Introducing the Boutique Hotel AI Visibility Index

Updated 7 July 2026. Since first publishing, we have rebuilt our analysis dataset and re-checked every initial figure against it. We have added a methodology section. The original run observations are unchanged.

For hotel marketers, visibility in AI responses has become an increasingly important part of discussions with hotel owners and management teams. Yet there is still very little practical data about how AI platforms select, rank, and cite hotel recommendations, especially in the boutique and independent hotel space.

Saying “we are not 100% sure” or “visibility depends on a variety of factors” may be technically true, but it does not give hotel teams much to work with. One-off screenshots and isolated examples are also difficult to build on.

So we decided to run our own research project: The Boutique Hotel AI Visibility Index.

What We Are Measuring

Our aim is to track hotel visibility, volatility, and discernible trends in AI-generated hotel recommendations.

Which hotels are recommended? Which ones appear repeatedly? Which platforms are more consistent? Which sources are cited? And how much does the answer change when the same question is asked again?

These are the kinds of questions hotel marketers and management teams will need to understand as AI-assisted hotel discovery becomes the norm.

For this first phase, we are running the same set of hotel recommendation prompts across 100 destinations worldwide. The destinations include a mix of cities, islands, resort areas, and well-known boutique travel regions.

Each run includes three prompt categories: boutique, luxury, and romantic. These were chosen because they reflect the type of properties Kollective works with most often: independent boutique hotels.

Is that setup shaped by our own area of focus? Yes. But it is also where we have the clearest frame of reference. Initial results suggest the findings are likely to be useful for hotels and marketing teams across much of the hospitality sector.

The study covers five AI platforms: ChatGPT, Copilot, Gemini, Google AI Overview, and Google AI Mode. These platforms were chosen based on what we currently see across our hotel clients’ analytics data, combined with the AI search platforms most visible in the wider travel search conversation.

Taken together, that gives us 1,500 individual recommendation queries per run.

One thing we are deliberately not doing is naming individual hotels. This study is about the behaviour of the platforms, not about producing a league table of properties. Published findings will report patterns, not names.

The Study Setup

The core of the study is repetition. We are running the same 1,500 queries 15 times across June and July 2026, using the same prompt structure, from a fixed US locale, in logged-out sessions with web search enabled. The point is to avoid changing our own inputs and then pretending that any movement came from the platforms.

Around that core, the study has grown several supporting arms:

• Google search comparison. On selected run days we also collect the full Google search results for keyword versions of the same queries, pulled the same day as the AI run, so we can ask whether the hotels AI recommends are the same ones Google ranks.
• Location testing. On selected days we run a subset of 60 queries from three additional locations (the UK, Australia and Singapore) to measure how much the answer depends on where you ask from.
• A same-day baseline. AI platforms do not answer the same question the same way twice. To separate genuine change over time from ordinary variation, we repeat the full query set twice on the same day under identical conditions and use the difference as our baseline.
• Answer capture. Every response is stored as a full page capture, so any number we publish can be traced back to the actual answer a platform gave.

Once the runs are complete, we will also review the public websites of hotels that appear frequently in the data for structured data, llms.txt adoption and other technical signals, to see whether the common advice about AI visibility matches what the platforms actually surface.

The Early Data Is Already Intriguing

Our first runs have shown a high level of variation. In Run 1, more than 4,000 distinct hotels were named and close to 3,000 different domains were cited. Run 2, a day later, named roughly 2,800 hotels that had not appeared the day before.

Part of that jump turned out to be a finding in its own right. Google AI Mode alternates between two answer formats: a short prose answer naming around six to ten hotels, and a card-style list of up to twenty. Which format you get changes how many hotels the answer contains, and Run 2 happened to return far more of the long format. Even comparing at equal list depth, though, more than 1,600 hotels were new on day two. However you count it, AI visibility is not a fixed position.

Those numbers are early observations, not final conclusions. The full analysis will use the complete dataset, and where a defensible methodological choice changes a number, we will publish the range rather than the most convenient point.

Methodology

AI answer data is collected via Rankscale and verified against full page captures. Google search results are collected via SE Ranking. Hotel names are matched through a normalisation process so that spelling and branding variants of the same property count once, and a blocklist removes non-hotel entities (booking sites, map providers, review platforms) that AI answers sometimes return as recommendations. Where we report anything involving chain brands, we will publish the exact brand list used to classify them.

Mid-study, we ran an adversarial audit of our own raw data and methods, and published corrections to our earlier posts where the recomputed numbers differed. Our collection protocol has not changed since Run 1: same prompts, same platforms, same cadence.

What this Study is Not

It is not a census of how AI recommends hotels. It covers three intent phrasings, in English, from defined locations, without traveller personas, on five platforms, over a defined period.

A family, a honeymoon couple, a solo traveller, and a price-conscious guest would all get different answers, and deliberately so. We are keeping the prompts broad because this is the area we know best, and because it keeps the research focused enough to run repeatedly.

AI answers are also phrasing-sensitive. Our Google comparison uses keyword forms of our question prompts, and we treat the two as related but not identical. And some destination names are ambiguous: where a name like Naples, for example, could refer to more than one place, we classify it by the destination the platforms actually returned. There is a lesson for hotels in that on its own, and we will return to it.

We also cannot see why each platform recommends a particular hotel. What we can measure is what appears: the hotels named, their ranking position, how often they return, and which sources are cited alongside. That matters because boutique hotels often depend on a more fragmented visibility ecosystem than large global brands: their own website, reviews, editorial coverage, destination content, PR, social platforms, OTAs, and third-party hotel guides.

What Happens Next

Over the coming weeks, Lynn is sharing selected observations from each run on LinkedIn, including corrections where our recomputation moves a number. Once the research period is complete, we will analyse the full dataset and publish a detailed report with the final findings, methodology notes, charts, and implications for boutique hotels and hospitality marketers.

We expect the final analysis to be published in early September, in time for the next phase of hotel marketing planning after the summer season.