Is Your Hotel Visible in AI Search? It Depends Which AI You Ask

The Kollective Boutique Hotel AI Index is an ongoing study of AI-generated hotel recommendations. We are 9 runs into our minimum of 15 runs for the initial findings. These are interim numbers: where a defensible methodological choice changes a figure, we give the range, and everything below will be recomputed on the complete dataset for the final report in September.

Nine runs into the study, the dataset covers 13,500 AI answers, just over 93,000 hotel mentions, and around 12,000 distinct hotel names, which merge to roughly 9,600 distinct properties once spelling and branding variants are combined. Some patterns are now stable enough to be worth sharing before the final report. Five stand out.

1. There is no single “AI visibility” for a hotel

Ask ChatGPT, Copilot, Gemini, Google AI Overviews and Google AI Mode the same hotel question, and all five name the same number one hotel on roughly 4% of queries (3 to 6% depending on how strictly hotel names are matched). Around seven in ten of the hotel names in our data appear on only one of the five platforms.

A hotel can be prominent in ChatGPT and absent in Gemini, first in AI Mode and invisible in Copilot. For hotel teams, this means “how do we appear in AI” is really a series of questions about individual AI engines and surfaces. Any tool or audit that reports a single AI visibility score is in danger of averaging away the analysis that matters.

2. The words guests use define the competitive set

We track three descriptors across the same 100 destinations: boutique, luxury and romantic. In every one of the nine runs, the romantic recommendation sets overlap both the boutique and the luxury sets more than boutique and luxury overlap each other. Romantic acts as the bridge between two otherwise distinct shortlists.

The platforms are not just matching keywords; they hold a different idea of the right shortlist for each intent. The same property can sit in very different competitive sets depending on the language a guest uses, which makes the prompts a hotel chooses to monitor a strategic decision rather than a reporting detail.

3. Boutique is where independent hotels still own the conversation

Chain-branded hotels take about 3% of boutique mentions, 12% of romantic and 25% of luxury under a majors-only chain definition, and about 6%, 18% and 33% under a broader one. The levels depend on the brand list, and we will publish the lists we use in due course. The gradient does not move: roughly 1:3:6 from boutique to romantic to luxury under every definition we tested.

For independent hotels, the boutique and romantic conversations are still overwhelmingly theirs. The luxury conversation is where chain brands concentrate.

4. The AI platforms have personalities

Gemini keeps the same top pick between runs most often, on about 59% of queries. Google AI Overviews is the least steady at about 41%, and it fails to produce an overview at all on roughly one in four boutique and romantic queries, against about 3% for luxury.

In our two same-day matched comparisons against ordinary Google search results, Google AI Mode was the outlier. Only a little over half of the hotels it recommends appear in Google’s own top 20 results for the equivalent search, about 53 to 57% across the two matched days, against roughly two-thirds to three-quarters (62 to 76%) for the other four platforms. AI Mode appears to draw on a noticeably different recommendation layer than the search results it sits on top of.

Along the way, we also found that Google AI Mode sometimes shows a separate panel of local hotel listings from Google alongside its written answer. These are directory-style cards rather than the model’s own recommendation, and our tracking initially counted them too, which is why some AI Mode results appeared to contain around 20 hotels. The figures above exclude those cards and count only the hotels AI Mode actually recommends in its answer. We will publish more on this separately.

5. The recommended set reshuffles on every ask

The most important result so far came from a control experiment. This week we ran the full query set twice on the same day, about an hour apart, under identical conditions. The same question produced a different number one hotel about 45% of the time, and only around six in ten of the hotels named the first time reappeared the second. That is almost exactly the churn we measure between runs three days apart.

This is one experiment and we will repeat it before the final report. But the direction is clear: the volatility in AI hotel recommendations is not a ranking drift over time. The platforms assemble a materially different answer nearly every time they are asked. For hotels, the practical question is not “how do we climb the ranking” but “how do we join the stable core of properties that keeps reappearing, ask after ask”.

What this means for hotel teams, so far

Three working conclusions we are already comfortable acting on with our own clients. First, measure AI visibility per platform and over repeated queries; a single screenshot is closer to a coin flip than a diagnosis. Second, treat intent language as strategy: the descriptors guests use place you in different competitive sets, and you should choose which shortlists you want to be actively competing on. Third, aim for consistency of presence rather than position: in AI systems that reroll their answers, the durable asset is being in the pool it keeps drawing from.

What we are not yet ready to say is why certain hotels keep reappearing. That is the second half of the study: more runs, repeated same-day baselines, location testing, and a technical review of the websites of frequently recommended properties, to test whether the common advice matches what the platforms actually surface.

The full report, with methodology notes, charts and the complete dataset behind every figure, is due in early September. We do not name individual hotels in published findings; the study reports the behaviour of the platforms, not a league table of properties.