Stop Screenshotting Your AI Visibility. It Is One Deal of the Cards.

A hotel owner searches ChatGPT for “best boutique hotels in [city]”. Their property does not appear. The screenshot is sent to the agency, the marketing team, or the owner’s WhatsApp group as evidence that something is wrong.

Another hotelier does the same search and appears first. That screenshot is framed as evidence that an AI visibility programme is working.

Our data suggests both reactions are flawed.

Had either asked again an hour later, there was a four in ten chance a different hotel would have been named first.

We spent a month measuring this systematically. The Boutique Hotel AI Index put the same hotel questions to five AI platforms - ChatGPT, Gemini, Copilot, Google AI Mode and Google AI Overviews - across 100 destinations, fifteen times over: 28,600 answers in total.

On one day we ran the entire panel twice, an hour apart, changing nothing, to measure how much the answers move on their own. The answer is: a great deal. The same question returned the same number one hotel only 60.2% of the time. An hour of nothing happening moved four in ten top slots.

If that were the whole story, AI visibility would be little more than theatre and we could all go back to worrying about OTA commissions. It is not the whole story. Underneath the churning answers sits a remarkably stable pool: a third of the roughly 9,300 hotels the platforms named carry 88% of all the visibility, and that core barely changed across the month. Within any single destination, around ten hotels take 57% of all AI mentions.

So the answer changes constantly, while the pool it is drawn from does not. The platforms deal a fresh hand on every ask, from a deck that is smaller and more stable than the surface suggests.

Once you see that, three practical conclusions follow for hotel teams:

First, a single AI answer, screenshot or "AI rank" is not a measurement. Not because the tools capturing them are dishonest, but because the thing being captured does not hold still. Anything worth calling measurement counts presence across repeated asks. One reading is one deal of the cards, whether it flatters your property or not.

Second, there is no such thing as your hotel's AI visibility in the singular. The five platforms agree on the same number one hotel about 9% of the time, and over half the hotels in our data appear on only one platform. What guests are shown also depends on what they ask for: luxury searches behave like a settled consensus, while boutique and romantic searches are wide open and re-dealt on almost every ask. It moves again with how the question is phrased, and again with the country it is asked from. A hotel with guests on three continents has many AI visibilities, and averaging them into one score hides everything a marketer would act on.

Third, the durable objective is the pool, not the position. Chasing a number one slot that reshuffles hourly is a losing game. Getting named at all is surprisingly easy: roughly a quarter of all hotels in our data were named exactly once in the month, then never again. The game that matters is joining the persistent core that the platforms keep drawing from, and staying in it, measured as the share of repeated asks in which your property appears, per platform, per intent, per market.

The uncomfortable part for our industry is that the loudest AI visibility conversation right now is structured around exactly the wrong unit: positions, screenshots and single scores. Some of that is innocent enthusiasm. Some of it is commercial convenience. Either way, hoteliers deserve to know that the volatility being shown to them is mostly the system's own noise, and that the signal underneath it is measurable, slower moving, and far more useful.

The full analysis, including the methodology, the confidence intervals and every caveat, is openly published, and the study continues weekly. We would genuinely welcome other teams checking these numbers against their own data. That, rather than another screenshot, is how this conversation moves forward.