The State of AI Representation in Travel: What the Published Evidence Shows

Travellers have started asking AI assistants the questions they used to type into a search box, and the assistants answer with confidence, detail and a short list of names. For most travel organisations, nobody has yet read those answers. This brief draws together what the independent research published so far tells us about who is asking, what they are being told, and where the assistants get their information.

Who is asking

The clearest Australian figure comes from a Compare the Market survey of more than 1,000 adults: 28.8 per cent said they use AI tools to plan travel, with destination recommendations (11.5 per cent) and accommodation (9.4 per cent) among the leading uses.¹ That is close to three in ten adults, and the uses they name are recommendation uses — where to go and where to stay — rather than administrative ones.

The point for an operator, a destination or a hotel is that the assistant is now sitting at the top of the funnel for a meaningful share of demand, and it is making a recommendation before the traveller has visited anyone’s website.

What they are told

Two independent studies have tested the answers directly.

Marti-Ochoa and colleagues put the same hotel-recommendation questions to five assistants — ChatGPT, Gemini, Claude, DeepSeek and Grok — across the ten most-visited cities in the world. Every one of the five understated average nightly prices, in some cases by a wide margin, and only two of the five disclosed the sources behind their recommendations.²

Nicolas Sitter’s AI Hotel Landscape 2026 ran 2,500 prompts across 25 cities, including Sydney and Melbourne, for a total of almost 20,000 answers. Among its findings: Perplexity routed roughly a quarter of its links to online travel agencies and metasearch sites rather than to hotels, cited Tripadvisor in more than 95 per cent of answers, and independent three-star hotels drew close to 30 per cent of their citations from OTAs.³

Taken together, the two studies describe the same shape. Assistants recommend readily and describe in detail; the detail is frequently out of date on the facts that matter commercially, most obviously price; and the organisation being described is rarely the source of its own description.

Where the answers come from

Assistants do not consult a single index. They retrieve a small set of documents at the moment of the question, extract facts from them and synthesise an answer, weighting sources by how consistent and authoritative they appear. Profound’s analysis of 680 million citations shows how concentrated that retrieval is: each assistant leans on a small set of community, reference and review domains, with Reddit alone accounting for close to half of Perplexity’s top-ten citation share.⁵

For travel specifically, Sitter’s data shows review platforms and booking intermediaries dominating the citations behind hotel answers, with the hotel’s own site among the least-cited sources.³ The practical consequence is that the “facts” an assistant states about an organisation are usually a third party’s copy of those facts — a listing, a review, an article — at whatever age that copy happens to be.

Why it matters

It would be possible to treat all this as a curiosity if travellers shrugged off wrong answers. The peer-reviewed evidence says they do not. Kim and colleagues, publishing in the Journal of Travel Research, found that a single inaccurate answer from a generative AI assistant measurably lowered participants’ intention to visit and their trust — and that the damage was greatest when the error was prominent in the answer.⁴ The traveller does not distinguish between the assistant’s mistake and the organisation’s.

What organisations can do

The research points to a discipline rather than a technology purchase. The organisations described accurately are the ones whose key facts — rates, inclusions, hours, access, policies — are stated the same way everywhere an assistant might read them, whose own pages state those facts plainly enough to be quoted, and who know which third-party sources are shaping their answers and keep them current.

That starts with a baseline: reading the answers assistants give today, for the questions travellers actually ask, and tracing each fact back to its source. Everything else — ownership, correction, monitoring — follows from knowing where you stand.

Tripicon’s External AI Representation Review applies exactly this method to a named organisation across the major assistants, with findings delivered within two weeks.