What the paper studied
Previous reviews of generative AI (GenAI) in tourism mostly mapped publication trends or discussed single applications. This review integrates methods, theories, benefits, risks and personalization in one analysis. The authors searched Web of Science and Scopus in June 2026, screened 395 records down to 119 eligible studies, and added six more through reference checking for a final set of 125 journal articles. Two researchers independently coded each study by method, theory, application context, benefits, constraints and personalization mechanism. The studies span 70 journals, led by Tourism Management (11), Journal of Travel Research (9) and Current Issues in Tourism (7).
Key findings
- Quantitative methods dominate. Survey-based structural equation models (SEM/PLS-SEM) and scenario experiments lead; 41 of the 125 studies combine more than one method. Qualitative and computational approaches are less common.
- Adoption theories still anchor the field. The most used frameworks are stimulus–organism–response (12 studies), the technology acceptance model (10), social exchange theory (8), UTAUT2 (7) and innovation resistance theory (5). More than half of the studies combine more than two frameworks.
- Four benefit categories. GenAI integrates fragmented travel information into personalized service, lowers decision-making costs, strengthens marketing communication and brand value, and supports operations and sustainability (for example automated replies, review management and demand forecasting).
- Four constraint categories. These are technical reliability and "AI hallucination", data security and privacy, copyright and ethical disputes (including homogeneous AI imagery that reinforces destination stereotypes), and weak industry adoption, with shortages of AI skills and organisational readiness.
- How personalization works. GenAI personalizes through situational adaptation, role-based interaction (the review cites six roles AI plays in trip planning: mentor, pathfinder, partner, acquisition, butler and implementer) and cognitive activation, or "Aha moments". It works best when users can see and adjust how recommendations are made.
- A TPO framework. The proposed Technology–Psychology–Organization framework links GenAI properties (multimodality, hallucination propensity, degree of personification), tourist responses (trust, perceived control, emotional resonance) and organisational implementation (staff capability, ethical governance, prompt-task fit).
Why it matters for hospitality
The field has moved beyond asking whether travellers will use GenAI toward questions of hallucination, privacy, authenticity and governance. The area that remains least studied is enterprise-level implementation: change management, workforce reskilling, governance frameworks and HR policy. That is exactly where hotels and tourism businesses need evidence most.
Practical takeaways
- Signal uncertainty in AI-generated information (for example confidence scores or source labels), calibrated to the risk of the task, and keep human review for high-stakes decisions such as bookings.
- If you use AI-generated review summaries, keep them balanced and flag possible positivity bias so guests can judge for themselves.
- Match prompts to tasks ("prompt nudging") to improve the quality of AI recommendations.
- When AI recommends local attractions or writes marketing copy, emphasize perceived benefits and trust, and set how human-like the AI appears together with how clearly it is disclosed, rather than maximizing either output volume or human-likeness.
Limitations
The review covers English-language journal articles indexed in Web of Science and Scopus, and its keywords centre on mainstream GenAI terms such as ChatGPT, so work on multimodal, affective or explainable AI may be missed. With 125 studies it offers limited cross-cultural comparison, and the authors present the TPO framework as a research agenda rather than a validated causal model.