Artificial Intelligence and Tourism: A Bibliometric Visualisation Review (2015-2024)

Bibliometric analysis

3.1

Bibliometric analysis enables visualisation of the intellectual structure of a given field of research and identification of current and potential lines of research (Al-Rousan & Khasawneh, 2024). It has a number of advantages over traditional literature review analysis because, based on a series of quantitative and statistical techniques, it allows us to objectively visualise a scientific field through different types of analysis (bibliographic coupling, co-citation or keyword analysis, among others) (Khanra et al, 2020). In addition, we can see which journals, authors, academic institutions or bibliographic references have a greater scientific impact (Florido-Benítez, 2022).

Web of Science will be the database of choice for bibliometric analysis due to its academic acceptance, both in systematic literature review work and as a scientific basis for the creation of scientific articles (Al-Rousan & Khasawneh, 2024; Zulfiqar et al., 2024). This is due to its longevity, with more than 100 years, and its scope, with 65 million articles and more than 1 billion cited reference connections (Kim and So, 2022). Comparison with other databases shows that WoS has advantages over Google Scholar due to transparency, ease and quality of documents (Harzing & Alakangas, 2016). On the other hand, in the case of Scopus, integrating both databases creates difficulties, as both lack a homogeneous structure, increasing the risk of information duplication (De Oliveira et al., 2019; Atsiz et al., 2022). Due to these reasons, most researchers use WoS as their database, as it has a much broader structure and content.

Regarding the data analysis programme, our article uses the Visualisation of Similarities (VOS) viewer software (version 1.6.20), which will allow us to analyse the scientific structure of a given scientific field through cocitation analysis and to predict future lines of research and their evolution over time, through keyword analysis (Baker et al., 2021). This programme is widely accepted by researchers due to the creation of scientific maps that allow them to visualise thematic clusters and observe different lines of research (Gao et al., 2021).

Search string

3.2

First, we identify keywords to obtain accurate and reliable results. Researchers use, interchangeably, two methods in order to select keywords in a paper (Chen & Xiao, 2016): a) a macro approach, by searching for all keywords related to the scientific field; b) a micro approach, by analysing the details of the main research topics in a domain. Based on this, our article will be based on the micro approach from the articles with the highest number of citations in this scientific field (Buhalis et al., 2019; Lu et al., 2019; de Kervenoael et al., 2020; Pillai & Sivathanu, 2020; Dwivedi et al., 2023).

Following the Web of Science search, the PRISMA protocol was employed to refine and filter the database for relevant records (Moher et al., 1999). First, the search query only included TOPIC, so the keywords “artificial intelligence” AND “tourism” were included in the abstract, keywords or title. Secondly, only papers in English were included, corresponding to the period 2015 to 2024. Thirdly, to eliminate bias, the papers were carefully screened, and those unrelated to the subject matter of the study were excluded. Finally, the Excel spreadsheet was used to check for duplicate records. The research questions and review protocol conducted for this study are shown in Figure 1.

Figure 1:

Performance analysis

4.1

Table 1 shows the information on the scientific field of AI and Tourism (Table 1). We also note that, although this field was born in 2015, there are a large number of citations and articles, indicating that it is an emerging topic and that its scientific output will increase in the future.

Table 2 shows the academic impact and scientific production of the different authors in the scientific field analysed. Additional bibliometric indicators from WoS and Scopus were incorporated into the published articles, with their respective citations, to assess and visualize the greater or lesser academic relevance of both the research and its authors. Rob Law, from the University of Macau (China), is the researcher with the highest number of papers; while Dogan Gursoy from Washington State University (USA); Seongseop Kim from Hong Kong Polytechnic University (China); and Dimitrios Buhalis from Bournemouth University (United Kingdom) are the researchers with the highest average number of citations per paper. According to the h-index metric, these authors have a consolidated and relevant scientific trajectory (h-index >20) (Hirsch, 2005; Díaz, 2014). Table-3 shows the academic institutions with the greatest scientific relevance with respect to the subject of analysis, with United Kingdom, China and the United States standing out.

Co-citation analysis

4.2

From the co-citation analysis, we will identify the intellectual structure of the scientific field analysed (Carvajal-Trujillo et al., 2024). In this work, we considered those documents that had been jointly cited in at least 40 papers. Accordingly, two distinct clusters of documents have been identified (Figure 2). The content of the most academically relevant papers in each of these clusters is described below:

Figure 2:

– Cluster Red: “Robots in the service sector”: This cluster is focused on analysing the benefits of robots in the service sector for consumers, as well as the opportunities and improved experience. One of the papers with the greatest academic impact is Wirtz et al. (2018), who analyse the benefits of robots for consumers and the ethical and moral issues. To do so, the authors conceptualise the Service Robot Acceptance Model (sRAM), which is based on the technology acceptance model (TAM; Davis, 1989). To the TAM, they add the social-emotional elements of service (Heerink et al., 2008; van Doorn et al., 2017) and relational elements (Heerink et al., 2010; Nomura & Kanda, 2016). Thus, this model consists of three elements: functional elements (perceived ease of use, perceived usefulness, and subjective social norms), social-emotional elements (perceived humanness, perceived social interactivity, and perceived social presence), and relational elements (trust and rapport). The authors conclude with a series of moral dilemmas which must be addressed in the future, such as privacy and security, and dehumanisation (consumers), technological gap, and loss of jobs (society), monopolies, loopholes, and loss of human contact (markets). A second relevant paper is by Van-Doorn et al. (2017), who introduce the concept of automated social presence, in which consumers feel that they are with another human when they establish a relationship with a robot. They build on Biocca and Harms’ (2002) conceptual framework of social presence, which they define as a “sense of being with another”. This article serves as a basis for the analysis of Wirtz et al. (2018), because it concludes that robot characteristics such as warmth of service, competence, responsiveness, attractiveness and manageability imply technological acceptance of the robot.

– Green cluster: “Benefits and risks of using Artificial Intelligence in Tourism”: This cluster analyses the use of AI in Tourism. One of the articles with the greatest academic impact is that of Pillai and Sivathanu (2020), who analyse the perceptions of customers and travel agency managers regarding the use of AI-powered chatbots in India. The proposed model is based on the Technology Acceptance Model (Davis, 1989), using the constructs of Perceived Ease of Use and Perceived Usefulness. In addition, they added context-specific variables such as anthropomorphism (Mori, 1970), perceived intelligence (Warner & Sugarman, 1986), perceived trust (Mayer et al., 1995) and technology anxiety (Igbaria & Parasuraman, 1989). A second relevant paper is Li et al. (2019), who analysed the substitution effect of AI on tourism employment. To do so, they analysed the psychological pressure AI exerts on workers to quit their jobs, considering the moderating role of perceived organisational support and competitive psychological climate. To do so, they drew on the estimates of Frey and Osborne (2017), who predict that almost half of all US jobs will be automated in the coming decades, and the forecasts of Bowles (2014), who predicted that 54% of European occupations could be computerised through rapid automation.

Co-occurrence analysis

4.3

The analysis of keyword co-occurrence enables identification of distinct research fields within the analysed scientific domain (Callon et al., 1991; Liu et al., 2015). Following Cordova-Buiza (2025), this analysis is based on the “author keywords” provided in the analysis documents and not on the “keyword plus” indexed automatically by WoS. The thesaurus was applied with the aim of eliminating synonymy problems, avoiding repetitive terms and standardising information (Van Eck & Waltman, 2022). Table 4 provides the 20 keywords with the highest number of occurrences and their total link strength, with “artificial intelligence”, “tourism” and “robot” being the most scientifically relevant keywords.

The analysis by keyword groups from a number of 12 co-occurrences allows the identification of four relevant clusters of heterogeneous keywords (Figure 3):

Figure 3:

– Red cluster: “Machine learning, deep learning and artificial neural networks in tourism studies”: Machine learning, deep learning and artificial neural networks are hierarchically related approaches within the field of AI. In this context, machine learning is used for AI to learn data systematically without being explicitly programmed for each task; artificial neural networks are computational models inspired by the functioning of the human brain, and deep learning is a subtype of neural networks characterised by multiple hidden layers, allowing it to process larger volumes of data and tackle more complex tasks (IBM, 2023). In this sense, the articles in this cluster use this tool as a methodology for the purpose of analysing tourism. One of the articles with the greatest impact is Seker (2024), which uses a hybrid methodological approach to analyse international tourism receipts in Turkey and predict future behaviour. This method uses, on the one hand, traditional mathematical models based on polynomial equations and, on the other hand, machine learning algorithms based on AI. The results predict that tourism revenues will not exceed USD 45 billion in 2027, despite the expected high number of visitors. This is due to factors such as the competitiveness of nearby tourist destinations (Spain or Greece) and, especially, all-inclusive hotels, which limit tourists’ additional spending. Furthermore, the methodology shows how these algorithms can be effective tools for predicting tourism demand, offering new opportunities for strategic planning and decision-making in the sector. Another relevant current paper is Puri et al. (2023), who perform a bibliometric analysis on the tourism blockchain using the VOSviewer tool and AI models with the help of machine learning tools. Using the Web of Science database, the authors conclude that the main applications of the tourism blockchain are related to investments, consumer feedback, hotel reservations, car rentals, donations and immersive technology.

– Green cluster: “Use of robots in tourism”: This cluster analyses the use of robots in the tourism sector, from their different applications and benefits to their disadvantages. They are also related to the COVID-19 pandemic because many jobs were replaced by robots due to protective measures such as social distancing or confinement. One of the most relevant articles is Ivanov’s (2024), which conceptually analyses the role of robots in tourism. The author shows that in developed countries, there is an increasing lack of workers in this sector due to factors such as low birth rate, increased bargaining power, higher educational level or harsh working conditions. In addition, the author highlights the fragility of jobs during negative external shocks, such as COVID-19, and the positive role of robots in such contexts. In terms of practical recommendations, the authors consider that robots can help in this context, e.g. by automating tasks, increasing employee efficiency, or improving working conditions by eliminating repetitive or dangerous tasks. Another relevant paper is Zeng et al. (2020), who highlight the relevant role acquired by robots during the COVID-19 pandemic. In this way, the negative connotations of robots before the pandemic, due to concerns about job losses and data privacy, have been lost due to the help they have provided. Thus, the authors highlight aspects such as the distribution of materials, the disinfection and sterilisation of public spaces, or the detection or measurement of body temperature, among others. All in all, the authors establish a series of practical recommendations for the use of robots during the tourist experience, such as the surveillance and protection of natural resources or navigation systems to avoid tourist overcrowding.

– Blue cluster: “Sustainability and AI in tourism”: This group of articles analyses the role of AI in promoting sustainability in the tourism sector. A relevant article is that of Siddik et al. (2025), who integrate artificial neural network (ANN) analysis with traditional econometric models in order to analyse the role of AI in promoting sustainable growth in tourism. To do so, they investigate the interaction between AI adoption, gross domestic product, foreign direct investment, inflation and urbanisation in the top ten global tourism destinations between 2010 and 2022 (Austria, France, Germany, Greece, Turkey, Spain, Mexico, Italy, the United Kingdom, and the United States). Among other findings, the authors highlight that the factor that has the greatest influence on AI is economic growth. Furthermore, in relation to sustainability, AI serves, on the one hand, to manage the influx of tourists in cities such as Barcelona and, on the other hand, to improve sustainable urban development. Another highly relevant article is that of Balsalobre-Lorente et al. (2023), who analyse the role of AI and Information and Communication Technologies (ICTs) in improving sustainability in tourism and assess whether digitalisation acts as a catalyst for achieving the Sustainable Development Goals (SDGs). The authors analysed the long-term effect of per capita income, tourism, natural resources, urbanisation, ICTs and AI on environmental sustainability in 36 OECD economies between 2000 and 2018. The authors highlight the role of AI in improving tourism efficiency and thus sustainability. This is highlighted in factors such as the creation of smart cities or the efficiency in the exploitation of natural resources.

– Yellow cluster “Acceptance of chatbots in tourism”: This cluster analyses the implementation of chatbots in tourism (such as ChatGPT), due to the advantages and risks they face. In addition, the authors of this cluster start from the TAM model of Davis (1989), with the aim of analysing the relationship between intention to use and actual consumer behaviour. In this way, Davis (1989) initially proposed that when users are confronted with a new technology, there is a set of factors that influence the intention to use and the actual use of the technology: perceived usefulness and perceived ease of use. One of the most prominent articles is Abed (2024), which establishes a behavioural model to analyse the intention to use and the actual use of Artificial Intelligence in tourism among Generation Z inhabitants in Saudi Arabia. Based on the TAM and DeLone and McLean’s Information Systems Success Model (DeLone & McLean, 2003), they analyse how system quality, service quality and information quality influence perceived usefulness and perceived ease of use, and these influence intention to use and actual use. All hypotheses were supported, except for the effect of service quality on service usability. Among the practical implications, the authors recommend the use of chatbots in tourism for the purpose of tourism planning and travel management. In addition, programmers should ensure that they provide accurate information while intelligently automating the tourism business with real-time solutions. A second important research study is that of Lei et al. (2021), who analyse the difference in the communication perceptions tourists get from chatbots on the one hand and a human agent on the other hand. To identify the factors that influence future usage intention, they draw on the TAM and the Unified Theory of Acceptance and Use of Technology (Venkatesh, 2003). In this sense, they analyse the effect of social presence and media richness on social attraction and task attraction, which in turn influence trust and future use intention. When comparing the difference in perception between the two sources of information, the causal relationships are fulfilled in both lines of communication, with the exception of the relationship between social presence and task attraction in human agents. Furthermore, those users who communicated with human agents scored significantly higher on social presence, media richness, social attraction, task attraction, trust and reuse intention than chatbot users. Among the practical implications, the authors recommend that chatbots should improve communication and problem-solving skills, allow users to comment on the service provided, improve the speed of response, and improve the ability to understand human language, among others.