The Paradox of Personalised Travel: The More AI Recommends, the More Valuable Human Expertise Becomes

Travel increasingly begins not with a printed guidebook or a travel agency, but with a question addressed to artificial intelligence: “Where should I go in October?”, “Which neighbourhood in Lisbon would suit me?”, or “Plan three days away from the main tourist attractions.”

This shift promises a simpler and more personalised travel experience. AI can analyse our preferences, remember our habits and compare vast amounts of information within seconds. Yet a paradox is emerging: the more capable recommendation systems become, the more visible the value of human judgement becomes.

This is one of the central questions explored in my thesis: how can artificial intelligence and trusted local expertise work together to create travel discovery that is more personalised, authentic and culturally meaningful, beyond mass tourism?

Personalisation is becoming scalable

Agentic AI systems can retain context, remember user preferences and carry out several stages of a travel journey. In a study conducted by McKinsey and Skift involving 1,002 travellers and 86 travel industry executives, more than 90% of surveyed travellers expressed some level of confidence in the accuracy of AI-generated travel information. However, only 2% were willing to give AI full autonomy to make or modify bookings without human supervision [1].

This difference is fundamental. Receiving a recommendation is one thing; allowing a system to choose, book and spend money on our behalf is another.

A study involving 319 tourists in the United Arab Emirates reached a similar conclusion. Personalisation positively influenced the adoption of AI-powered tourism technologies, but trust played a central mediating role. Ethical practices, transparency and perceived value therefore matter as much as the technical quality of the recommendation [2].

The challenge is no longer simply to produce a relevant answer. It is to make that answer credible, transparent and acceptable.

The risk of personalisation creating uniformity

We often associate personalisation with greater diversity. However, a system may personalise the wording of a recommendation while repeatedly drawing from the same limited collection of destinations.

A study published in 2026 compared 420 travel recommendations generated by ten AI systems in response to fourteen structured travel queries. Despite the apparent variety of destinations mentioned, the answers frequently converged around a relatively narrow set of iconic or semi-iconic places [3].

A recommendation can therefore feel personal while still reproducing collective popularity. When AI models are trained primarily on information that is widely available, highly referenced and frequently consulted, they may reinforce the visibility of places that are already visible.

Yet travel is not only about receiving the most accurate answer. It is also about encountering something we would not have known to ask for.

Research on recommender systems has shown that accuracy alone is not sufficient to define a good recommendation. Diversity, novelty, coverage and serendipity must also be considered [4]. A genuinely personalised travel experience should not merely confirm a traveller’s previous preferences. It should also introduce relevant forms of surprise.

Personalisation becomes limiting when it confines us to an optimised version of ourselves.

Local expertise is not a database

Claiming that humans are naturally superior to machines would, however, be too simplistic. A local expert can also be biased, reproduce stereotypes or recommend what they know best rather than what genuinely suits the traveller.

Their distinctive value lies elsewhere.

A local expert does not simply know a list of addresses. They understand the rhythms of a city, its unspoken rules, the moments when a place changes character and the behaviours required to enter a community without disrupting it. They can interpret an imprecise intention and recognise when a theoretically perfect recommendation will not suit a particular person in a particular context.

A study published in Tourism Management, based on in-depth interviews with fifteen certified tour guides in Turkey, examines guides as intermediaries between travellers, local communities, tourism professionals and policymakers. Their role depends on reciprocity, trust, mutual benefit and balanced relationships between stakeholders [5].

This relational dimension transforms information into cultural mediation.

Authenticity, therefore, does not automatically reside in a “hidden” place. Once a secret address is distributed at scale, it can reproduce the same mechanisms of mass tourism it was supposed to avoid. The objective should not be to replace conventional tourist hotspots with a new list of supposedly exclusive hotspots. It should be to create more appropriate connections between travellers, contexts and local communities.

Traveller expectations already appear to be moving in this direction. According to research commissioned by Booking.com in January 2026 involving 32,500 travellers across 35 countries and territories, 43% planned to avoid overcrowded tourist destinations or attractions, while 42% intended to travel outside peak seasons [6].

Because this research was commissioned by an industry platform, it should be treated as a market indicator rather than independent academic evidence. Nevertheless, it points towards a growing desire to move beyond dominant travel routes and distribute tourism flows more responsibly.

The real model: complementary intelligence

The future of travel should therefore be neither fully automated nor dependent on a purely human service that is difficult to scale.

Research into human–AI collaboration defines complementarity as a situation in which the combination of human and artificial intelligence produces better results than either could achieve independently. It identifies two principal sources of complementarity: differences in available information and differences in capabilities [7].

Applied to travel, this logic becomes particularly powerful. AI can process large volumes of data, remember preferences, compare alternatives and reduce friction. Local experts contribute tacit knowledge, cultural interpretation, contextual judgement, trusted networks and human accountability.

AI may learn that I prefer quiet places, contemporary design and long conversations. It does not necessarily know which table, in which neighbourhood, with which person and at what moment will transform those preferences into a meaningful memory.

This is where the paradox lies. As recommendations become abundant, immediate and technically personalised, scarcity shifts elsewhere. What becomes valuable is no longer access to information, but access to judgement we can trust.

The next transformation in travel may therefore not involve removing the human intermediary. It may involve using AI to enable the right human intermediaries to exercise their expertise with greater precision, reach and relevance.

Sources

[1] McKinsey & Company and Skift, Remapping Travel with Agentic AI, September 2025. The report draws in part on surveys involving 1,002 travellers and 86 travel industry executives.
https://www.mckinsey.com/industries/travel/our-insights/remapping-travel-with-agentic-ai

[2] Inhyouk Koo, Umer Zaman, Hojung Ha and Shahid Nawaz, “Assessing the Interplay of Trust Dynamics, Personalization, Ethical AI Practices, and Tourist Behavior in the Adoption of AI-Driven Smart Tourism Technologies”, Journal of Open Innovation: Technology, Market and Complexity, Volume 11, 2025.
https://doaj.org/article/6d175ff0b6fd433598ef9d1983232849

[3] Maryam Najafi and Carlos Costa, “Digital Overtourism in AI Travel Recommendations: Evidence from a Comparative Analysis”, Current Issues in Tourism, published online in April 2026.
https://www.tandfonline.com/doi/abs/10.1080/13683500.2026.2654066

[4] Marius Kaminskas and Derek Bridge, “Diversity, Serendipity, Novelty, and Coverage: A Survey and Empirical Analysis of Beyond-Accuracy Objectives in Recommender Systems”, ACM Transactions on Interactive Intelligent Systems, Volume 7, Issue 1, 2016.
https://research.ucc.ie/en/publications/diversity-serendipity-novelty-and-coverage-a-survey-and-empirical/

[5] Ayse Sengoz, Tarik Dogru, Makarand Mody and Cem Isik, “Guiding the Path to Sustainable Tourism Development: Investigating the Role of Tour Guides within a Social Exchange Theory Paradigm”, Tourism Management, Volume 110, 2025.
https://doi.org/10.1016/j.tourman.2025.105162

[6] Booking.com, Travel & Sustainability Research 2026. The research was commissioned by Booking.com and conducted in January 2026 among 32,500 travellers across 35 countries and territories.
https://news.booking.com/bookingcoms-latest-travel-and-sustainability-research-reveals-unexpected-generational-paradox/

[7] Patrick Hemmer, Max Schemmer, Niklas Kühl, Michael Vössing and Gerhard Satzger, “Complementarity in Human–AI Collaboration: Concept, Sources, and Evidence”, European Journal of Information Systems, Volume 34, Issue 6, 2025.
https://www.tandfonline.com/doi/full/10.1080/0960085X.2025.2475962

For a deeper look at how trust shapes the entire travel journey, from initial inspiration to final booking, read my previous article: AI in Travel: From Inspiration to Booking, Trust Still Wins.