The AI-Search Visibility Study for Short-Term Rentals

Nobody had measured this for vacation rentals. So we did.

When a traveler asks ChatGPT, Perplexity, or Google’s AI “where should we stay in Gatlinburg?”, a handful of companies get named and the rest are invisible. That answer now shapes real bookings. Yet the only vacation-rental study of it published before this one tested 40 prompts, ran each a single time, hand-picked five brands, kept its data private, and never touched Google’s AI answers at all — the surface with the widest reach.

So we ran a bigger, more honest measurement, and we are publishing all of it. In July 2026 Cavmir put 550 real questions to Google’s grounded AI answer engine: 450 destination questions across 15 U.S. vacation-rental markets and six question types, each asked five identical times, plus 100 questions about 20 named companies. For every answer we recorded the exact set of websites the AI cited — 4,157 citations across 720 different domains. The full dataset is downloadable below. Five findings stand out.

1. Independent operators win, not the OTAs. Local and regional rental companies took 82% of all citations. Airbnb and Vrbo appeared in just 17% of answers.
2. The funnel is everything. The vaguer the question, the more the AI leans on Airbnb, Vrbo, and Reddit. The more specific the intent, the more it cites operators — up to 100%.
3. The same question gives different answers every time. Ask five times and about 20 companies surface across the runs, but any single answer names only about nine. Two runs overlap only 47%.
4. Name a company and it usually shows up — usually. Named directly, a company’s own site was cited 89% of the time. One established operator’s site never appeared under its own name.
5. A few operators own each market. In most markets one company is cited in the large majority of answers, and everyone else fights over the remaining slots.

The takeaway for anyone who owns, manages, or markets a short-term rental: on the AI surface most people will use to plan trips, being an independent operator is an advantage, not a disadvantage — but only if the AI can find and trust your own website. Most can’t yet.

The single strongest pattern in the data is that how a traveler phrases the question decides whether an operator has any chance at all. We asked six question types in every market. The share of answers that cited Airbnb, Vrbo, or another OTA fell from two-thirds to zero as the intent got more specific.

Read top to bottom, that is the discovery funnel. A traveler who types a vague “where should we go” gets handed to Airbnb, Vrbo, and Reddit. A traveler who signals intent — “book directly,” “a local company,” “the top managers” — gets handed to operators, and the OTAs vanish entirely. Reddit, meanwhile, was the most-cited single third-party source overall (11.6% of answers) and dominated head-to-head comparisons, cited in 48% of “Airbnb vs Vrbo vs a local company” answers.

In most markets, one company is cited in the large majority of answers and everyone else competes for the remaining slots. Below is the single most-cited operator in each of the 15 markets, and how many of that market’s 30 answers named it. These are the incumbents an operator has to unseat — not Airbnb.

Good work exists in adjacent fields, and we built on its standards. AIVO’s April 2026 hotel study ran three passes per prompt; FlyDragon’s real-estate study ran five. Kevin Indig, Profound, and Ahrefs have all shown the same portability problem we see here — a citation on one engine rarely repeats on another. The peer-reviewed Princeton “GEO” paper showed that adding statistics, quotations, and cited sources measurably raises how often AI engines quote a page. Our method — five passes, variance reported, open dataset — is meant to meet the best of that bar, in a vertical that had nothing like it.

Against the one prior vacation-rental study, the contrast is simple and we’ll state it without heat: it ran 40 prompts once each, on five hand-picked brands, across engines that did not include Google’s, and it keeps its data private. This study ran 550 queries with five passes each, across 15 markets and hundreds of companies, on the largest AI answer surface, and publishes every row. We think research should be checkable. Ours is.