Cornell’s Center for Hospitality Research has published An Examination of AI in Travel Planning Across Traveler Spending Segments, by Young Jang and Christopher Anderson. The study surveyed 1,029 U.S. travelers through Prolific and grouped them by their approximate nightly accommodation spend in a large U.S. city: Budget under $170, Premium $170 to $250, Aspirational $251 to $350, and Luxury $351 and above.
What Cornell found
Respondents were asked what stops them from using AI in travel planning. Three concerns dominate.
Accuracy of the information, cited by more than 60 percent. Lack of transparency in how the recommendation was generated, above 40 percent. Recommendations that feel too generic or insufficiently tailored, also above 40 percent. Data privacy, algorithmic bias, and a perceived lack of nuance follow behind.
The pattern is worth stating precisely, because it is easy to overstate. The three largest barriers are not objections to AI in the abstract. They concern confidence in the information and recommendations it produces.
What Cornell recommends
The report’s managerial prescriptions are segment-aligned and specific.
For Budget travelers, AI interfaces should lead with a value-first display, prioritizing price comparison and deal alerts. For Premium travelers, seamless organization and efficient information retrieval. For Aspirational travelers, discovery, curation, and personalization inside a visually engaging and secure interface. For Luxury travelers, AI positioned as an intelligent enabler that handles back-end logistics and rapid information retrieval while human advisors take the high-touch work.
The report’s summary line is that hospitality firms should replace one-size-fits-all bots with prescriptive, segment-aligned approaches. That is sound advice. It is also worth noticing what it assumes.
The assumption underneath
Every one of those recommendations assumes the hospitality company can influence the AI experience the traveler is using.
Design the display. Tune the personalization. Build the handoff to a human advisor. Position AI as an enabler. Each is a decision a hotel or a brand makes inside a system it operates.
That is a real and legitimate scope. It is also a boundary. The report examines how hospitality organizations should design AI experiences travelers will trust. It does not examine what happens when the AI experience belongs to somebody else.
The upstream question
Cornell does not determine what causes the three leading concerns, or which part of the recommendation process produces them. Transparency in how a recommendation was generated could refer to source provenance, to ranking criteria, to model logic, to commercial influence, or to explainability. The study does not isolate which, and it does not need to. It establishes the traveler-side problem, and it establishes it well.
The next question is upstream. What determines the information an external AI system has available when it forms the recommendation in the first place?
That question sits outside the report’s scope entirely. It is the question our own work calls Knowledge Formation Optimization (KFO), and it is a different inquiry from the one Cornell conducted, not a conclusion drawn from it.
AI does not have to book the room
Two findings in the report explain why the upstream question matters economically.
Across the sample, AI Chatbots and Assistants ranked fourth in how often they appeared among respondents’ top three travel planning tools, behind Search, Reviews and Review Websites, and Official Hotel Websites. The report notes that the four segments produced largely consistent rankings. AI is already in the consideration set. Taken together, Cornell’s data suggest its strongest role is upstream of the transaction.
That shape holds across the use-case data. Respondents were asked whether they would feel comfortable using AI for specific parts of the planning and booking process. Comfort was generally highest around discovery and information tasks, while booking activities and tours drew the lowest comfort across every segment. The report characterizes its Luxury respondents as comfortable with rapid, fact-based research while preferring human involvement for complex logistical coordination and the final orchestration of a trip.
This is stated comfort, not tracked behavior, and it should be read that way. But it points somewhere specific.
An AI system can shape which properties a traveler considers without ever completing a transaction. That shortlist can form before the booking environment loads. A hotel-controlled interface, however well segmented, operates after that point.
The Luxury label deserves a qualification
The study’s top spending bucket is $351 or above per night, with 89 respondents.
The issue is not that high-end travelers are absent from the sample. Someone spending $900 a night falls inside that bucket. The issue is that the bucket has no ceiling. A traveler spending $375 and a traveler spending $1,500 are classified identically.
Cornell did not set out to isolate the top of that range, and segmenting by accommodation spend is a defensible design choice. It does mean the study cannot tell us whether the upper luxury market behaves differently from travelers sitting just above the threshold. For independent luxury hotels and resorts operating at the upper end of the market, that distinction matters.
Cornell measured the person asking
There is one more distinction worth drawing, and it is not a criticism.
Cornell measured travelers. Self-reported usage, stated comfort, perceived barriers, 1,029 respondents. That is the demand side of the interaction.
The AGR Luxury Hotel AI Visibility Index measured the other side. Ten questions travelers actually ask were put to ChatGPT, Google AI Mode, and Gemini in each of six U.S. luxury markets: New York City, Los Angeles, Chicago, Miami, Maui, and Napa Valley. Every query was run by hand, logged out, with zero account state, in a fresh private browsing window, from a New York origin, on the consumer web surface a traveler uses rather than an API. All 180 answers were captured on July 29, 2026, in a single day, and preserved. Those answers contained 824 ranked hotel recommendations, counting each ranked hotel in each answer as one slot.
That is one day of capture and should be read as one day, not as a trend line. What it records is which properties AI systems named when a traveler asked which hotels to consider. Not what travelers say they would do. What the machines say when nobody is asking them to be diplomatic.
These are not competing studies. They examine opposite ends of the same exchange. Cornell measured the person asking. The Index measured the machine answering. Read together, they expose a part of the travel decision that neither traveler-attitude research nor model-output research describes on its own.
What this leaves open
Cornell’s research explains why travelers hesitate to trust AI recommendations. It is careful, well constructed, and the segment-level framing is genuinely useful.
It also leaves the harder question standing. Hotel consideration can now form inside AI systems the hotel does not control: the model, the retrieval process, the citations, the competing properties, and the synthesized answer. A hotel cannot resolve that by redesigning its own AI interface.
What those systems know about a property before a traveler ever reaches the hotel’s website is a separate problem, and it is the one worth solving next.
Source: Young Jang and Christopher Anderson, “An Examination of AI in Travel Planning Across Traveler Spending Segments,” Cornell Center for Hospitality Research. The report discloses that Curacity supported its data collection.
