ChatGPT usage from both linear and non-linear perspectives using a hybrid PLS-ANN approach.

This research expands the current understanding of how ChatGPT is applied within the hospitality and tourism sectors. To do so, the study extends the UTAUT2 model by incorporating personal innovativeness and risk aversion. The findings from this study deepen our knowledge of the vital factors that influence and promote the adoption of ChatGPT by travelers. The study adopts a hybrid technique, integrating both linear (e.g. PLS) and non-linear (e.g. ANN) methods to study the desired outcome. The findings extend the knowledge by underscoring key drivers of ChatGPT adoption and ranking their relative importance. This study provides insightful implications for technology developers, practitioners, and various stakeholders in the hospitality and tourism industry about key drivers of ChatGPT adoption, supporting them in formulating effective ChatGPT-centric strategies.

ChatGPT, developed by OpenAI, has caused a sensation worldwide. This chatbot, powered by AI, was made available to the public in November 2022 and trained on human-written internet data, including conversations. By integrating language models and deep learning based on the “Generative Pre-training Transformer” (GPT) architecture, ChatGPT has significantly expanded the capabilities of chatbots. ChatGPT relies on a mixture of unsupervised pre-training and supervised fine-tuning to produce responses that mimic human-like reactions to queries and deliver expert-like insights on topics. According to Calvaresi et al. (2021), the utilization of chatbots in the tourism and hospitality sector has transformed how tourists interact with service providers. Ukpabi et al. (2019) asserted that AI-powered chatbots enable real-time communication with hotel guests via text messaging. Calvaresi et al. (2021) explained that chatbots can assist tourists in obtaining information about their intended activities. Chatbots can efficiently understand and respond to customer requests through the use of NLP. The use of chatbots in the travel planning process has gained widespread popularity in recent years. The advancement of NLP and ML has resulted in more intelligent and interactive chatbots.

ChatGPT is one of the most promising chatbots. ChatGPT has the potential to create travel and tourism suggestions tailored to a user’s interests and preferences by drawing on a vast pool of travel and tourism data (Iskender, 2023). The model can suggest personalized options for destinations, accommodations, and activities by training on a comprehensive set of travel-related information. ChatGPT has the potential to enhance customer service by generating individualized replies to customer queries. ChatGPT can facilitate trip planning and enrich travel experiences (Dwivedi et al., 2023). Buhalis and Moldavska (2022) suggested that ChatGPT can be combined with voice assistants and contextual real-time services to provide clients with groundbreaking smart customer service. ChatGPT enables tour and activity providers to analyze customer feedback and preferences, allowing them to determine the most popular tours and activities (Carvalho and Ivanov, 2024). As a result of this information, existing offerings can be improved, and new services can be developed.

Venkatesh et al. (2003) developed the UTAUT model, which has become a widely used framework for comprehending users’ intentions and actual use of technology. The UTAUT considers four primary constructs that affect the adoption and usage of technology by users, which are performance expectancy, effort expectancy, social influence, and facilitating conditions. Lu et al. (2009) reported that compared to other models in predicting behavioral intention (BI) towards information systems, UTAUT exhibits superior explanatory capability. Although UTAUT has been widely employed to comprehend user acceptance and adoption of technology, it has some limitations. UTAUT mainly concentrates on individual factors that influence technology adoption and does not take into account the influence of social and cultural factors on adoption (Tamilmani et al., 2021). In order to address the constraints in UTAUT, the UTAUT2 model incorporates three new constructs, namely habit, price value, and hedonic motivation (Ain et al., 2016). This study adopted UTAUT2 to assess the influential drivers of using ChatGPT for travel planning. While UTAUT2 holds considerable relevance, Venkatesh et al. (2012) emphasized that contextual factors should be incorporated into UTAUT2 as the factors influencing the acceptance of novel information systems can vary based on distinct situational conditions and contexts. This study incorporated personal innovativeness and risk aversion (users’ characteristics) into the original UTAUT2 model as independent variables and moderators. Prior studies have acknowledged the role of personal innovativeness (Yang et al., 2022) and risk aversion (Prince and Kim, 2021) in determining travelers’ decisions to adopt new technology. Personal innovativeness reflects the user’s openness to change, adaptability, and willingness to try out novel solutions which can significantly influence their acceptance of new technologies such as ChatGPT (Singh, 2022). Innovative individuals typically demonstrate greater adaptability towards adopting novel technologies (e.g. ChatGPT) and may be less encouraged by other traditional factors such as ease of use or perceived performance in their decisions, indicating its moderating role (Tran Xuan et al., 2023). Moreover, people who are more risk-averse are less likely to use new technologies (e.g. ChatGPT) even if they believe they could be beneficial (Niu et al., 2022). Furthermore, highly risk-averse travelers are less likely to use ChatGPT even if they believe it could be beneficial, suggesting that risk aversion can play a moderating role. In essence, incorporating personal innovativeness and risk aversion can enhance the predictive power and contextual relevance of the UTAUT2 model. The conceptual framework of the study is presented in Figure 1.

“Performance expectancy” (PE) refers to individuals’ belief in the system’s ability to enhance their performance and help them achieve gains (Venkatesh et al., 2003). PE is one of the primary determinants of intentions to use novel technology (Venkatesh et al., 2003). Numerous studies have shown the positive impact of PE on technology adoption (Alalwan et al., 2016; Tarhini et al., 2017). Terblanche and Kidd (2022) found PE to be a key factor influencing users’ intentions to use coaching chatbots. PE has been widely examined in the context of tourism research (Chung et al., 2018; Melián-González et al., 2021). In Malaysia, where efficient travel planning is crucial due to the diverse tourist attractions and cultural sites, the ability of ChatGPT to provide accurate and quick responses can significantly influence its adoption. Therefore, we hypothesize that:

“Effort expectancy” (EE) pertains to the level of simplicity connected with utilizing the system and the degree to which a person perceives the use of technology to be effortless (Yadav et al., 2016). The literature identifies EE as a significant driver of intention to use (Sharma et al., 2016). This association has been extensively supported across various research contexts, such as intention to use chatbots (Almahri et al., 2020), e-learning systems (Tarhini et al., 2017), and voice-based digital assistants (Vimalkumar et al., 2021). Shneiderman et al. (2016) have shown that users prefer user-friendly systems that require less effort. This is because users have limited time and attention, and they aim to complete their tasks quickly and efficiently. When a system is too complex or challenging to use, users may experience frustration and discouragement, leading to an unfavorable user experience. In the Malaysian context, where users may have varying levels of technological proficiency, ensuring that ChatGPT is user-friendly and intuitive is essential for its widespread adoption. Thus, we hypothesize that:

“Social influence” (SI) is the extent to which individuals believe influential people consider it necessary to use a specific technology (2003). The technology adoption literature commonly identifies SI as a critical determinant of adoption intention (Zhao, 2023). The significant positive effect of SI on usage intention has been widely supported within the chatbot adoption and traveling literature (Almahri et al., 2020; Medeiros et al., 2022). Al-Emran et al. (2023) stated that society and community acceptance and use are critical to the success of chatbot adoption. Kim et al.’s study (2021) found that SI plays a significant role in predicting the acceptance of chatbots for knowledge sharing. In a collectivist culture like Malaysia, where decisions are often influenced by family, friends, and community, the approval and recommendations from these groups can strongly impact the adoption of ChatGPT. Therefore, we propose that:

“Facilitating conditions” (FC) measure whether an individual believes that adequate infrastructure is in place for effective use of technology (Venkatesh et al., 2003). FC provides external resources that assist in performing a specific behavior. These resources can involve technological infrastructure, training, and support from colleagues or IT experts (El-Masri and Tarhini, 2017). Studies on chatbots have found FC to be a significant driver of usage intention (Kim et al., 2021; Rahim et al., 2022). In Malaysia, providing adequate support and resources to users can ensure they feel confident using ChatGPT for their travel planning needs. This includes access to tutorials, FAQs, and responsive customer support. Therefore, we hypothesize that:

“Hedonic Motivation” (HM) is the enjoyment or pleasure an individual experiences while using technology (Venkatesh et al., 2012). Tamilmani et al. (2019) found HM as a crucial driver of technology acceptance and sustained use. Jain et al. (2022) verified that the presence of enjoyment has a considerable impact on the overall perceived value, which, in turn, positively influences the intention to use voice assistants. HM has been identified as a predictor of chatbot and voice assistant usage intentions (Maroufkhani et al., 2022; Melián-González et al., 2021). Given that ChatGPT offers a unique and enjoyable way to interact with technology through natural language conversations, users are likely to be motivated by its fun and engaging experience. This is particularly relevant in a diverse and vibrant tourism market like Malaysia. Thus, we suggest that:

Habits (HB), as Venkatesh et al. (2012) states, refer to “automatic response to contextual cues that have been associated with prior behavior.” de Blanes Sebastián et al. (2023) found that HB has a noteworthy influence on technology usage intention (de Blanes Sebastián et al., 2023). Studies on technology adoption have identified HB as a crucial driver of users’ intention to use innovations (Singh et al., 2023; Ramírez-Correa et al., 2019). If the use of a system becomes a habit for users, they are more likely to use it (Tarhini et al., 2017). In the same vein, integrating ChatGPT into users’ routines as a common travel planning tool can increase its likelihood of being used regularly. Therefore, we proposed that:

“Personal innovativeness” (PI) refers to an individual’s willingness and ability to alter current circumstances and take risks (Bommer and Jalajas, 1999). Noh et al. (2014) described PI as an individual trait that significantly influences technology adoption and acceptance. Dajani and Abu Hegleh (2019) identified PI as a critical factor influencing the usage of animation technology. Likewise, Kasilingam (2020) found PI as a diver of chatbot usage. Strzelecki (2023) found that PI significantly predicts students using ChatGPT in higher education. Thus, PI is expected to influence travelers’ willingness to use it for travel planning purposes. Furthermore, we expect PI to play a notable moderating role. It means that individuals with higher levels of PI are more interested in innovative ideas and prone to using ChatGPT regardless of the extent of PE, EE, SI, FC, HM, and HB. It is expected that PI moderates negatively the impacts of influential factors on ChatGPT usage. In Malaysia, targeting individuals who are open to adopting new technologies can help accelerate the diffusion of ChatGPT. Innovative users are often early adopters who can influence others in their social circles (de Jong et al., 2023). Hence, we hypothesize that:

Hofstede and Bond (1984) defined “risk aversion” (RA) as “the extent to which people feel threatened by ambiguous situations and have created beliefs and institutions that try to avoid these”. RA involves anticipating adverse outcomes that may arise from technological innovations (Pennings and Smidts, 2000). Although it is commonly believed that humans are generally risk averse, the degree of this aversion may differ among individuals (Sun, 2014). Previous studies on technology adoption have suggested that RA can have a negative impact on the adoption of information technology (De Jong et al., 2003; Schleich et al., 2019; Yu et al., 2021). Cui (2022) found a negative relationship between RA and chatbot adoption. Accordingly, using ChatGPT for travel planning may be affected negatively by travelers’ RA. Furthermore, we expect RA to moderate negatively the impacts of influential factors on ChatGPT adoption. This means that highly RA individuals are less likely to use ChatGPT for travel planning, even if it can improve their performance or is easy to use. This is because high RA individuals are less open to taking risks and trying new things in uncertain situations. In Malaysia, addressing concerns about the reliability and security of ChatGPT can mitigate the impact of risk aversion. Providing clear information about data privacy and the accuracy of ChatGPT’s recommendations can help reduce users’ perceived risks. As such, we proposed that:

A survey questionnaire was used to collect data. The measurement items were adapted from previous studies in order to ensure that the constructs were valid and reliable ( Appendix A). A 5-point Likert scale ranging from 1 “Strongly Disagree” to 5 “Strongly Agree” was used to measure the items. The draft version of the questionnaire was pre-tested by three academicians who specialize in the adoption of technology in tourism and hospitality. The items were modified based on their comments. A pilot test was conducted with 53 Malaysians using the revised questionnaire. Cronbach’s alpha of measures was above 0.7, meaning that the measures were reliable.

The population of the study consisted of Malaysians aged 18 and above. The sample was limited to individuals aged 18 and over to ensure the respondents have the legal authority to plan travel and make autonomous decisions about travel-related activities, including the use of technologies like ChatGPT. Selecting adults ensures that the participants have the necessary experience and responsibility associated with travel planning and technology adoption. To collect data, we employed a convenience sampling technique, which involves selecting participants who are readily available and willing to participate. Although convenience sampling has limitations regarding representativeness, it is effective for gathering a large number of responses quickly and is suitable when there is no proper sampling frame. The data collection was conducted through an online survey administered via social media platforms such as Facebook, Instagram, and LinkedIn. These platforms were chosen due to their widespread use and accessibility among the target population. To ensure a broad reach, the survey link was shared on several social media pages and groups with a significant number of Malaysian followers, including travel-related groups, community pages, and university groups. Additionally, we employed a snowball sampling technique by encouraging initial respondents to share the survey link within their networks. The survey questionnaire included a video demonstrating how ChatGPT could be used for travel planning. This record contained some prompts related to travel posed to ChatGPT and the responses it provided. Furthermore, the ChatGPT login link was provided, and respondents were asked to practice getting travel information by entering their travel inquiries. This interactive component was designed to ensure that participants had a firsthand experience with ChatGPT before responding to the survey questions. We collected 410 usable datasets. Table 1 presents the profile of respondents.

The present study employed both linear and non-linear methods to investigate the importance of the suggested associations. The linear analysis was carried out using PLS. The choice of PLS was driven by its ability to handle the features of the gathered data (non-normal data), the exploratory focus of the study (e.g. identify drivers of ChatGPT adoption), and the model complexity (e.g. nine constructs) (Hair et al., 2019). Based on the multivariate normality test results, the data distribution was not normal, providing evidence that PLS is an appropriate method for the current research. Although the PLS technique provides valuable insights concerning causality, it ignores the likelihood of non-linear relationships. Therefore, an ANN method was conducted following the initial PLS methodology to pinpoint any considerable non-linear relationships that might influence the findings. While ANN surpasses regression-based techniques (e.g. PLS) in terms of accuracy and prediction power, their limitation lies in their inability to examine causal connections, causing them to be ineffective in testing hypotheses (Dadhich and Hiran, 2022). To tackle this shortcoming, we adopted a hybrid technique of PLS-ANN to discover the linear and non-linear relationships that lead to travelers’ ChatGPT adoption. First, the PLS method was employed to evaluate the measurement model (i.e. reliability and validity) and structural model (i.e. suggested hypotheses). The verified hypotheses from PLS analysis were further evaluated using ANN to rank them based on their significant importance.

The present research assessed validity and reliability of the constructs by determining “indicator loadings,” “composite reliability” (CR), and “average variance extracted” (AVE). As illustrated in Table 2, the loadings of the items exceed the suggested threshold of 0.7. Moreover, all constructs fulfilled the accepted thresholds for CR and AVE, with CR exceeding 0.7 and AVE surpassing 0.5.

Discriminant validity was assessed by calculating the Heterotrait-Monotrait (HTMT) ratio. As presented in Table 3, all HTMT values fell below the 0.85 threshold, thereby fulfilling the criteria for discriminant validity.

The proposed hypotheses were examined by applying the “bias-corrected and accelerated” (BCa) bootstrap method using 5,000 subsamples. The findings of this study highlighted that performance expectancy (β = 0.267; p < 0.01), facilitating condition (β = 0.151; p < 0.01), hedonic motivation (β = 0.297; p < 0.01), personal innovativeness (β = 0.178; p < 0.01), and risk aversion (β = −0.073; p < 0.1) all significantly contribute to the intention to use ChatGPT. Contrary to the anticipated outcomes, the research results suggest that effort expectancy (β = 0.038; p > 0.1), social influence (β = 0.026; p > 0.1), and habit (β = 0.038; p > 0.1) do not have a statistically considerable effect on the intention to use ChatGPT (Table 4). As reflected by the R² value, the presented model accounts for 55.5% of the variance in the intention to use ChatGPT.

The moderating results supported that personal innovativeness negatively moderates the impact of effort expectancy on the intention to use ChatGPT (Figure 2). Personal innovativeness does not moderate the impacts of other predictors of ChatGPT usage intention. Furthermore, risk aversion negatively moderates the effect of effort expectancy, social influence, and hedonic motivation on the intention to use ChatGPT (Figure 3). However, risk aversion does not moderate the impacts of performance expectancy, facilitating conditions, and habit.

To address the limitations of PLS in capturing non-linear relationships, this research employed an ANN analysis following the initial PLS methodology. ANN is capable of capturing complex non-linear relationships and provides robust predictive power, which makes it an ideal complement to PLS. The ANN analysis was conducted using SPSS software. First, the PLS scores of the constructs were exported and imported into SPSS to serve as the input for the ANN model. The ANN model was configured with one hidden layer, which is often sufficient for capturing non-linear patterns in social science research (Hew et al., 2023). The number of neurons in the hidden layer was determined based on a trial-and-error method to optimize model performance. The dataset was divided into training and testing sets, with 90% of the data used for training and 10% for testing to evaluate the model’s predictive accuracy. The training set was used to train the model, while the testing set was used to validate the model’s performance. The performance of the ANN model was assessed using metrics such as Root Mean Square Error (RMSE) to ensure the model’s robustness and predictive capability. Based on the research framework, we established a single ANN model for behavioral intention. In Table 5, the average RMSE values derived from the training and testing data were 0.097 and 0.096, respectively. These values underscore the high degree of predictive accuracy and confirm the model’s excellent fit.

Table 6 presents the exogenous variables according to their normalized relative importance in relation to the endogenous variable. The findings indicate that performance expectancy is the leading contributor to behavioral intention, with a normalized importance of 100%. It is closely followed by hedonic motivation at 99.29%, personal innovativeness at 74.36%, risk aversion at 39.50%, and facilitating conditions with a value of 34.30%.

Table 7 compares the PLS results based on the path coefficient and ANN, guided by normalized relative importance. When analyzed through the PLS and ANN models, the findings suggest a discrepancy in the rank of factors such as performance expectancy, facilitating conditions, hedonic motivation, and risk aversion.

By assisting travelers, providing informative responses, and making tailored advice regarding their journey, ChatGPT contributes to the promotion of tourism. However, due to its recent emergence, there is less empirical evidence on drivers that significantly affect the use of ChatGPT for travel planning. Understanding the factors influencing its adoption by travelers can help close this gap. Therefore, the present study developed a theoretical model, extending the UTAUT2 with personal innovativeness and risk aversion to explain how travelers adopt ChatGPT. The model is evaluated via a complementary approach involving PLS and ANN. The PLS results showed that performance expectancy, facilitating conditions, hedonic motivation, perceived innovativeness, and risk aversion significantly affect travelers’ intention to use ChatGPT. Personal innovativeness moderates negatively the impact of effort expectancy on ChatGPT usage. The PLS results also supported the significant role of risk aversion in moderating the impacts of effort expectancy, social influence, and hedonic motivation on travelers’ behavioral intentions. The ANN results further reinforced these results, suggesting that performance expectancy is the most important determinant affecting behavioral intention, followed by hedonic motivation, perceived innovativeness, risk aversion, and facilitating conditions.

The findings highlighted a positive relationship between performance expectancy and travelers’ intention to use ChatGPT. This result aligns with the findings of Terblanche and Kidd (2022) and Chung et al. (2018). The ANN results indicated that performance expectancy emerged as the most crucial driver of behavioral intention. This suggests that people are more likely to use ChatGPT if they believe that it can enhance the planning process. For instance, if a traveler believes that ChatGPT can provide accurate and tailored information about potential destinations, suggest the best routes, or help them budget effectively, they are more likely to use it for their travel planning. So, if a user is planning a trip to London, he/she may expect ChatGPT to provide a list of the best hotels in the city, suggest the best times to visit famous landmarks to avoid crowds, recommend local cuisine restaurants, or help with transport options from the airport to their hotel. As such, ChatGPT should be trained with travel information to meet or exceed travelers’ expectations.

Unlike the findings of Almahri et al. (2020) and Vimalkumar et al. (2021), which supported the relationship between effort expectancy and behavioral intention, this research did not support this relationship. There are several reasons for this insignificant relationship. First, ChatGPT, based on NLP, offers an intuitive textual user interface. Unlike other technologies that require understanding specific tools, icons, or navigation styles, interacting with ChatGPT might be as simple as typing a sentence. Consequently, “ease of use” may no longer be considered a determinant of intention to use. Second, the respondents are generally tech-savvy and familiar with AI technologies. They might not view effort expectancy as a significant factor affecting their intention to use ChatGPT. For example, a user would like to know the best places to visit in Dubai. With traditional travel planning tools, they may have to navigate through different tabs, enter specific search terms, or apply filters to get the needed information. However, with ChatGPT, they can ask, “What are the best places to visit in Dubai?” just as they might ask a human travel agent. This straightforward interface suggests that ease of use or usage difficulty does not significantly affect their intention to use ChatGPT.

The findings do not support the relationships between social influence and ChatGPT usage, contradicting the findings of Medeiros et al. (2022) and Al-Emran et al. (2023). One possible explanation for this result might be the individualistic nature of travel planning. When planning a trip, individuals often make decisions based on personal preferences rather than what others think they should do. Travel planning is a highly subjective task, and while travelers might take recommendations from others, they may not be heavily influenced by what technology others think they should use for this process. For instance, a user might use ChatGPT to plan their trip to Sydney not because their friends or influencers suggest it but because they find it convenient, efficient, and capable of delivering accurate and comprehensive information. Their decision is more influenced by their personal experience and assessment of the tool rather than the opinions of others.

We found facilitating conditions as an enabler of ChatGPT usage. This comes in agreement with prior findings (Kim et al., 2021; Rahim et al., 2022). In this context, facilitating conditions involve resources related to ChatGPT accessibility, reliable internet connection, availability of user guides or support, and other updates and improvements. For instance, a traveler planning a trip to Tokyo with ChatGPT would find it much easier to access ChatGPT on their smartphones, have a steady internet connection, and know they can quickly find help if they face any issues or have any questions.

In the same vein, the study found hedonic motivation as a significant driver of behavioral intention, which is consistent with Melián-González et al. (2021) and Maroufkhani et al. (2022). This indicates that travelers who enjoy ChatGPT are likelier to use it and explore its various features. For instance, a user planning a trip to Paris might find it enjoyable to interact with ChatGPT because it helps them plan their trip effectively and because they have fun asking about local culture, history, or traditional expressions. They might also enjoy getting information in an interactive, conversational way rather than passively reading from a screen.

Contrary to Ramírez-Correa et al. (2019) and Singh et al. (2023), the findings did not support the relationship between habit and ChatGPT usage. One possible explanation for this observation could be the relative novelty of chatbots. Due to its recency, there is a high probability that the users surveyed have not been exposed to chatbots for an extended period. In other words, they might not have formed a habit of using chatbots yet. Another reason might be the infrequent nature of travel planning for many people. Travel planning is not typically a daily or even monthly task, so it’s less likely that users will have formed strong habits around using chatbots for this purpose. For example, a person might be accustomed to using a specific app for daily tasks like email or social media. However, when planning a vacation that happens only once or twice a year, they may not have a specific tool or app they habitually use. Instead, they might explore different options each time, including using ChatGPT.

As per the results, the relationship between perceived innovativeness and ChatGPT usage was supported. This result agrees with Sitar-Taut and Mican’s (2021) and Kasilingam’s (2020) findings. People with high personal innovation are usually more open to new experiences, including trying out novel technologies. They enjoy learning and figuring out how to use a new tool, making them more likely to use ChatGPT for travel planning. For instance, innovative users might be intrigued by using an AI model like ChatGPT to plan their trips. They might enjoy exploring its features, trying out different ways of interacting with it, and discovering innovative ways it can assist them in their travel planning. However, personal innovativeness did not moderate the relationship between UTAUT2 factors and behavioral intention except for the impact of effort expectancy. The insignificant moderating effects indicate that whether a person is innovative or not, the extent to which these factors influence their intention to use ChatGPT remains the same. This finding could be because ChatGPT’s functionalities could similarly impact all users, irrespective of their innovative level. For instance, both innovative and less innovative users might appreciate the capability of ChatGPT to offer detailed and personalized travel recommendations and enjoy the interaction. Personal innovativeness moderates negatively the impact of effort expectancy on ChatGPT usage. As innovative users enjoy exploring new technology, ease of use less affects their decision to use ChatGPT.

Risk aversion was found to have a significant negative influence on ChatGPT usage, which comes in agreement with the findings of Yu et al. (2021) and Schleich et al. (2019). A user might be concerned about the reliability of the information provided by ChatGPT. As a generative AI technology, ChatGPT generates responses based on the data it has been trained on but does not have real-world experience or the ability to verify information independently. A risk-averse user might worry that the information provided could be inaccurate or outdated. Another concern might be data privacy. Users share information with ChatGPT during their interaction, and a risk-averse user might worry about how this information is used and protected. A risk-averse traveler might hesitate to use ChatGPT to plan their trip because they are unsure if it can provide reliable information on travel requirements or other issues, such as COVID-19 regulations. They might also worry about sharing details of their planned trip with ChatGPT.

Risk aversion also moderates the relationship between effort expectancy, social influence, hedonic motivation, and traveler behavioral intentions negatively. For instance, users might perceive ChatGPT as easy to use, indicating high effort expectancy. However, if they are risk-averse, they might still hesitate to use it due to concerns about potential risks like data security or reliability. Similarly, while individuals generally use technologies adopted by their social environment, a risk-averse user may resist such social influence if they perceive potential risks in using ChatGPT, despite seeing their peers use it. Even the enjoyment derived from using the ChatGPT might not be enough to convince a risk-averse user to adopt it if they have concerns about aspects such as the accuracy of information. On the other hand, risk aversion did not moderate the impacts of performance expectancy, facilitating conditions, and habit on ChatGPT usage. This suggests that no matter how risk-averse individuals are, their intention to use ChatGPT remains consistent if they expect the platform to enhance their travel planning and if there is adequate infrastructure and support. Habit has effect on decision to use ChatGPT regardless of the extent to which users are risk averse. This reinforces the importance of promoting ChatGPT’s benefits and providing supportive conditions for its use to enhance user adoption across varying levels of risk aversion.

This research significantly contributes to the theoretical understanding of generative AI, specifically ChatGPT, and its adoption for travel planning. First, it reinforces the UTAUT2 model by confirming the role of performance expectancy, facilitating conditions, and hedonic motivation in influencing behavioral intentions, but with a new focus on generative AI-based travel planning tools. This expands the application of UTAUT2, showing its validity in the emerging context of AI-driven applications. Second, the study offers new insights by challenging common beliefs regarding effort expectancy and social influence. These findings broaden our understanding of user behavior in the AI context, suggesting that “ease of use” and “social influence” may not always play a significant role, especially when AI interfaces are naturally intuitive, and the task is highly personal, like travel planning.

Third, adding personal innovativeness to the model reveals the role of individual differences in technology adoption. It shows that those more inclined to try new things are likelier to adopt ChatGPT, furthering our understanding of individual variability in technology acceptance. Fourth, this study introduced risk aversion into the model, uncovering its significant negative impact on behavioral intention. Furthermore, it identifies risk aversion as a moderating variable affecting the relationship between several UTAUT2 factors and behavioral intention. This draws attention to the importance of risk perception in the context of generative AI adoption. Fifth, the study went beyond the conventional linear regression models and used a hybrid PLS-ANN approach. The ranking of the influence of factors on ChatGPT adoption was not consistent between PLS and ANN results. Although, according to PLS, hedonic motivation is the most influential driver of ChatGPT usage, ANN revealed that performance expectancy is the most influential driver. According to Sharma et al. (2021), ANN is a more accurate approach to ranking factors in comparison to PLS as it considers non-linear relationships. This finding confirms the importance of using a non-linear approach as a complementary technique to PLS.

This study offers several practical contributions, particularly within the Malaysian context, where the adoption of AI technologies like ChatGPT is still emerging. By understanding the factors that influence the adoption of ChatGPT for travel planning, stakeholders in the tourism and hospitality industry can develop targeted strategies to enhance user engagement and satisfaction. Firstly, the findings highlight the significant role of performance expectancy and hedonic motivation in driving ChatGPT adoption. Malaysian tourism services providers, such as travel agencies, hotels, and tour operators, can leverage this insight by ensuring that their AI-driven services are not only functional but also enjoyable to use. Incorporating interactive and entertaining elements in ChatGPT-based applications can enhance user experience and encourage more widespread adoption. Secondly, the study underscores the importance of facilitating conditions and personal innovativeness. For the Malaysian market, this suggests that providing adequate support and resources, such as user guides and customer service, can enhance the ease of use and perceived value of ChatGPT. Additionally, targeting tech-savvy individuals who are more likely to experiment with new technologies can accelerate the diffusion of ChatGPT in travel planning. Promotional campaigns and educational workshops can be effective in reaching this demographic. Thirdly, the influence of risk aversion on ChatGPT adoption suggests that Malaysian service providers need to address users’ concerns about the reliability and security of AI technologies. Clear communication about data privacy measures and the reliability of ChatGPT’s recommendations can mitigate these concerns. Implementing robust security protocols and providing transparent information about data handling practices will be crucial in building user trust. Moreover, for Malaysian policymakers and industry leaders, these insights can inform the development of supportive regulations and initiatives to promote AI integration in tourism services. Collaborating with technology providers and investing in AI infrastructure can position Malaysia as a leader in innovative travel solutions. Finally, the insights from this study can help Malaysian educational institutions design curricula that prepare students for careers in AI and tourism. By integrating AI-related courses and practical training programs, universities and vocational schools can equip future professionals with the skills needed to develop and manage AI-driven applications in the tourism sector.

The conceptual model of the study was tested with a sample of Malaysian respondents. Future studies are recommended to test the study model in other countries and compare the findings. In contrast to our expectations, the findings revealed that habit does not affect ChatGPT usage. Investigating the influential factors at the early stages of ChatGPT diffusion is the potential reason for this insignificant association. Future studies can assess the effect of habit after further maturation of ChaptGPT usage within the tourism context. One of the limitations of this study is that it did not explicitly consider the humanlike nature of ChatGPT, which can significantly impact user interactions and perceptions. The UTAUT2 model was selected for its robustness in explaining technology adoption, but future research should incorporate factors that capture the anthropomorphic qualities of ChatGPT, such as perceived human-likeness or social presence. These factors could provide deeper insights into how the humanlike nature of ChatGPT influences its adoption and user experience. Finally, the study showed the importance of two personal characteristics, namely personal innovativeness and risk aversion, in choosing ChatGPT. Future studies are recommended to assess the impacts of other personal characteristics such as technological anxiety, self-efficacy, and propensity to trust.