Only 6% of Hotels Show Up When You Ask AI Where to Stay
Skift found that 94% of hotels never surface in AI trip planning answers. That concentration of demand is quietly moving what you pay.
Ask an AI assistant where to stay in Lisbon and it will hand you four hotel names. Lisbon has more than a thousand places to book a room. The distance between those two numbers is the most interesting thing happening in travel right now, and almost nobody is pricing it in.
Skift Research put a figure on the gap in its State of Travel 2026 report. Only 6% of hotels surface in AI search. The other 94% never appear in the answer to a traveller’s first question. Over the same stretch, 62% of travellers now say they are familiar with AI trip planning tools.
A shrinking answer set and a growing audience. That combination does something to price, and it is already showing up.
What being invisible actually means
Type “where should I stay in Lisbon for four nights, walkable, under $150” into a chat window and the model is not querying live hotel inventory. It is assembling an answer out of what it has read. Hotel pages, review aggregators, magazine features, forum threads, city guides.
A hotel that has been written about a lot exists inside that body of text. A hotel that has never been written about does not exist there at all, no matter how good it is or how many rooms it has sitting empty tonight.
That is a harsher filter than the one search engines apply. Google indexes nearly everything and ranks it, so a determined traveller can reach result number 40. An AI answer names three to six properties and stops. There is no page two to click.
Why the same hotels keep coming back
Four things push a property into that 6%.
Independent coverage. Being named in editorial content, guidebooks, and neighbourhood guides carries far more weight than anything a hotel publishes on its own site. Self-description barely registers.
Extractable facts. A page stating “double rooms in the Chiado location run $140 to $190 in shoulder season” gives a model something to lift and cite. A page promising “competitive rates in a prime location” gives it nothing at all.
Brand scale. Chains publish consistent, crawlable property pages across hundreds of locations, all following the same template. An independent hotel publishes one page and hopes.
Review density, not review score. A property with 4,000 reviews generates vastly more source text than one with 90, even when the smaller property scores higher. We went through how hotel review scores actually work recently. This is the same distortion viewed from a different angle.
The part that lands on your bill
Demand that used to spread across a wide field of properties now lands on a narrow one. When millions of people ask a near-identical question and receive overlapping shortlists, those specific hotels fill earlier in the booking curve. Filling earlier is exactly what lets a revenue manager hold higher rates for longer.
The wider market already has room to push. US hotels ran 69.7% occupancy in July 2026 at an average daily rate of $171.74, with RevPAR up 8.2% year over year, according to CoStar and STR data. Concentrated demand hands the properties inside that 6% more pricing power than the market average suggests.
The 94% are not cheaper because they are worse. They are cheaper because far fewer people are being shown them. That is a visibility discount, and you can collect it.
How to use AI search without paying the visibility premium
Use these tools for the question they are actually good at, then switch.
AI is strong at neighbourhood reasoning. “Which part of Lisbon is quiet after 10pm but still a walk from dinner” rewards synthesis across thousands of scattered opinions, and synthesis is the thing these models do best. Ask that question and the answer will be useful.
They are weak at price and availability. A model does not know what a room costs tonight. Ask anyway and you get figures pulled from articles written eight months ago, quoted back at you with total confidence.
So run it in two steps. First, ask the AI to narrow the neighbourhood and the type of property you want. Second, take that neighbourhood to a booking platform and compare live rates across every hotel in it, not just the three that happened to get named.
We built Best around that second step. Search a destination and you see the lowest available rate across the properties, sorted by price rather than by how much editorial coverage each one has accumulated. If an AI just handed you three names for a city, it is worth twenty seconds to see what the other forty look like.
The squeeze this puts on independents
Small operators feel this first. A 30-room hotel cannot out-publish a chain, and it has no marketing budget to buy its way into the training data. Its rooms are good and its price is fair and the question never reaches it.
Some of them are responding by joining soft brands, which is one reason independent hotels keep signing with the big groups. Distribution used to mean shelf space on booking sites. Increasingly it means existing in a sentence that a model can generate.
For travellers, the practical read is simple. The interesting, smaller, better-value properties are disproportionately the ones an AI will not mention. Finding them takes one extra sort, and it is usually the sort by price.
What we would actually do this autumn
Ask the AI where in the city, not which hotel. Get the neighbourhood, the walk times, and the honest note about which area is loud on weekends. That part is genuinely good and would have taken an hour of forum reading a few years ago.
Then ignore the hotel names entirely and price the whole neighbourhood yourself. On a four-night stay in a European city this autumn, the spread between the most-cited property and a comparable one two streets away is routinely $30 to $60 a night. Over four nights that is a very good dinner.
Rates are also moving underneath all of this. Our note on why fall 2026 hotel prices are running above summer covers the timing side.
Questions we get about this
Does asking an AI in a different way surface different hotels? Slightly. Naming a budget, a neighbourhood, and a property type will pull in a wider set than a generic question. But you are still drawing from the same 6%. Rephrasing widens the sample a little. It does not change the pool.
Are the hotels AI recommends bad recommendations? No. They are usually solid, well-run properties, which is part of why they got written about. The problem is not quality. The problem is that you are choosing from 6% of the options and paying a premium for the privilege.
Can a hotel pay to appear in AI answers? Not directly, in the way search ads work. But money spent on press coverage, partnerships, and content ends up in the training data eventually, so scale still buys visibility. It just takes a longer route.
Will this get better? Probably, as AI tools connect to live inventory rather than static text. Several are already testing it. Until that is standard, treat any hotel an AI names as a starting point rather than a shortlist.
How much does the visibility premium cost? There is no clean published figure yet, and we are not going to invent one. What we can say from watching rates is that the most-cited property in a neighbourhood is almost never the best-priced one in it.
Images: Hero by ArtHouse Studio via Pexels. Hotel lobby via Pixabay. Reception desk by Andrea Piacquadio via Pexels. Used under licence.
