Mapping generative AI research in tourism and hospitality

You’re listening to “Mapping generative AI research in tourism and hospitality,” by M. Malekpour and colleagues. Published in 2026.

ISSN: 1368-3500 (Print) 1747-7603 (Online) Journal homepage: the linked source

Mehrgan Malekpour, Oswin Maurer, Hatice Kizgin & Fevzi Okumus

To cite this article: Mehrgan Malekpour, Oswin Maurer, Hatice Kizgin & Fevzi Okumus (23 May 2026): Mapping generative AI research in tourism and hospitality, Current Issues in Tourism, DOI: 10.1080/13683500.2026.2674077

© 2026 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group View supplementary material

Published online: 23 May 2026.

Submit your article to this journal Article views: 2116 View related articles View Crossmark data Citing articles: 1 View citing articles

RESEARCH ARTICLE

Mapping generative AI research in tourism and hospitality b and Fevzi Okumus a, Oswin Maurer a, Hatice Kizgin c,d,e,f Mehrgan Malekpour aDepartment of Economics and Management, Free University of Bozen-Bolzano, Bozen-Bolzano, Italy; bDepartment of High Tech Business and Entrepreneurship, University of Twente, Enschede, The Netherlands; cCollege of Hospitality, Retail and Sport Management, University of South Carolina, South Carolina, USA; dCollege of Hotel and Tourism Management, Kyung Hee University, Seoul, Republic of Korea; eHumanities and Social Sciences Research Center (HSSRC), Deanship of Scientific Research, Imam Mohammad Ibn Saud Islamic University (IMSIU), Riyadh, The Kingdom of Saudi Arabia; fThe Cesar Ritz Colleges, Brig-Glis, Switzerland

ABSTRACT.

Generative AI is reshaping the tourism and hospitality (T&H) industry. This study synthesises existing research, identifies key trends and challenges, and proposes a framework to guide future research. Using the SPAR-4-SLR protocol, we triangulate a systematic literature review with bibliometric mapping, qualitative content analysis, and lexicometric analysis of 83 peer-reviewed articles. We introduce and apply an extended ABCDE + D–TCM framework, integrating Antecedents, Barriers, Drivers, Decisions, and Effects with Theories, Contexts, and Methods to provide a structured, theory-grounded analysis. Findings reveal five dimensions of GenAI adoption in T&H: antecedents (e.g. personalisation demand), barriers (e.g. trust issues), drivers (e.g. perceived usefulness), decisions (stakeholder adoption), and effects (e.g. enhanced experiences).

Research is largely customer-focused, quantitative, and concentrated in the United States, China, and South Korea, with limited attention to employees, organisations, and underrepresented regions. This is the first study to propose and apply the extended ABCDE + D– TCM framework, and offers a structured, multi-dimensional lens to assess how GenAI is understood and applied across tourism and hospitality research.

1. Introduction.

ARTICLE HISTORY

Received 11 June 2025 Accepted 9 May 2026

Generative AI (GenAI); systematic literature review; tourism and hospitality; ABCDE+D–TCM framework; SPAR-4-SLR

Generative AI (GenAI) is reshaping the T&H sector. Unlike earlier AI systems, GenAI offers new possibilities for efficiency, personalisation, and engagement and enables firms to generate high-quality marketing content, analyze large-scale data and deliver tailored experiences. Practical applications are already evident, such as Pegasus Airlines’ FlyBot for personalised travel assistance and AI-driven platforms like Google’s Gemini and Booking.com, which enhance itinerary planning and user experience. Despite this progress, GenAI adoption in T&H remains constrained by technical complexity, ethical concerns, high implementation costs, and industry structure, as many hospitality firms are small, often family-owned enterprises. Given its growing diffusion and strategic importance, overcoming these barriers is an urgent priority for research and practice.

Still, the understanding and application of generative technologies within the T&H industry remain underdeveloped, despite rapid advancements in other sectors. It highlights the need for a timely and holistic investigation of its theoretical and practical implications. This study addresses this gap through a comprehensive multi-method review, combining (i) a systematic literature review, (ii) bibliometric analysis, (iii) qualitative content analysis, and (iv) lexicometric analysis. Building on this foundation, this study introduces a novel analytical lens combining the ABCDE model, the decision component of the ADO framework, and the theoretical grounding of the TCM framework. This structure enables a comprehensive analysis of GenAI adoption in T&H by capturing antecedents, barriers, drivers, decisions, and effects within a broader theoretical and methodological context.

Several frameworks have structured prior research, including ADO, TCM, TCCM, and the integrated ADO–TCM. While valuable, these frameworks have limitations for analyzing complex technologies such as GenAI. ADO captures antecedents, decisions, and outcomes but overlooks barriers and drivers. TCM and TCCM provide strong theoretical and methodological classifications yet remain largely descriptive and do not capture adoption dynamics. Although ADO–TCM offers a more holistic view, it lacks sufficient granularity regarding barriers, challenges, and drivers in recent GenAI research.

Building on these limitations, this study adopts the ABCDE framework to capture antecedents, barriers, challenges, drivers, and effects. The decision component from ADO is retained as a distinct dimension, reflecting its key role in shaping GenAI adoption in T&H contexts. Finally, the TCM framework is maintained as the theoretical–methodological lens, while TCCM is not adopted to avoid conceptual redundancy, as ABCDE + D already captures relevant constructs and characteristics.

While ADO focuses on antecedents, decisions, and outcomes, and TCCM emphasises theoretical and methodological classification, the ABCDE + D–TCM framework integrates both by linking content-level adoption mechanisms with their theoretical and methodological foundations. It is inductively grounded in patterns identified through qualitative content analysis and supported by bibliometric and lexicometric evidence across the corpus.

Accordingly, the study explores the following research questions: (RQ1) What is known about the impact of GenAI on customer experiences and business transformation in the T&H industry? (RQ2) What is the current body of knowledge on the impact of GenAI on T&H research? (RQ3) What should future research focus on to deepen the understanding of the opportunities and challenges posed by GenAI in T&H?

This study distinguishes itself from prior systematic reviews on GenAI in T&H (Web Appendix

2. summary of previous literature reviews) in three fundamental ways.

First, while earlier works

have largely adopted a broad or purely bibliometric approach, this study applies a focused, stakeholder-based, and theory-driven lens. Specifically, it examines the impact of GenAI across four core groups: consumers, businesses, employees, and educators. Second, it introduces a novel, and author-developed analytical framework-ABCDE + D–TCM and integrates four complementary methods (systematic review, bibliometric mapping, qualitative content analysis, and lexicometric analysis) under this unified structure. Third, by applying the SPAR-4-SLR protocol and analyzing 83 studies up to October 2025, the review ensures both methodological rigour and up-to-date relevance, to capture the latest developments that are missing or were overlooked in earlier reviews. Hence, it fills a conceptual gap in previous studies.

These features provide three contributions: (i) advancing theoretical understanding of GenAI’s impact across stakeholders, (ii) strengthening the methodological foundation through a dual-framework approach, and (iii) offering an integrated synthesis of GenAI’s opportunities and challenges in T&H.

2. Methodology.

2.1. Review approach.

This study employs a hybrid review integrating SLR, bibliometric, qualitative content, and lexicometric analyses to ensure methodological rigour and coherence.

2.1.1. SLR.

The SLR forms the foundation and addresses RQ1 by examining GenAI’s impact on customer experiences and business transformation. Following the SPAR-4-SLR protocol, it enabled the identification and synthesis of 83 peer-reviewed articles, providing insights into publication trends, key journals, authors, and themes.

2.1.2. Bibliometric analysis.

Complementing the SLR, the bibliometric analysis addresses RQ2 by mapping the intellectual and methodological structure of GenAI research in T&H. Using VOSviewer (v.1.6.20) and Biblioshiny (bibliometrix), it analyzes performance indicators, citation patterns, and intellectual linkages.

2.1.3. Lexicometric analysis.

The lexicometric analysis addresses RQ3 by identifying gaps, challenges, and future research directions through textual patterns and concept co-occurrence. Based on a corpus of 15,237 word occurrences, it applies hierarchical descending classification (CHD) and similarity analysis to reveal lexical structures and thematic clusters.

2.1.4. Qualitative content analysis.

To structure GenAI adoption in T&H, we conducted a qualitative thematic analysis following Braun and Clarke’s (2006) framework. Guided by the ABCDE + D model, data were organised into five dimensions across four stakeholder groups: customers, employees, businesses, and academics, addressing RQ1 and RQ3.

Figure 1 illustrates the hybrid approach and analytical framework.

2.2. Search strategy.

To select papers for the literature review, a comprehensive search procedure called the SPAR-4-SLR framework was used. SPAR-4-SLR is a clear, step-by-step process structured into three key stages: assembling, arranging, and assessing (Figure 2). The Web of Science (WoS) database was chosen for its reliability and extensive collection of peer-reviewed journals, preferred for its high-quality standards compared to other databases.

2.2.1. Organising framework.

An organising framework provides a structured lens for synthesising fragmented literature. Building on this, the study adopts the ABCDE + D–TCM framework as the main analytical structure. The ABCDE + D model guides qualitative content analysis by classifying evidence on antecedents, barriers, challenges, drivers, decisions, and effects of GenAI adoption across stakeholders, while TCM maps theoretical foundations, contexts, and methods of the reviewed studies.

This combined approach reflects an inductive, data-driven process based on recurring patterns in the corpus, ensuring a comprehensive, coherent, and transparent synthesis of GenAI research in T&H.

2.2.2. SPAR-4-SLR.

The assembling stage consists of two sub-stages: identification and acquisition. Identification involves defining the domain and research questions to establish the scope of the review. The research is guided by three key questions. Acquisition involves the process of collecting relevant literature based on predefined search parameters. The literature search was conducted up to October 2025, ensuring coverage of the most recent advancements in the field. The search targeted title, abstract, and keywords and it yielded 992 articles. Related keywords are shown in Web Appendix 1. Keywords were developed through a two-step process: reviewing prior T&H studies to identify common terms (e.g. ‘Generative AI’, ‘ChatGPT’) and conducting pilot searches to refine recall and precision, ensuring comprehensive and relevant coverage.

The arranging stage includes purification and organisation. Purification filters studies by language (English, 988), document type (articles, 695), and research area, reducing the sample to 437 articles. Restricting to SSCI-indexed T&H journals further narrowed it to 240 articles. Based on the CABS Academic Journal Guide, 109 journals were identified, and screening titles and abstracts reduced the sample to 112, with 83 articles selected for full-text review (Web Appendix 3). In the organization stage, the extended ABCDE + D–TCM framework guides the analysis.

The assessment stage is made up of two sub-stages: evaluation and reporting. In the evaluation stage, content analysis, bibliometric and lexicometric analysis were used. In the reporting stage, several methods including the use of figures, tables, and word visualisations were used to provide a clear and comprehensive overview of the current research landscape (see Figure 2).

3. Findings.

3.1. Results from the SLR.

3.1.1. Article and yearly publication trends.

Figure 3 shows a sharp rise in GenAI research in T&H. Publications began in 2023 and increased to 18 in 2024 (↑200%), reflecting a shift toward more structured academic engagement. Growth peaked in 2025, indicating accelerating interest and GenAI’s transition to a core research priority aligned with broader AI developments.

This surge reflects rapid diffusion following the release of large language models in late 2022. Transformer-based systems expanded accessibility and utility across academia and industry, while advances in infrastructure and model architectures enhanced applicability. Considering publication cycles, the 18–24-month lag explains the concentration of studies in 2025, indicating a structural diffusion trend. One article has appeared in 2026 but is excluded as the year is ongoing.

3.1.2. Journal publication trends.

Publication trends of GenAI–T&H research (see Web Appendix 4) show that 83 articles were published across 18 peer-reviewed journals, with the highest concentration in International Journal of Hospitality Management (n = 14), followed by Current Issues in Tourism (n = 13), and Journal of Travel Research and Tourism Management (n = 7 each). CABS and ABDC rankings indicate a strong presence of high-quality journals (e.g. 3–4; A/A).

3.1.3. Most cited articles.

The most cited articles indicate that the leading study, ‘Autonomous travel decision-making: An early glimpse into ChatGPT and generative AI’, has 136 citations (45.33 per year). Other influential works address trust in ChatGPT (71 citations) and AI hallucinations (68 citations). Most highly cited studies were published in 2023–2024, reflecting the rapid rise of GenAI research (see Web Appendix 5).

3.2. Results from the bibliometric analyses.

Following the systematic review, bibliometric techniques were applied to map the intellectual structure of GenAI research in T&H.

3.2.1. Co-citation analysis.

The co-citation analysis examines how frequently pairs of references are cited together. In this study, 4,573 cited references were detected in the dataset, of which 103 met the minimum threshold of four co-citations and were included in the network. The results revealed three major clusters (see Figure 4).

3.2.1.1. Cluster 1: GenAI adoption and trust dynamics.

This cluster (red-coloured) examines GenAI adoption in T&H, focusing on decision processes, trust formation, and value co-creation. Key studies highlight factors such as prior experience, perceived intelligence, credibility, and recommendation quality in shaping trust and usage intentions.

3.2.1.2. Cluster 2: GenAI capabilities and adoption pathways.

This cluster (green-coloured) focuses on AI capabilities and adoption drivers in T&H. Studies emphasise perceived usefulness, ease of use, anthropomorphism, and trust, alongside risks such as bias, privacy concerns, and misinformation.

3.2.1.3. Cluster 3: GenAI foundations, risks, and transformation.

This cluster (blue-coloured) addresses GenAI’s broader implications, including ethics, transparency, and digital transformation. Key studies highlight both benefits and risks, reflecting the strategic and societal dimensions of GenAI in T&H.

3.2.2. The results of bibliographic coupling analysis.

The bibliographic coupling analysis identified seven thematic clusters, each reflecting a distinct research focus within GenAI scholarship in T&H (Figure 5).

3.2.2.1. Cluster 1: drivers and barriers of GenAI adoption.

This cluster (red-coloured) brings together studies on drivers and barriers of ChatGPT adoption across tourists, employees, and organisations. Key works highlight attitudinal barriers and motivational drivers.

3.2.2.2. Cluster 2: GenAI capabilities and system-level functions.

This cluster (green-coloured) focuses on GenAI’s technical and methodological capabilities, including recommendation quality, automation, and efficiency. Studies highlight its methodological utility and advances over traditional approaches, underscoring the growing sophistication and practical value of GenAI tools.

3.2.2.3. Cluster 3: effects of GenAI features on tourist decision-making.

This cluster (Dark blue- coloured) examines how features such as personalisation, anthropomorphism, and accuracy influence perceptions and behaviour. Studies emphasise effects on trust, satisfaction, and intention.

3.2.2.4. Cluster 4: broader roles of GenAI in tourism services and education.

This cluster (Yellow– coloured) highlights diverse applications of GenAI (e.g. ChatGPT) from value co-creation to pedagogical innovation. Demir et al. (2023) demonstrate ChatGPT’s role in personalised service design, whereas Dalgıç et al. (2024) examine its contribution to tourism education.

3.2.2.5. Cluster 5: GenAI-driven service recovery.

This cluster (purple-coloured) examines

ChatGPT’s role in service recovery. Studies show that

AI-generated responses can be as effective as human ones, but awareness of AI authorship may reduce customer acceptance (algorithm aversion). These findings highlight both opportunities and perceptual risks.

3.2.2.6. Cluster 6: personalization and creative capabilities.

This cluster (light blue–coloured) captures GenAI’s creative and experiential potential. Miao and Yang (2023) show how text-to-image tools enrich anticipation and visual engagement, while Wang (2025) illustrates the role of GenAI in delivering deeply personalised hotel experiences.

3.2.2.7. Cluster 7: practical implications and organisational transformation.

This cluster (light- brown coloured) reflects broader practical and organisational impacts of GenAI. Shin et al. (2025) show how ChatGPT simplifies complex travel planning, while Keiper (2023) highlights its role in event management education.

3.3. Results from qualitative content analysis.

A qualitative content analysis of the 83 studies was conducted following Braun and Clarke’s (2006) framework (see Web Appendix 6 for procedural details). Constructs related to GenAI were identified, coded, and grouped into seven themes, which were organised into four analytical levels: micro (individual users), meso (organisational actors), macro (sector-level perspectives), and meta (education and research) (Web Appendix 7).

3.3.1. Thematic synthesis (ABCDE + D framework).

Using the extended ABCDE + D framework proposed in this study, the reviewed articles were systematically analyzed and structured. This framework guided the organisation of findings across seven themes, which were further mapped onto four stakeholder groups: customers (Themes 1, 2, 4, and 5), organisations (Themes 3 and 6), employees (Theme 3), and education/academics (Theme 7) (see Table 1).

3.3.1.1. Customer-related factors.

3.3.1.1.1. Antecedents.

Customer-related antecedents reflect travellers’ expectations, technological familiarity, and psychological readiness to use GenAI. As travel planning becomes more complex, tourists increasingly rely on GenAI for real-time assistance, personalisation, and convenience. Key predictors include perceived usefulness, ease of use, and trust, while accessibility features (e.g. multilingual and inclusive design) broaden adoption across diverse users. Prior experience and emotional readiness further strengthen willingness to rely on GenAI.

3.3.1.1.2. Barriers & Challenges.

Despite these enablers, adoption is constrained by concerns over privacy, misinformation, and lack of human warmth. Doubts about authenticity and accuracy remain key barriers in hospitality contexts. Technical issues such as complex interfaces reduce trust, while limited domain knowledge lowers satisfaction. Inaccuracies and hallucinations further reduce usefulness and may lead to disengagement, and preference for human interaction continues to hinder adoption.

3.3.1.1.3. Drivers.

GenAI adoption is driven by convenience, personalisation, and efficiency in decision- making. GenAI simplifies complex planning, reduces cognitive effort, and delivers tailored recommendations. Beyond functional value, human-like interactions enhance enjoyment and engagement, encouraging repeated use. Social influence further accelerates adoption, highlighting the joint role of functional and emotional drivers.

3.3.1.1.4. Decision.

Adoption decisions depend on whether perceived benefits outweigh concerns.

When GenAI is seen as useful and trustworthy, travellers use it for planning, comparison, and personalised itineraries. Conversely, concerns about security or accuracy limit use to low-risk tasks, although real-time and conversational features still enhance user confidence.

3.3.1.1.5. Effects.

GenAI generates mixed outcomes for travellers depending on how antecedents, drivers, and barriers interact. When usefulness, accessibility, and personalisation align, GenAI improves decision quality, reduces uncertainty, and enhances satisfaction and loyalty. However, concerns over authenticity, errors, and impersonal interactions can reduce trust and perceived value, distancing users from human-centred experiences (Kim et al., 2023 2024; Lu et al., 2024).

Overall, GenAI outcomes emerge from the balance between enabling and constraining factors: strong functional and emotional benefits enhance adoption and loyalty, while persistent trust concerns limit usage depth.

3.3.1.2. Business-related factors.

3.3.1.2.1. Antecedents.

Business-related antecedents highlight organisational capacity, technological readiness, and strategic orientation toward digital transformation. Firms with strong infrastructure, API integration, and skilled human capital are better positioned to deploy GenAI for personalised services and marketing automation. Efficiency goals and data-driven decision-making further drive adoption, while leadership support and innovation culture enable long-term integration.

References: Zhu and Yu (2025); Zhu et al. (2024); Dalgıç et al. (2024). Note: References shown are illustrative, not exhaustive. References: Author’s elaborations.

3.3.1.2.2. Barriers.

Despite its benefits, adoption is constrained by concerns over data accuracy and credibility. Internal factors such as financial constraints and weak digital strategies, along with regulatory uncertainty, increase hesitation. In hospitality, limited emotional intelligence and domain knowledge may lead to misuse, while unclear accountability reduces organisational trust.

3.3.1.2.3. Drivers.

GenAI adoption in T&H is driven by efficiency, creativity, and competitiveness. Cognitive drivers include faster decision-making, predictive analytics, and automation, while affective drivers involve enhanced employee creativity and reduced operational strain. Normative pressures from industry trends and customer demand for personalisation further accelerate adoption, highlighting the role of both technological ambition and market forces.

3.3.1.2.4. Decisions.

T&H firms integrate GenAI into marketing, operations, and customer service to gain competitive advantage. Key decisions include deploying chatbots and voice assistants for 24/7 multilingual communication, using recommendation systems and predictive analytics for demand forecasting, and adopting hybrid AI–human models to balance efficiency and authenticity. GenAI is also used for content generation, documentation, and data analysis, reflecting a shift toward AI-augmented service ecosystems.

3.3.1.2.5. Effects.

GenAI enhances service personalisation, decision speed, and efficiency in T&H firms

. Chatbots provide 24/7 multilingual support, reduce workload, and enable upselling, while dynamic content generation supports marketing and service innovation. However, excessive automation may undermine authenticity and human connection.

Overall, outcomes depend on the balance between strategic readiness and constraints. Strong infrastructure and leadership enable efficiency and competitive advantage, whereas unresolved governance and authenticity concerns may limit large-scale impact.

3.3.1.3. Employee-related factors.

3.3.1.3.1. Antecedents.

Employee-related antecedents reflect conditions enabling GenAI adoption.

Employees are motivated by automation of routine tasks, improved accuracy, and reduced cognitive load. Adoption also depends on innovation openness, digital literacy, and self-efficacy, while organisational support through training and clear roles strengthens readiness, enabling a shift toward more analytical and value-adding roles.

3.3.1.3.2. Barriers.

Despite these enablers, employees face challenges such as job insecurity and fear of displacement, along with digital skill gaps. Unclear legal and organisational guidelines create role ambiguity, while ethical concerns over bias and cultural insensitivity raise issues of fairness and inclusivity. Together, these barriers highlight technological and socio-ethical tensions in AI–human collaboration.

3.3.1.3.3. Drivers.

GenAI adoption among employees is driven by efficiency, accuracy, and improved decision-making. Employees view AI as a supportive tool that reduces repetitive tasks and enhances autonomy, leading to greater empowerment and performance satisfaction. Positive emotions such as confidence, creativity, and technological competence encourage use, while organisational support and peer learning reinforce sustained engagement.

3.3.1.3.4. Decisions.

Employees increasingly use GenAI for operational and communication tasks, including handling inquiries, automating data processing, generating service scripts, and producing marketing content. Writing, translation, and interaction tools improve multilingual communication, while managerial use supports data analysis and decision-making. However, increased reliance may lead to overconfidence and reduced professional judgment. Therefore, maintaining effective human–AI collaboration remains essential for ensuring accountability and service quality.

3.3.1.3.5. Effects.

GenAI generates both positive and negative outcomes for employees. It reduces routine workload, improves accuracy, and supports more analytical tasks, enhancing job satisfaction. However, automation and unclear human–AI boundaries may increase job insecurity and reduce confidence, while biased outputs can threaten fairness. Overall, outcomes depend on the balance between empowerment and insecurity: strong support and digital skills enhance performance, whereas limited training and unclear roles increase anxiety.

3.3.1.4. Academic and student-related factors.

3.3.1.4.1. Antecedents.

GenAI adoption in education reflects its growing role in teaching, learning, and research. Students benefit from personalised support, instant feedback, and 24/7 access, while educators value its ability to streamline content creation and assessment. Institutional readiness further supports adoption.

3.3.1.4.2. Barriers.

Despite its benefits, excessive reliance on GenAI may reduce critical thinking and creativity, while AI-assisted writing and data fabrication threaten academic integrity. These risks are amplified by unclear guidelines and weak monitoring.

3.3.1.4.3. Drivers.

Adoption is driven by improved learning quality, engagement, and efficiency.

GenAI enables rapid information processing, personalised learning, and better performance, while institutional initiatives strengthen AI literacy.

3.3.1.4.4. Decisions.

Educators and students increasingly integrate GenAI into learning workflows.

Teachers use tools like ChatGPT to adapt teaching strategies, create materials, and design assessments, while students use them for self-paced learning, translation, feedback, and simulations. This shift toward AI-supported platforms reflects a move to hybrid, data-driven pedagogy.

3.3.1.4.5. Effects.

GenAI enhances learning efficiency, engagement, and inclusivity, but overreliance may reduce critical thinking and increase misconduct risks.

Overall, outcomes depend on balancing learning benefits with ethical risks: with proper guidance and AI literacy, GenAI enhances higher-order learning; without governance, it may undermine academic integrity. A summary across stakeholder domains is presented in Table 1.

3.4. Results of lexicometric analysis.

A lexicometric analysis of the 83 articles was conducted using IRaMuTeQ (v0.8 alpha 7). The corpus included 15,237 word occurrences across 422 text segments and 2,560 distinct forms, with 49.26% hapaxes, indicating moderate lexical diversity (Web Appendix 8).

3.4.1. Similarity analysis.

The similarity analysis identified five main lexical clusters (plus two minor sub-clusters), revealing the conceptual structure of GenAI research in T&H (Figure 6). The central cluster, dominated by ChatGPT, is associated with terms such as trust, recommendation, traveler, and acceptance, highlighting user interaction and adoption. A second cluster, centred on study, includes framework, methodology, and relationship reflecting theoretical–methodological focus. The third cluster, linked to AI and generative, connects with technology, employee, and performance, emphasising organisational and operational applications. A fourth cluster, organised around tourism, research, and hospitality, experience, sustainability, and practice, indicating contextual grounding.

Finally, two smaller sub-clusters, one related to analysis and model, and another to data and survey capture the methodological diversity of the field, pointing to the coexistence of qualitative, quantitative, and hybrid approaches.

3.4.2. Factorial correspondence analysis.

To complement the similarity analysis, a Factorial Correspondence Analysis (FCA) with Hierarchical Descending Classification (CHD) was conducted to identify key lexical structures. The CHD dendrogram (Figure 7) reveals five classes: learning and application (e.g. learn, guest, content), reflecting engagement and GenAI-assisted services; quantitative research (e.g. data, survey, hypothesis), capturing methodological foundations; practical and theoretical insights (e.g. stakeholder, tourism); AI and hospitality context (e.g. AI, generative, hospitality); and perception and intention (e.g. intention, usefulness), representing user evaluation and behaviour.

The FCA (Figure 8) maps these relationships across two axes. Factor 1 (32.73%) contrasts methodological constructs with application and perception themes, while Factor 2 (27.53%) separates conceptual insights from user engagement. Overall, the structure shows a shift from methodological foundations to applied and perception-driven research.

Together, bibliometric and lexicometric analyses complement the ABCDE + D synthesis by reinforcing key patterns. Bibliometric results map the intellectual structure (RQ2), lexicometric analysis reveals semantic patterns (RQ3), and qualitative analysis integrates these insights into a stakeholder-oriented framework (RQ1, RQ3).

In turn, building on this integrated evidence, three implications emerge: trust-building mechanisms (e.g. transparency and hybrid AI–human models) for customers and organisations; training and clear role boundaries for employees; and governance and AI literacy in education to ensure responsible adoption.

3.5. Theoretical, contextual, and methodological analysis (TCM framework).

3.5.1. Theories.

A strong theoretical foundation enhances the practicality and relevance of research. The current review reveals that 57 out of 83 studies explicitly applied a theoretical framework, while 26 studies lacked a clear theoretical foundation (Web Appendix 9). The most widely applied theories were the Technology Acceptance Model (TAM), cited in eight studies, followed by the Theory of Planned Behaviuor (TPB) (used in six articles). In addition, Stimulus-Organism-Response (S–O–R) Theory, Innovation Resistance Theory and Media Richness Theory were employed in three articles each. This theoretical concentration largely stems from the dominance of adoption-focused studies (22% of the total sample), where TAM and TPB have been most frequently used to explain drivers and barriers of GenAI acceptance in T&H contexts. Full list of articles and associated theories are presented in Web Appendix 10.

3.5.2. Contexts.

The context of a study reflects where and how it is conducted. In this review, contexts are categorised by population and geography (Figure 9; Web Appendix 11). Most studies (72%; n = 58) focus on customers (tourists), while other stakeholders, particularly employees, remain underexplored. Geographically, studies cover 15 countries, with the highest concentration in

China (30%; n = 21), followed by the United States (26%; n = 18) and South Korea (17%; n = 12), with limited representation from other regions.

This distribution reflects strong AI investment, infrastructure, and university–industry collaboration in these countries. However, it should be interpreted cautiously, as reliance on Web of Science, SSCI-indexed journals, and English-language criteria may bias visibility toward dominant academic systems. Consequently, findings may primarily reflect digitally advanced, English-publishing contexts rather than a fully representative global landscape.

3.5.3. Methods.

Methods refer to the analytical approaches used to investigate research questions. Six methodological categories are identified (Web Appendix 12). Quantitative methods dominate (n = 55), followed by qualitative (n = 15), mixed methods (n = 8), and conceptual/review studies (n = 5). Quantitative studies mainly use surveys (n = 33) and scenario-based experiments (n = 21), while qualitative research relies on semi-structured interviews (n = 5) and secondary textual data (n = 4). Some studies adopt novel approaches, such as using ChatGPT as virtual interviewees. Mixed-method studies combine quantitative and qualitative techniques.

4. Future research agenda (RQ3).

Building on the thematic synthesis (Web Appendix 13), GenAI–T&H research is concentrated on consumer adoption and marketing, while organisational transformation and educational innovation remain underexplored. To address this imbalance, six priority domains are proposed (see Table 2).

4.1. New theoretical perspectives.

Despite drawing on 48 frameworks, the field remains fragmented and dominated by TAM and S–O– R, which overlook emotional, motivational, and organisational dimensions. Future research should adopt integrative, multi-level frameworks capturing cognitive, affective, and systemic mechanisms (see Table 2). Priority levels are assigned based on theoretical gaps, stakeholder imbalance, and strategic relevance.

4.1.1. Dual-path adoption and resistance models (High priority; under-theorized resistance).

Research on GenAI adoption in tourism has predominantly relied on TAM and UTAUT, which explain adoption but overlook resistance and leaving the motivational roots of rejection unaddressed. Behavioral Reasoning Theory (BRT; Westaby, 2005) fills this gap by uncovering the reasons for and reasons against using GenAI – capturing both motivational drivers and psychological inhibitors. Yet BRT alone lacks the predictive strength of TAM in modelling behavioural intention. Combining TAM’s predictive clarity with BRT’s reasoning-based structure therefore enables a dual-path approach that simultaneously models adoption and resistance. Incorporating Personality Trait Theory further strengthens this integration by explaining why individuals weigh benefits and risks differently.

4.1.2. Barriers and change dynamics in GenAI diffusion (High priority; process over time.

missing)

Despite growing interest in GenAI adoption, most studies approach diffusion as a static outcome rather than a dynamic process shaped by competing forces. Diffusion of Innovation Theory provides a macro-level view of innovation spread but underplays psychological barriers. Innovation Resistance Theory complements this by identifying functional and psychological constraints. However, neither theory conceptualises the dynamic tension between driving and restraining forces. Integrating these with Force Field Theory provides a dynamic view of how barriers evolve and how diffusion unfolds over time.

4.1.3. Communication trust and engagement mechanisms (Medium priority; marketing-rich.

but mechanism-light)

Marketing messages generated by GenAI rely on different cues that influence how consumers judge trustworthiness and authenticity. Signaling Theory explains how these cues such as tone the message appear human or AI-generated and shape consumers’ initial impressions. However, this theory overlooks the channel through which the message is delivered. Media Richness Theory fills this gap by showing that richer formats (e.g. interactive chatbots, or voice assistants) create greater clarity, and emotional engagement than simple text. Therefore, combining these theories can explain both what signals GenAI should use and how different communication channels influence consumer trust and engagement.

4.1.4. Organisational sensemaking and employee adaptation (High priority; employee lens.

underexplored)

Employee responses to GenAI remain underexplored compared with customer-facing outcomes. A useful way to advance this area is to combine Organisational Sensemaking Theory with Technology– Work Interaction Theory. Sensemaking explains how employees interpret and give meaning to AI-driven changes, but it pays limited attention to how technology affects daily tasks. Technology–Work Interaction Theory fills this gap by showing how the design and features of new technologies reshape work practices and professional identities. Integrating these theories allows exploration of how employees cognitively and behaviourally adapt to GenAI implementation.

4.1.5. Strategic agility and capability transformation (High priority; macro-level gap).

Strategic and industry-level insights on GenAI remain limited. A combined application of the Resource-Based View (RBV) and Dynamic Capabilities Theory can clarify how tourism firms convert GenAI tools into sustainable competitive advantages. RBV highlights the role of valuable and rare resources but assumes relatively stable environments. Dynamic Capabilities Theory complements this by explaining how firms sense opportunities, seize them, and reconfigure resources in rapidly changing conditions. Together, these perspectives offer a dynamic explanation of how GenAI capabilities can be developed, adapted, and leveraged to build long-term strategic advantage in tourism.

4.1.6. Knowledge creation and transformative learning (Emerging priority; education.

meta-level)

Educational and research applications of GenAI are expanding, but the field still lacks a cohesive theoretical foundation. Knowledge Creation Theory (SECI; Nonaka & Takeuchi, 1995) explains how organisations externalise and share tacit knowledge yet pays limited attention to individual cognitive transformation. Transformative Learning Theory complements this gap by focusing on reflection and critical dialogue that reshape learners’ frames of reference. Integrating these two perspectives offers a multi-level lens for examining how GenAI supports both organisational knowledge co-creation and deeper, reflection-driven learning in T&H education.

4.2. New research contexts (fill geographic and sectoral gaps).

GenAI–T&H research is geographically concentrated in the U.S., China, and South Korea. Future studies should expand to underrepresented regions (e.g. Southeast Asia, Africa, South America), where cultural, institutional, and infrastructural differences may reshape adoption patterns. For example, religiosity and ethical expectations influence trust in Muslim-majority contexts, while limited digital readiness alters adoption dynamics in emerging markets. These contexts also offer fertile ground for advancing research on Halal tourism, where AI-mediated personalisation must balance convenience with religious compliance, authenticity, and perceived integrity. Similarly, in emerging markets with lower digital readiness and infrastructural disparities, issues of access, capability constraints, and trust may shift the relative weight of antecedents and barriers within the ABCDE + D framework.

Current research also focuses on marketing and customer service, neglecting experiential domains such as gaming, entertainment tourism, and casino management, where personalisation raises unique ethical concerns. Moreover, the literature remains customer-centric, overlooking employees, managers, policymakers, and developers. Future work should adopt a multi-stakeholder perspective to capture diverse perceptions and governance needs.

4.3. New methods.

The review reveals a strong quantitative bias in GenAI–T&H research, with most studies relying on surveys and scenario-based experiments (Web Appendix 12). While these approaches enhance predictive modelling, they often provide static views of behaviour. Future research should diversify methods. First, longitudinal and process-based designs are needed to capture how perceptions, trust, and behaviour evolve over time. Second, qualitative approaches (e.g. interviews, ethnography, netnography) can uncover emotional, ethical, and cultural dimensions. Third, cross-stakeholder studies should compare perspectives across customers, employees, managers, and policymakers. Fourth, conceptual and theory-building remains crucial to integrate fragmented constructs and advance theory development.

Taken together, these directions form a structured research trajectory. Customer-focused domains are suited to experimental and survey methods, while employee and organisational themes require qualitative and longitudinal approaches. Strategic transformation calls for firm-level and multi-level analysis, and educational domains benefit from mixed-method and design-based research. The identified domains vary in theoretical maturity and complexity, forming two broad trajectories. Near-term studies should focus on empirically tractable areas such as adoption mechanisms, trust dynamics, and employee adaptation, while long-term research should address strategic transformation and educational change through longitudinal and multi-level approaches.

As summarised in Table 2, key research questions are outlined for each domain to guide future inquiry.

5. Implications.

Despite growing research, a comprehensive understanding of GenAI’s transformative impact remains limited. Building on prior work, this study offers implications for scholars and practitioners.

5.1. Theoretical implications.

This study addressed the three research questions through a comprehensive analysis of GenAI adoption in T&H. Firstly, by integrating Antecedents, Barriers & Challenges, Drivers, Decisions, and Effects with theories–contexts–methods (TCM), current research synthesises a fragmented literature into a cumulative programme. The proposed ABCDE + D–TCM framework provides a novel theoretical lens that bridges behavioural constructs with contextual and methodological dimensions.

Secondly, prior reviews in GenAI and tourism have mainly relied on frameworks such as TCCM, which provide structural overviews but do not capture adoption as a process linking antecedents, barriers, drivers, decisions, and effects. By extending the ABCDE logic with a Decisions dimension and integrating it with TCM, the ABCDE + D–TCM framework shifts the field from predictor-based to process-oriented, multi-level analysis. This reveals a concentration on antecedents and intentions, while decision pathways, barrier resolution, and long-term effects remain underdeveloped, highlighting opportunities for cumulative theory building.

Thirdly, the TCM-based analysis reveals a strong reliance on behavioural models such as TAM and UTAUT, with most studies focused on intention-based explanations. Advancing the field requires integrative theorisation that links behavioural reasoning with communication, organisational sensemaking, and capability development perspectives.

Fourthly, beyond theoretical integration, this review highlights critical contextual blind spots that constrain theory development in the GenAI–T&H domain. Empirical evidence remains heavily concentrated in technologically advanced regions, such as the U.S., China, and South Korea limiting generalizability to emerging tourism economies with distinct digital and cultural environments. Addressing these gaps requires expanding theoretical testing across diverse geographical, cultural, and sectoral contexts, including Halal, gaming, and entertainment tourism, where behavioural, ethical, and regulatory dynamics may differ substantially.

Finally, methodological limitations also constrain theory development. The dominance of cross-sectional quantitative designs provides predictive but static insights. Advancing theory requires longitudinal, qualitative, and multi-stakeholder approaches to capture evolving interpretations and institutional dynamics.

5.2. Practical implications.

This review provides actionable implications for managers, employees, and policymakers in T&H.

First, the findings indicate that the value of GenAI for customers extends beyond search efficiency or recommendation quality. Trust, transparency, and perceived authenticity remain central. Managers should complement technical investments with trust-building practices such as human-in-the-loop validation, transparent disclosure of AI-generated content, and clear data-use communication. These measures reduce misinformation risks, strengthen credibility, and alleviate customer concerns.

Second, businesses should adopt hybrid service models rather than full automation. Competitive advantage arises when GenAI complements human roles, not replaces them. Managers should delegate routine tasks to AI while reserving human effort for high-contact, empathy-driven interactions. This balance maintains efficiency, preserves authenticity, and reduces negative customer perceptions.

Third, GenAI is reshaping skill requirements across the T&H workforce, creating both empowerment and stress as job demands evolve. Organisations should strengthen AI literacy, provide targeted digital training, and establish feedback mechanisms to track employee concerns. Such practices help reduce techno-stress and fears of displacement, while fostering a workplace culture in which GenAI is viewed as an enabler rather than a threat.

Fourth, the findings highlight critical ethical and regulatory challenges, especially around privacy, accuracy, and bias. Policymakers and managers should therefore work together to define clear standards for responsible AI use, including data governance rules, bias-mitigation procedures, content-verification protocols, and consumer-protection guidelines. Such governance mechanisms are essential for maintaining public trust and reducing reputational or legal risks.

Finally, GenAI has strategic implications for innovation and competitiveness in T&H. Firms that integrate generative tools into service design, marketing, and personalisation can achieve faster differentiation. Managers should therefore treat GenAI as a strategic capability rather than a purely operational tool.

6. Limitations.

Despite the rigour of this study, some limitations are to be acknowledged. First, the review relied solely on the Web of Science database. While WoS is widely recognised for its reliability and comprehensive coverage of high-quality journals, relevant studies indexed in other databases (e.g. Scopus, Google Scholar) may have been excluded. Second, although the keyword strategy was carefully refined through prior reviews and pilot testing, the fast-evolving nature of GenAI means that some emerging terms may not yet be fully captured. Finally, as with any systematic review, our synthesis is constrained by the scope, quality, and reporting of the primary studies included. These limitations provide opportunities for future reviews to adopt a broader database coverage and continuously update the scope as the field evolves.

7. Conclusion.

The principal contribution of this study is the introduction of the ABCDE + D–TCM framework as a process-based, multi-level approach to understanding GenAI adoption in T&H. By integrating antecedents, barriers, drivers, decisions, and effects with theories–contexts–methods (TCM), the framework clarifies how adoption unfolds across stakeholders, moving beyond intention-focused explanations (Figure 10).

Drawing on a systematic review of 83 articles using the SPAR-4-SLR protocol and a four-method approach (systematic, bibliometric, qualitative, and lexicometric analyses), this study synthesises fragmented insights into a coherent analytical structure. Rather than replacing established approaches such as ADO or TCCM, the ABCDE + D–TCM framework extends them by explicitly modelling the decision stage and highlighting where tensions between drivers and barriers become consequential. In doing so, it provides a structured foundation for cumulative and multi-level theorising in GenAI–T&H research. In doing so, the study addressed three central questions concerning GenAI’s impact, the state of art in the field, and future research priorities.

Findings show that GenAI research in T&H is growing rapidly but remains uneven. Adoption is driven by convenience and personalisation but constrained by privacy and authenticity concerns. Methodologically, the field is dominated by quantitative, TAM-based studies, while contextually it is concentrated in technologically advanced countries with a strong customer focus.

Future research should expand to underrepresented regions and stakeholders, including employees, managers, and policymakers, to better address governance, ethics, and accountability. Methodologically, longitudinal, qualitative, and mixed-method approaches are needed to capture dynamic adoption processes. Overall, the ABCDE + D–TCM framework provides a shared foundation for more coherent, cumulative, and ethically grounded research in GenAI–T&H.

CRediT: Mehrgan Malekpour: Conceptualization, Formal analysis, Methodology, Software, Supervision, Visualization, Writing – original draft, Writing – review & editing; Oswin Maurer: Conceptualization, Methodology, Supervision, Writing – review & editing; Hatice Kizgin: Conceptualization, Methodology, Supervision, Writing – review & editing; Fevzi Okumus: Conceptualization, Methodology, Supervision, Writing – review & editing.

Disclosure statement

No potential conflict of interest was reported by the author(s).