AI sends Hawaii hotel-seekers to the OTAs. The boutiques it does name beat the chains.
13-hotel Hawaii cohort (boutique and chain), 18 hash-locked questions, 3 usable runs on 2026-06-21. Figures generated from the aggregate tooling. Individual properties anonymized. Counts and distributions named.
Why this category matters as a measurement subject
Hawaii hotels are a high-value, fiercely competitive category across every island, and a booking is worth far more than a single lead. Travelers ask AI the way the queries name it: by island (Maui, Kauai, Oahu, the Big Island), by type (boutique, luxury, resort), and by occasion (honeymoon, adults-only, ocean-view). It is also one of the most aggregator-captured categories on the internet: the OTAs and review platforms have spent two decades becoming the default answer, and a hotel that wins a direct booking keeps the 15 to 25 percent commission an OTA would take. That makes hotels a clean test of whether an individual property, especially a boutique, can be named by AI at all when Booking and Expedia own the category.
Methodology summary
Same 7-AI-tool methodology applied across all NeverRanked teardowns:
- 5 web-searching AI tools: Perplexity (sonar API), ChatGPT search (gpt-4o-mini-search-preview), Gemini grounded (gemini-2.5-flash with Google Search), Microsoft Copilot via Bing organic results, Google AI Overviews.
- 2 training-data AI tools: Claude (claude-haiku-4-5, via Anthropic API), Gemma (open-weight, via Together AI).
18 questions a Hawaii hotel shopper would actually ask AI, mixing generic intent (best hotel in Waikiki) with boutique and occasion intent (adults-only, honeymoon, ocean-view) across all the major islands, locked at hash 45ed5eee... so every run compares apples to apples. 3 repetitions per question per AI tool to separate signal from noise. 3 usable runs on 2026-06-21. Pattern-readiness rule of 3 usable runs cleared per MOAT.md rule 5. The 13-hotel cohort was built from the citations themselves, then hard-audited to real hotel own-sites and tagged boutique or chain, with the OTAs and directories excluded as aggregators.
Full methodology + open-source measurement code at /methodology/.
Source-type distribution (cohort-wide)
Across all 7 AI tools and 5,775 citations, AI pulled answers from these source types:
| Source type | % of mentions | Count |
|---|---|---|
| Independent web (OTAs, editorial, directories) | 74% | 4,287 |
| Hotel-owned websites | 17% | 958 |
| Review directories | 4% | 230 |
| Social | 2% | 100 |
| 2% | 90 | |
| YouTube | 1% | 83 |
| Wikipedia | 0% | 27 |
This is aggregator domination on the scale of real estate. The OTAs (Booking, Expedia, Hotels.com, Travelocity) and the review and editorial platforms (TripAdvisor, Conde Nast Traveler, US News, Forbes Travel, plus boutique-hotel directories) make up the bulk of that 74 percent. For most of the answer, AI is recommending a platform, not a property. A hotel’s own website earns 17 percent of the total, which is the share a direct-booking strategy competes for.
Per-AI-tool breakdown
| AI tool | Hotel-owned share | Third-party share | Total mentions |
|---|---|---|---|
| Claude (training data) | 47% | 53% | 583 |
| Gemma (training data) | 42% | 58% | 395 |
| ChatGPT search | 30% | 70% | 722 |
| Google AI Overviews | 20% | 67% | 111 |
| Gemini grounded | 12% | 86% | 1,702 |
| Perplexity | 6% | 80% | 1,452 |
| Microsoft Copilot (Bing) | 0% | 95% | 810 |
The training-data engines lead, the same inversion the real-estate teardown found. Claude (47%) and Gemma (42%) cite hotels’ own sites far more than the web-searching engines do, because iconic and well-marketed Hawaii hotels have the editorial and historical footprint a model learns from. The live-search engines lean on the OTAs instead: Gemini 12 percent, Perplexity 6 percent, and Microsoft Copilot 0 percent, which is the cohort-wide Copilot pattern we see in every category. The practical read for a hotel: the editorial and brand presence that feeds the training-data engines is the most reachable surface, and it is exactly where a distinctive property can out-perform its size.
The finding for a boutique: you are not losing to the chains
The natural fear for a boutique hotel is that AI favors the big chains. The data says the opposite. Among the 958 hotel-owned citations, the boutique and independent properties drew 493 and the chains drew 457. The single most-cited hotel in the entire measurement is a Maui boutique, ahead of every chain brand, and the two most-cited hotels overall are both independent. When a boutique gets named, it competes with the chains on equal footing and often wins, because the AI tools reward a distinctive, well-documented identity more than corporate scale. The real adversary for any Hawaii hotel is not the Four Seasons down the road. It is Booking.com. And on the surface that decides it, the editorial-presence surface, a boutique has a structural advantage a generic chain property does not.
The OTA wall, and the opening
Microsoft Copilot returned zero hotel-owned citations across 810 mentions, and Perplexity only 6 percent, because the live-search results for hotel queries are dominated by Booking, Expedia, and TripAdvisor. That is the wall. The opening is the training-data engines and ChatGPT search, where hotels already earn 30 to 47 percent, and where a property with real editorial coverage, a clear identity, and a well-structured own-site can be the named answer. Whether a given hotel can move from absent to named there is a measurement question we keep answering month over month, not a promise.
Top recurring hotels (anonymized)
The 5 hotels AI cited most often across the 18 questions and 7 tools, by total mentions across the 3 runs, with their bucket:
| Hotel (anonymized) | Total mentions | Questions cited on |
|---|---|---|
| Boutique A | 135 | 7/18 |
| Boutique B | 132 | 10/18 |
| Chain A | 112 | 12/18 |
| Chain B | 89 | 10/18 |
| Chain C | 74 | 11/18 |
The top two are boutique and independent properties, both ahead of the most-cited chain. Across the full cohort the boutiques (7 properties) and the chains (6 properties) finish nearly level on total citations, with the boutiques ahead. The chains appear on slightly more distinct questions (broader coverage), while the top boutiques appear deeply on the questions they own. Both shapes are a path: breadth or depth.
What this teardown does and does not prove
What it supports:
- The 13-hotel cohort is the set of Hawaii hotels AI cites for buyer-shaped questions, surfaced from the citations and hard-audited to real properties, with OTAs and directories excluded.
- Hotels are heavily aggregator-dominated: 17 percent hotel-owned against 74 percent independent web (OTAs and editorial).
- Among hotels that are cited, boutiques out-cite chains (493 to 457), and the most-cited hotel is a boutique. This held across all 3 runs.
- The training-data engines (Claude 47%, Gemma 42%) cite hotels far more than the web-searching engines, the same inversion the real-estate teardown found, which makes it a second travel/high-aggregation data point rather than a one-off.
What it does not yet support:
- That AI behavior on these questions stays stable over months. Models refresh training data and search indices on schedules outside our control. Re-measurement is the only honest answer.
- That changing a hotel’s site or its editorial presence would cause AI to cite it differently. We measured what AI cites. Causation requires pre-registered experiments against named properties with control for confounds. Different scope.
- That the OTA dominance is closable by any one property. Competing with Booking for the live-search slot is a category-level condition this measurement names but does not solve.
Why this is anonymized
None of the 13 hotels in this cohort are paying NeverRanked customers. The non-customer anonymization rule applies: counts, distributions, per-AI-tool numbers, and the boutique-versus-chain split are public. Individual property names are not. The OTAs and directories are named because they are public infrastructure, not cohort businesses. A hotel that becomes a customer gets a 1:1 deliverable that names every property in the cohort, names the questions it is missing on, and ranks the closable conditions. That deliverable is private to the customer.
Measurement window: 3 usable runs on 2026-06-21. Figures generated from the aggregate tooling (teardown-data.mjs) and drift-monitored. Pattern-readiness rule of 3 runs cleared per MOAT.md rule 5. Refresh cadence is monthly or on customer request.
Substantiation: question set locked by hash 45ed5eee..., open-source measurement code, named AI tools on named dates. Gemma is open-weight, so any auditor can independently re-run the same questions and verify the training-data numbers.
Anonymization: the 13-hotel cohort is kept anonymized at the property level per the non-customer rule. Counts, distributions, and the boutique-versus-chain split are public. Individual property names are not.