A hotel can earn an AI recommendation without receiving the accompanying link. Americas Great Resorts examined the source and link records associated with 1,177 assessed answers to 100 travel questions across ten destinations, collected from ChatGPT, Google AI Mode, Gemini and Copilot on September 27, 2026. The answers featured hotels 6,110 times. This follow-up examines a different part of that visibility: whether the accompanying destinations were hotel websites, publishers, rating guides, booking intermediaries or the AI platform’s own travel interface.
Our first study examined which hotels appeared in AI recommendations. This analysis finds that a hotel’s appearance, its citation and its website link are separate things, and sometimes the same URL serves more than one role.
Findings at a glance
- Publishers and editorial guides were prominent in ChatGPT citations. They appeared in 217 of 291 assessed answers (74.6%); hotel or brand websites appeared in 92 (31.6%). An answer could include both categories.
- Copilot website buttons usually also appeared as listed sources. Of 699 distinct answer–URL pairs with verified hotel/brand “Website” buttons, 689 (98.6%) also matched a displayed source list unambiguously. This establishes an answer-level source association, not attribution to a particular sentence.
- Gemini frequently linked to Google Travel. Google Travel links appeared in 270 of 296 assessed answers (91.2%). A hotel’s name on a card did not make its destination a hotel-owned website.
- Google AI Mode offered both Google viewer links and inline hotel-site links. Google viewer URLs appeared in 235 of 290 assessed answers (81.0%). Separately, we identified 60 inline links associated with hotel or brand destinations across at least 31 answers (10.7%); their visible links used Google redirects.
These findings distinguish who was recommended, which sources were listed and where the immediate links led. They do not measure clicks or bookings, and the different interface measures are not a ranking of the engines.
A recommendation, a citation and a website button do different jobs
Consider three things a traveler might encounter in an AI answer.
The answer names a hotel. A citation accompanies a paragraph describing that hotel. A separate button offers a route to the hotel’s website.
The hotel may receive all three forms of visibility, or only one. A citation can lead to a publisher’s review. A card bearing the hotel’s name can lead to a travel platform. A hotel website can appear as both supporting information and a destination link.
| Element in the answer | What it establishes | What it does not establish |
|---|---|---|
| Hotel recommendation | The hotel was featured in the answer | That its website received a link |
| Citation or source group | A source was associated with the answer or a passage | That the source caused the recommendation or supports every claim |
| Hotel “Website” button | The saved interface offered a link to that website | That a traveler clicked, booked or avoided an intermediary |
| Platform hotel card | The hotel appeared within a platform’s travel interface | That the card linked directly to the hotel |
Counting every hotel-named link as a direct hotel link would erase these distinctions.
Gemini: the hotel’s name, Google’s travel destination
Google Travel links appeared in 270 of the 296 assessed Gemini answers. The saved HTML confirmed these as travel-card links, rather than relying solely on their titles.
In a Los Cabos answer about romantic getaways, a card named Las Ventanas al Paraíso, A Rosewood Resort. Its URL led to Google’s travel interface. The property received visibility, but that particular link did not lead directly to the resort’s own website.
That is a finding about the immediate link destination. It says nothing about which booking options a traveler might encounter next, which option they would choose, or whether a booking occurred.
Google AI Mode: Google viewer links and inline hotel-site links serve different roles
Google AI Mode included cited or attached Google /searchviewer links in 235 of 290 assessed answers, or 81.0%.
One checked Punta Cana answer included a source card naming Sanctuary Cap Cana. The URL captured on September 27, 2026 was a Google /searchviewer address. When retrieved on September 28, 2026, it returned a page titled “Place Viewer.” The hotel’s name was visible, while the immediate link destination remained on Google.
AI Mode also offered a different route within the written answer. We identified 60 inline links associated with hotel or brand destinations across at least 31 of 290 answers, or 10.7%. This is a confirmed minimum from the captured evidence.
In a Las Vegas answer, the text said, “Explore bookings at the,” followed by a link labeled “Nobu Hotel Las Vegas Site.” The corresponding source record identified the Caesars page for Nobu. The visible hyperlink itself was an encoded Google redirect, rather than a plain hotel URL.
For these 60 links, the hotel or brand destination was identified in the captured source record; we did not independently decode the Google redirect to verify its final destination. The finding therefore distinguishes the immediate Google link from the hotel destination associated with it in the records. It does not establish that a traveler clicked through or booked directly.
Copilot: a website button usually also appears as a listed source
Saved HTML confirmed hotel or brand “Website” buttons in at least 172 of 300 assessed Copilot answers, or 57.3%. Within those answers, the audit identified 699 distinct answer–URL pairs carrying a verified hotel/brand Website button.
Of those 699 pairs, 689 also matched a displayed source list unambiguously: 98.6%. The overlap was the dominant pattern, rather than an isolated example.
| Copilot website-link measure | Count |
|---|---|
| Assessed answers with a verified hotel/brand Website button | At least 172 of 300 |
| Distinct answer–URL pairs with verified hotel/brand Website buttons | 699 |
| Pairs also matched unambiguously to a displayed source list | 689 |
| Share with an unambiguous source-list match | 98.6% |
An answer–URL pair counts a URL once within an answer. The same URL appearing in another answer counts again, so these figures do not represent 699 different websites or hotels. Nine additional pairs had ambiguous source-list matches, and one was unmatched; the remaining ten should not all be described as absent from the source lists.
The Peninsula Chicago illustrates the pattern. Its hotel card included a “Website” button linked to the property’s page on peninsula.com. That same URL also appeared among the answer’s listed sources.
A hotel website could therefore provide both a navigation link and a listed source in the same answer. An answer-level source list does not, by itself, show which sentence a source supports or why the engine selected the hotel.
ChatGPT: hotel information and independent coverage can both appear beside recommendations
The saved records contained paragraph-level citation evidence for 290 of 291 assessed ChatGPT answers. That coverage includes 286 answers using citation-button records and four using an older citation-pill format.
Across all 291 assessed answers, publishers and editorial guides appeared in the paragraph-linked citation evidence for 217 answers, or 74.6%. Hotel or brand websites appeared in 92 answers, or 31.6%.
| Source category identified in paragraph-linked citations | Assessed answers containing the category | Share of all 291 assessed answers |
|---|---|---|
| Publishers and editorial guides | 217 | 74.6% |
| Hotel or brand websites | 92 | 31.6% |
The categories can overlap within an answer, so these percentages do not add to 100%. If every unresolved domain were assigned to the publisher category, its share would rise to 75.3%. Assigning every unresolved domain to the hotel/brand category would raise that share to 36.8%. These are separate hypothetical allocation limits, not statistical confidence intervals.
The checked examples show what those source associations looked like beside individual hotel discussions.
| Hotel discussed | Linked source | What the saved answer showed |
|---|---|---|
| Finest Playa Mujeres | Finest Resorts | A citation attached to information about the resort’s children’s programming and family facilities |
| The Setai, Miami Beach | Forbes Travel Guide | A citation associated with a discussion of the hotel’s service and recognition |
| Pendry Chicago | Time Out | A citation beside a paragraph discussing the hotel and Michelin recognition |
These examples demonstrate source association. They do not establish that those pages caused the model to select the hotels, or that every statement in the answer was independently verified.
They also show why hotel visibility cannot be reduced to a contest between hotel websites and third parties. A hotel’s own page can provide details about its facilities; a rating guide or publisher can appear alongside a discussion of its reputation. Both can be visible in the same research environment.
A hotel’s name in a domain is not proof of hotel ownership
The audit also encountered URLs that incorporated a hotel’s name within a broader booking-directory hostname.
For example, a Copilot answer about Honolulu included a Halekulani rooms page on halekulanihotel.ratedotels.com. The reviewed page presented accommodation information and reservation functions. Its hotel-specific name was not sufficient evidence to classify it as a hotel-operated website.
We kept that booking-directory role separate and did not infer ownership, affiliation or misconduct from the hostname. The lesson for measurement is straightforward: identify the actual destination rather than treating a hotel name in a title or URL as proof of a direct hotel link.
What hotel teams should measure next
A useful AI visibility assessment should answer three separate questions:
- Was the hotel recommended? Track the property’s appearance for relevant destinations and traveler needs.
- Which sources accompanied the recommendation? Distinguish hotel information, publishers, rating guides, review platforms and booking intermediaries.
- Where did the visible links lead? Separate hotel website buttons, citations and platform destinations, allowing a URL to serve more than one role.
The first question measures recommendation visibility. The other two describe how that visibility is presented and where a traveler could go next. None, by itself, measures traffic or revenue.
Together, these questions provide a more useful assessment than a hotel-mention count alone. AGR’s AI visibility research and resources explore the broader relationship between hotel information, entity representation and AI discovery.
Method and scope
This analysis uses the same September 27, 2026 collection as the first article: 100 distinct questions across ten destinations, with three repetitions per question and engine. The assessed-answer totals were 291 for ChatGPT, 290 for Google AI Mode, 296 for Gemini and 300 for Copilot.
The audit retained original source URLs and examined the associated saved HTML. It distinguished explicitly marked citations, attached links, other returned references, and observable interface roles. Repeated URLs within an answer were consolidated for destination analysis while their original records and multiple roles were preserved.
The Google Travel and viewer percentages use cited-or-attached records. The AI Mode inline count requires a matching label in the answer text and a hotel/brand destination in its source record. The Copilot website-button figure requires a matching “Website” anchor in the saved HTML; its source-list overlap counts distinct answer–URL pairs. The ChatGPT category figures count answers containing paragraph-linked citation evidence for a category, using all 291 assessed ChatGPT answers as the denominator. These measures describe different features and should not be compared as competing shares of one metric.
Source categories describe the identified role of a website or organization. A travel seller can also publish an editorial guide, so its business category alone does not establish the function of a particular linked page. We have not published a ChatGPT OTA/travel-seller percentage on that basis. The unresolved-domain allocation limits above also do not account for possible classification errors among already classified sites.
All 278 domains previously awaiting classification received a review attempt. Fifty-one remained uncertain because access or evidence was insufficient. Those uncertain domains accounted for 143 of 6,730 distinct cited-or-attached answer–URL pairs, or 2.1%. The uncertainty was concentrated in Copilot: 97 of 300 assessed Copilot answers, or 32.3%, contained at least one uncertain domain. A small share of links can therefore affect a much larger share of answers. We have not published Copilot source-category percentages because those unresolved domains could materially change them. Source-page checks conducted on September 28, 2026 establish what could be inspected then, not that every page was unchanged from the collection date.
The study does not measure clicks, booking transactions, commissions, training sources or the reasons an engine selected a hotel. A source associated with an answer is not proof of the answer’s underlying causal origin.
Disclosure: Americas Great Resorts’ own website appeared in cited-or-attached records for six assessed answers: five ChatGPT answers and one Google AI Mode answer. Those appearances were identified separately rather than treated as independent validation of AGR’s research.
