The other 74 percent went to the OTAs, the travel press, and the directories. Same guest, same question, someone else's answer. This is the direct booking fight you already fund, on a surface your reports have never covered.
See what AI sees on your property →
We asked seven AI surfaces the eighteen questions a hotel shopper actually asks, three times each, across the three runs that clear our pattern-readiness rule. Five of those surfaces search the live web, which is where most guests already are. Two answer from model memory. Then we counted every source the answers named. This is the result, at one block per percent.
Real measurement, not a mockup. 5,775 citations counted across 7 AI tools (5 that search the live web, 2 that answer from model memory) and a 13-hotel Hawaii cohort of boutique and chain properties, 18 questions locked at hash 45ed5eee before the first run, 3 repetitions per question per tool, 3 usable runs dated June 21, 2026. The wall's third block is the teardown's remaining source types added together, since the wall has to total 100: review directories 4, social 2, Reddit 2, YouTube 1, Wikipedia 0. This is a published teardown of a market, a dated photograph of one category. A property under engagement is measured differently: the same locked panel is re-run repeatedly against its frozen baseline, because one reading is a data point and the pattern is the product. Hotels are anonymized here, the way every non-customer is on our public pages. Where the teardown names an OTA or an editorial platform it does so because they are public infrastructure rather than businesses in the cohort. Counts and distributions are public at /claims/.
For most of the answer, the citation goes to a platform rather than a property. The 17 percent is the share a direct booking strategy is competing for. When a guest lands on an OTA instead of your site, the booking can still happen. The commission goes with it.
Among the hotels the AI surfaces did name, the independents finished level with the chains, 493 citations to 457. The two most-cited hotels in the cohort are both independents, and the single most-cited hotel is an independent, ahead of every chain in the set.
So the contest is not boutique against chain. It is hotel against platform. A big brand budget did not decide who got cited in this cohort, which is exactly why it is worth measuring rather than assuming.
Independent and chain hotels finished close on total citations across the measured cohort, with the independents ahead. Seven independents and six chains, so read it as level rather than a rout. The finding that survives either reading is who sits at the top: the two most-cited hotels in the set are both independents.
The same eighteen questions, the same day, the same cohort. The share of citations going to hotel-owned websites ranged from 47 percent down to zero, depending only on which tool the guest happened to open.
| AI surface | Hotel-owned share | Third-party share | Mentions |
|---|---|---|---|
| Claude (model memory) | 47% | 53% | 583 |
| Gemma (model memory) | 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 |
Read the bottom of that table again. On the five surfaces that search the live web, where most guests already are, hotel websites carried between zero and 30 percent of the citations. On the two that answer from model memory, they carried more than 40. A property can look healthy on one surface and be absent on another, on the same question, on the same day. That is the reason a single-engine tool cannot tell you where you stand, and the reason we run all seven.
Hotels sell into source markets. The AI surfaces answer each of those markets in its own language, from its own sources, with its own set of names. Measuring the English-language question tells you nothing about the Japanese-language one.
Each additional market panel measures the same category the way that market actually asks it. It runs as its own line and carries no second kickoff, because it reuses the competitor cohort the first kickoff already built. Same discipline, a different room of the same house.
You already benchmark your comp set on rate and on occupancy, and you trust those numbers precisely because the party reporting them does not operate your hotel. This is that same discipline, on the surface where the booking decision now starts.
The questions are committed to a public timestamp before we measure. The method is documented in full at /methodology/, every published number is dated and tied to the run that produced it, and your team or your agency executes the punch list. We never grade our own work.
Hawaii is where the method was proven, in one of the most aggregator-captured hotel markets there is. The method itself is geography blind. Lock the questions your guests ask, run the surfaces, count the names. It reads the same from Waikiki as it does from Napa, Charleston, or Kyoto.
Paste your URL and see in seconds what the seven AI tools can read from your site. Free, no signup.
Prefer a hand-built diagnostic of your category and comp set? Email Lance@hi.neverranked.com.