Which hotels do AI systems recommend, and why those and not others, is now a demand question for every luxury property. Americas Great Resorts measured it directly. In 824 AI hotel recommendations captured across six US luxury markets on July 29, 2026, AGR found that 25 properties accounted for 53.4% of everything recommended, that 44.7% of the properties named appeared on only one of the three AI surfaces tested, and that the three systems disagreed on the lead hotel in 70.0% of comparable query sets.
AGR then audited the named properties. It found explicit AI crawler blocking in 1.8% of retrievable robots.txt files, and no lodging structured data on 46.7% of retrievable primary property pages.
Americas Great Resorts captured 824 AI hotel recommendations across six United States luxury markets on July 29, 2026: New York City, Los Angeles, Chicago, Miami, Maui, and Napa Valley. Ten question intents per market, across three surfaces: ChatGPT, Google AI Mode, and Gemini. Every capture was run manually by AGR in a logged-out browser session, with results recorded as returned.
On August 18, 2026, AGR audited the public technical configuration of the properties those captures named: robots.txt directives and structured data markup.
This page reports what AGR measured and what those measurements do and do not permit.
Part one: who AGR’s captures named
1. Americas Great Resorts found that 25 properties accounted for 53.4% of all 824 AI hotel recommendations it captured across six US luxury markets on July 29, 2026.
AGR’s captures named 152 distinct properties in total. The AGR dataset’s top twenty properties accounted for 46.0% of its recorded recommendations, and its top ten for 26.2%.
2. Americas Great Resorts found that 49 of the 152 properties named in its July 29, 2026 capture set – nearly a third, were recommended exactly once.
Those 49 properties accounted for 5.9% of all recommendations AGR recorded. Americas Great Resorts found that the modal outcome in its July 29, 2026 capture set was a property receiving a single recommendation, on a single surface, in a single market.
3. Americas Great Resorts found that 44.7% of the properties named in its 824 July 29, 2026 captures appeared on only one of the three AI surfaces tested, while 32.2% appeared on all three.
AGR’s Los Angeles captures named 31 properties, of which 17, or 54.8%, appeared on a single surface, and six appeared on all three.
4. Americas Great Resorts observed disagreement on the lead property in 70.0% of the 60 three-surface query sets it tested on July 29, 2026.
AGR recorded ChatGPT, Google AI Mode, and Gemini naming the same lead property in 30.0% of those sets, and three different properties in 20.0%.
5. Americas Great Resorts measured a 28.7-percentage-point difference in recommendation concentration across the six markets it studied on July 29, 2026: five Maui properties received 70.6% of that market’s recommendations, against 41.9% for the top five in New York City.
AGR’s captures named 14 distinct properties in Maui across 136 recommendations, and 31 in New York City across 136. AGR’s measurements placed Napa Valley near Maui at 69.6% across 20 properties, and Chicago at 63.4% across 25.
6. Americas Great Resorts found that varying question intent across ten intents and three AI surfaces produced only five distinct lead properties in Maui across 30 query-surface combinations on July 29, 2026.
AGR’s cross-intent measurements returned 12 distinct lead properties for New York City and Los Angeles, and 9 for Chicago and Napa Valley.
| Market | Recommendations | Distinct properties named | Top-five share | Distinct lead properties |
|---|---|---|---|---|
| Maui | 136 | 14 | 70.6% | 5 |
| Napa Valley | 135 | 20 | 69.6% | 9 |
| Chicago | 142 | 25 | 63.4% | 9 |
| Miami | 138 | 31 | 52.2% | 11 |
| Los Angeles | 137 | 31 | 48.9% | 12 |
| New York City | 136 | 31 | 41.9% | 12 |
Source, findings 1-6: AGR Luxury Hotel AI Visibility Index, 824 captures, six US luxury markets, July 29, 2026.
Part two: what Americas Great Resorts observed on those properties’ sites
AGR audited 148 of the 152 named properties on August 18, 2026. Four were excluded: two that were closed or demolished, one whose name is withheld under AGR editorial policy, and one that had changed operating identity.
7. Americas Great Resorts found that 2 of 111 recommended luxury properties with retrievable robots.txt files fully blocked GPTBot in its August 18, 2026 audit, and that 93, or 83.8%, referenced no AI crawler at all.
AGR recorded blocking of any of the thirteen AI user-agents it tested in 1.8% of that retrievable set. AGR does not offer this as evidence about whether crawler policy affects recommendation. AGR’s measurements record that among properties AI systems did recommend, explicit blocking directives were rare.
8. Americas Great Resorts found that 42 of 90 recommended luxury properties with retrievable primary pages, or 46.7%, published no Hotel, Resort, or LodgingBusiness structured data on those pages in its August 18, 2026 audit.
Americas Great Resorts observed these 42 properties lacking lodging schema on their audited primary page on August 18, 2026, twenty days after AGR’s captures recorded them as recommended.
9. Americas Great Resorts found that two properties not operating on July 29, 2026 nevertheless received six recommendations in its AI hotel capture set that day.
AGR recorded five Miami recommendations for Mandarin Oriental, Miami, which closed in 2025 and was demolished in April 2026, and one for The Ritz-Carlton Bal Harbour, Miami, which closed in April 2026 with a reopening announced for January 2027. AGR excluded both from the technical audit. As of AGR’s August 18, 2026 audit date, The Ritz-Carlton Bal Harbour maintained a live official property page and the Mandarin Oriental property did not. Operating-status information in this finding comes from public record and property communications, not from AGR’s audit.
10. Americas Great Resorts found that the 48 audited properties publishing lodging schema received a mean of 5.48 recommendations each in its July 29, 2026 captures, against a mean of 5.81 for the 42 without.
Median recommendations within the AGR capture set were 3.0 and 2.5 respectively. AGR publishes this as an uncontrolled observation, not as a measurement of effect. AGR considers the figure confounded by property prominence, brand affiliation, market, and the coverage gap described below, and advises against reading it as evidence that structured data does or does not affect recommendation.
Source, findings 7, 8 and 10: AGR technical audit, 148 properties named in the AGR Luxury Hotel AI Visibility Index, August 18, 2026. Finding 9 draws recommendation counts from the AGR capture set and operating status from public record.
What AGR Found About Which Hotels AI Systems Recommend
Americas Great Resorts measured two things on the same set of properties: how often AI systems named them, and what those properties published on their own websites. The two did not move together.
On the recommendation side, AGR found concentration and instability at the same time. Twenty-five properties took more than half of all 824 recommendations, while a third of the properties named appeared exactly once. Nearly half appeared on only one of the three surfaces tested, and the three systems disagreed on the lead property in 70.0% of comparable query sets. Concentration varied across markets by 28.7 percentage points, and changing the question did little to widen the field within a market.
On the technical side, AGR found no corresponding pattern. Explicit AI crawler blocking appeared in 1.8% of the properties it could audit. Lodging structured data was absent from the audited primary pages of 46.7% of them. Properties publishing that markup received a mean of 5.48 recommendations against 5.81 for those without, a difference AGR reports as uncontrolled and confounded. Two properties that were not operating on the capture date received six recommendations between them, and one of those had been demolished four months earlier.
AGR draws no causal conclusion from this. What its measurements record is that in this dataset, on these dates, the variation in which properties AI systems named was not accounted for by the site-level technical attributes AGR audited. Establishing what does account for it requires the matched comparison set described below, which AGR has not run.
What these measurements do not establish
The audit postdates the captures by twenty days. AGR observed site configurations on August 18, 2026 and recommendations on July 29, 2026. AGR has not established that the observed configurations were in force on the capture date, or during whatever retrieval or indexing preceded it. Findings 7, 8 and 10 describe configurations as observed on August 18 and recommendations as observed on July 29, and nothing more.
Finding 8 is limited to a specific claim. AGR’s observation differs from what would be expected if currently published primary-page lodging schema were universally required for AI recommendation. It is not a refutation of historical necessity, because AGR does not know what those pages published in July. AGR also did not measure structured data elsewhere on those domains, on third-party listings, or in any other source an AI system might draw from, and makes no claim about those.
AGR did not measure whether these factors provide any advantage. Determining that requires a matched set of comparable properties AGR’s captures did not name, audited identically. AGR has not run that set.
AGR’s robots.txt parsing evaluated named user-agent groups. AGR reports the presence or absence of named directives in its audit and does not report an access determination for crawlers that were not named.
Coverage is incomplete and its gaps are not random. AGR retrieved robots.txt for 111 of 148 audited properties (75.0%) and primary pages for 90 of 148 (60.8%). Failures concentrated among chain-operated domains. Those properties may differ systematically from the retrievable set in both crawler policy and markup. The prevalence figures in Findings 7, 8 and 10 describe the retrievable subset only. The retrieval gaps limit AGR’s prevalence estimates but do not eliminate the individual cases AGR observed and reports in Findings 8 and 10.
AGR tested one request configuration. Pages and robots.txt files were requested with a standard desktop browser user-agent. AGR does not know how those domains respond to other clients, including named AI crawlers, and did not test the causes of retrieval failure. Failures are reported as failures and excluded from denominators rather than counted as absences.
The measurement records which properties were named, not why. The Americas Great Resorts capture data records which properties were named in its July 29, 2026 testing, not why they were named. It contains no mechanism.
It is a single dated snapshot. AI answer surfaces can vary between runs. AGR has documented the same query returning a different answer format nine minutes apart, and a citation landing on one phrasing of a question while a nearly identical phrasing returned nothing.
Scope is six US luxury markets and ten question intents. These findings describe those six markets on the dates measured. They do not describe US luxury hotels generally, or any market AGR did not capture.
Method
Captures. Run manually by AGR across ChatGPT, Google AI Mode, and Gemini on July 29, 2026, in logged-out browser sessions. Ten question intents per market, from “top five luxury hotels” to “under the radar.” Results were recorded as returned, whether or not they favored any party.
Technical audit. Run August 18, 2026 against 148 of the 152 properties named in the capture set. Where a property maintained multiple domains or booking paths, AGR selected the primary consumer-facing property page. For each: robots.txt requested from that property’s domain and parsed for full-disallow directives against thirteen named AI user-agents: GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, anthropic-ai, Claude-Web, PerplexityBot, Perplexity-User, Google-Extended, CCBot, Bytespider, Applebot-Extended, and meta-externalagent. The property page requested and parsed for JSON-LD declaring Hotel, Resort, LodgingBusiness, BedAndBreakfast, Motel, or Campground. All requests carried a standard desktop browser user-agent.
Full study design, capture protocol, scoring rules, and per-market concentration results: AGR Luxury Hotel AI Visibility Index. The underlying capture dataset of 824 ranked recommendations is available from AGR on request.
Americas Great Resorts, founded 1993, is a demand infrastructure firm for independent luxury hotels, resorts, and cruise lines. It does not sell or resell rooms.
Related: Why Doesn’t My Hotel Show Up in ChatGPT? · How to Get Your Hotel Recommended by AI
