AGR Research | AI Hotel Recommendations | November 2026
AI assistants increasingly help shape which hotels a traveler considers. This study measures where the links in those answers point. Of 900 answers from ChatGPT, Google AI Mode and Gemini, 899 named at least one hotel, and 536 of those contained no link to any hotel or hotel-brand website. Each engine pointed somewhere different. Google AI Mode cited Google. ChatGPT cited publishers and Tripadvisor, and Tripadvisor was the recorded provider for its hotel maps. Gemini often cited nothing.
Basis: 900 answers from ChatGPT, Google AI Mode and Gemini to 100 hotel questions across 10 destinations, each question asked three times per engine, collected November 11, 2026. The answer dataset is available upon request.
Quick Answers
Do AI hotel recommendations link to hotel websites?
Usually not. In AGR’s November 11, 2026 study, 536 of 899 answers that named at least one hotel (60%) contained no link to any hotel or hotel-brand website, either among the citations or in the answer text. Counting citations only, the figure was 705 of 899 (78%). In this collection, Google AI Mode answers were the most likely to carry a hotel website link and Gemini answers the least likely, although 96% of Google AI Mode answers cited Google’s own pages.
How can widespread AI use coexist with low measured referral traffic?
Industry research reports that 56% of active US travelers used AI on at least one trip in the prior 12 months, while AI referrals remain a small share of the traffic measured at large travel companies. This study does not measure traffic. It documents one structural difference: AI answers can name a hotel without containing a link to the hotel’s website. Most answers in this study linked to Google, Tripadvisor, publishers or nothing at all.
What should a hotel check first?
Its Google business information and its Tripadvisor listing. Google AI Mode presented hotels in cards inside Google’s interface, and the hotel’s website appeared in the card data for 1,303 of 1,309 cards. Tripadvisor was the recorded provider for 95% of the pins on ChatGPT’s hotel maps. The study did not test how a change to either record affects an AI answer.
Does the same question return the same hotels across three runs?
Rarely. Each question was asked three times on each engine on the same day. In 292 of 299 complete question-and-engine groups (97.7%), at least one hotel was added or removed across the three answers. That does not mean the entire list changed.
Do the same hotels keep leading?
Some do. The Peninsula Chicago, Halekulani, Hotel Bel-Air and Mandarin Oriental, New York led their destinations in AGR’s September 27, 2026 study, which pooled four engines, and led or co-led again in this three-engine November sample. In the other six destinations, the leader changed.
Scope: These findings describe one November 11, 2026 collection across 10 destinations, 100 questions and three logged-out consumer AI experiences. They measure hotels named and links present in the saved answers. They do not measure clicks, website traffic, bookings or the causes of any recommendation.
At a glance
- AI answers named hotels and rarely linked to hotel websites. 60% of answers that named a hotel had no hotel website link anywhere in the answer. 78% had no hotel website among their citations.
- Google AI Mode cited Google. 95.7% of its citation links pointed to Google pages. Its hotel cards held the hotel’s website in their data, and no card’s link field pointed to a hotel website. More than half its answers did include a hotel website link in the answer text.
- ChatGPT cited publishers and Tripadvisor. Those two destinations received 68.3% of its citation links. Tripadvisor was the recorded provider for 95.0% of its hotel map pins.
- Gemini often gave no sources. 34.3% of its answers carried no citations. When it cited, 62.4% of links went to publishers and guides.
- Hotel lists moved between runs, and six destination leaders changed. At least one hotel changed across same-day repeats in 292 of 299 complete groups. Four of 10 September leaders led or co-led again in November.
What this study measures
Hotel marketers usually judge AI by the referral traffic it sends. That counts only the travelers who clicked a link that happened to point at the hotel. Before any click, the AI answer names hotels and attaches links: citations to sources, links in the answer text, map pins and hotel cards. Those links shape what a traveler can check directly from the answer and where an immediate click leads.
AGR asked ChatGPT, Google AI Mode and Gemini the same 100 hotel questions used in its September 27, 2026 ten-destination study, three times each, and recorded the hotels named and the links present in each saved answer. Each link was classified by destination: hotel or brand website, Google, Tripadvisor, OTA or booking site, Yelp, Reddit or YouTube, or publisher and other site.
The study uses two link measures, labeled the same way throughout.
Any hotel website link. A link to a hotel or hotel-brand website anywhere in the answer, either among the citations or in the answer text.
Hotel website citation. A link to a hotel or hotel-brand website among the sources the engine attached to the answer.
The gap between them matters. In 169 answers, a hotel website link appeared in the answer text while no hotel website was among the citations. Counting citations alone would understate the hotel links present in the full answer by 18.8 percentage points.
The traffic puzzle
Exhibit 1
Travelers report using AI. Travel companies report little AI referral traffic.
| Measure | Finding | Source |
|---|---|---|
| Active US travelers who used AI for planning, booking or in-destination help on at least one trip in the prior 12 months | 56% in the first half of 2026, up from 43% in the second half of 2025 and 33% in the first half of 2025 | Phocuswright, March 2026 |
| Travelers using generative AI platforms for trip research | 33%, against 35% for general search engines | Phocuswright, as reported by TravelPulse, March 26, 2026 |
| AI referral share of traffic to the hotel and short-term rental websites Bain tracked | Quadrupled over a year, from a low base | Bain & Company, September 30, 2026 |
| AI traffic to US travel websites | Up 119% year over year in July 2026. AI-referred visits converted at a rate 1% lower than non-AI visits | Adobe, August 19, 2026 |
| External AI referrals at Booking.com | Significantly less than 1% of Booking.com room nights, as stated by management | Booking Holdings third-quarter 2026 earnings call, November 3, 2026, as reported by PhocusWire |
Source: figures as published by each source. Bain’s analysis used Similarweb data for three to five operators per travel category.
More than half of active US travelers report using AI on a trip. Booking Holdings reports AI referrals at well under 1% of Booking.com room nights. The rest of this study shows where the links in AI hotel answers point.
Where the links point
Exhibit 2
Per answer: AI answers that named a hotel, by the hotel website links they contained
| Engine | Answers that named a hotel | Any hotel website link | Hotel website citation | No hotel website link of any kind | No hotel website citation |
|---|---|---|---|---|---|
| ChatGPT | 300 | 117 (39.0%) | 100 (33.3%) | 183 (61.0%) | 200 (66.7%) |
| Google AI Mode | 300 | 166 (55.3%) | 14 (4.7%) | 134 (44.7%) | 286 (95.3%) |
| Gemini | 299 | 80 (26.8%) | 80 (26.8%) | 219 (73.2%) | 219 (73.2%) |
| All three | 899 | 363 (40.4%) | 194 (21.6%) | 536 (59.6%) | 705 (78.4%) |
Source: AGR’s November 11, 2026 answer dataset. “Named a hotel” means the answer named at least one hotel from AGR’s verified hotel list for that destination.
Exhibit 3
Answers with no hotel website link, compared with answers with no hotel website citation
On smaller screens, swipe or scroll horizontally to view the full chart.
In 169 answers, the engine named a hotel and linked to a hotel website in the answer text while citing something else. 152 of those were Google AI Mode answers.
Exhibit 4
Per link: where each engine’s citation links pointed
On smaller screens, swipe or scroll horizontally to view the full chart.
| Destination of citation link | ChatGPT (2,035 links) | Google AI Mode (1,173 links) | Gemini (1,167 links) |
|---|---|---|---|
| Publishers and other sites | 47.2% | 2.0% | 62.4% |
| Hotel and brand websites | 25.6% | 1.6% | 16.9% |
| Tripadvisor | 21.0% | 0.3% | 0% |
| Google pages | 0.1% | 95.7% | 0% |
| OTAs and booking sites | 3.1% | 0.3% | 15.5% |
| Yelp, Reddit and YouTube | 3.0% | 0.1% | 5.2% |
Source: AGR’s November 11, 2026 answer dataset. Each distinct URL counted once per answer. Totals may differ from 100% because of rounding.
Three engines, three different destinations. In this collection, Google AI Mode pointed its citations to Google. ChatGPT pointed them to publishers and Tripadvisor. Gemini pointed them to publishers, to booking sites, or to nothing. A hotel’s own website was a minority destination on all three.
Google AI Mode: the website is in the card. The card links to Google.
Google AI Mode showed hotel cards in 237 of its 300 answers, 1,309 cards in all. The hotel’s website appeared in the card’s website field for 1,303 of them, 99.5%. No card’s link field pointed to a hotel website. 794 pointed to a Google page. The other 515 were blank.
Exhibit 5
Google AI Mode hotel cards: what the card data held and where the card link pointed
On smaller screens, swipe or scroll horizontally to view the full chart.
Source: AGR’s November 11, 2026 answer dataset. A small number of cards described restaurants or spas inside hotels rather than the hotels themselves.
Google AI Mode also carried hotel website links in its answer text more often than either other engine. 166 of its 300 answers included at least one. In one Cancún answer, the text read “Check availability and packages on the NIZUC Resort & Spa Website,” linked to nizuc.com, while the citation beside it pointed to Google’s own hotel page. The hotel’s link was present. The citation still went to Google.
AI Mode showed paid ads in five answers, nine ads in all: four from Expedia, three from Booking.com and two from IHG Hotels & Resorts.
ChatGPT: publishers, Tripadvisor and Tripadvisor’s map data
ChatGPT attached citations to 255 of its 300 answers. Publishers, guides and ratings bodies received 47.2% of its citation links. Tripadvisor received 21.0%. Hotel and brand websites received 25.6%. 183 of its 300 answers (61.0%) had no hotel website link of any kind.
ChatGPT showed a hotel map in 235 answers, with 1,550 pins. Tripadvisor was the recorded data provider for 1,472 of them, 95.0%. Yelp supplied 57 and other providers 21.
Exhibit 6
Recorded data provider for ChatGPT hotel map pins
On smaller screens, swipe or scroll horizontally to view the full chart.
Source: provider recorded for each pin in AGR’s November 11, 2026 answer dataset. The study did not trace individual map fields to specific Tripadvisor pages.
For 95% of ChatGPT’s hotel map pins, the recorded provider was Tripadvisor. A hotel’s Tripadvisor presence is part of how it appears in ChatGPT, not only on Tripadvisor.
ChatGPT’s background search results, which the traveler does not see as citations, held a second signal. In 27 answers (9.0%), those results included third-party pages built on hotel-named web addresses, such as four-seasons-hotel-chicago.hotel-in-chicago.com and the-miami-beach-edition.miamibeachfl-hotel.com. None appeared as a displayed citation in this study.
Gemini: one answer in three cited nothing
Gemini attached no citations to 103 of its 300 answers, 34.3%. 102 of those 103 still named at least one hotel from AGR’s verified list. In those answers, the traveler received a recommendation with no displayed source to check. The hotel’s name, description and standing existed only in the answer text.
When Gemini did cite, 62.4% of its links went to publishers, guides and travel blogs, and 15.5% to OTAs and booking sites. Hotel and brand websites received 16.9%. 219 of 299 Gemini answers that named a hotel (73.2%) had no hotel website link of any kind, the highest share of the three engines in this collection.
The answers moved. Some leaders held.
Each question was asked three times on each engine, all within one 91-minute collection window. In 292 of 299 complete question-and-engine groups (97.7%), at least one hotel was added or removed across the three answers. A group counts as changed if even one hotel differs. The result does not mean the entire list changed.
Exhibit 7
Most frequently named hotel or hotels by destination, November 11, 2026, compared with the September 27 leader
| Destination | November 11 leader or co-leaders | Answers naming it (of 90) | September 27 leader (four engines) |
|---|---|---|---|
| Chicago | The Peninsula Chicago | 69 | The Peninsula Chicago |
| Honolulu | Halekulani | 75 | Halekulani |
| Los Angeles area | Hotel Bel-Air | 47 | Hotel Bel-Air |
| New York City | Mandarin Oriental, New York; The Greenwich Hotel (tie) | 36 each | Mandarin Oriental, New York |
| Greater Miami | Faena Miami Beach | 58 | The Setai, Miami Beach |
| Las Vegas | Waldorf Astoria Las Vegas; Four Seasons Hotel Las Vegas (tie) | 63 each | Wynn Las Vegas |
| Orlando area | The Ritz-Carlton Orlando, Grande Lakes | 74 | Four Seasons Orlando |
| Cancún and Costa Mujeres | Atelier Playa Mujeres | 59 | Excellence Playa Mujeres |
| Punta Cana | Eden Roc Cap Cana | 57 | Tortuga Bay |
| Los Cabos | Montage Los Cabos | 71 | Waldorf Astoria Los Cabos Pedregal |
Source: November counts from AGR’s November 11, 2026 answer dataset, three engines. September leaders as published in AGR’s September 27, 2026 study, which pooled four engines including Copilot. A September leader counts as holding if it led or co-led in November.
Four of 10 September leaders held: The Peninsula Chicago, Halekulani, Hotel Bel-Air and Mandarin Oriental, New York, which now shares the top with The Greenwich Hotel. AGR’s five-week follow-up on November 2, using all four September engines, found seven of 10 leaders held. The engine sets differ, so the two results are not directly comparable.
What this means for a luxury hotel
The study measures links, not lost bookings. What follows is interpretation.
If the traveler just searches the hotel’s name and books direct, what did the hotel lose? Possibly nothing. One potential risk sits in the pages between the AI answer and the booking. A traveler who follows a link to a Google hotel page, a Tripadvisor listing or a publisher’s list meets rate comparisons, reviews, competing hotels and OTA offers before reaching the hotel. Some of those travelers book direct. Some book elsewhere. Standard last-click analytics credit the final route, not the AI answer that helped form the shortlist. This study does not measure how often that sequence occurs.
In order of effort, a hotel can act on these findings:
1. Check the Google information. Observed: Google AI Mode presented hotels in cards inside Google’s interface, and nearly every card held the hotel’s website. Check: the website, phone number, hours and description Google holds for the property. Not tested: how a change to that information affects an AI Mode answer.
2. Check the Tripadvisor listing. Observed: Tripadvisor was the recorded provider for 1,472 of 1,550 ChatGPT map pins and received 21.0% of ChatGPT’s citation links. Check: the property name, category, description, amenities and recent reviews. Not tested: whether editing a Tripadvisor listing changes ChatGPT’s answer.
3. Search for lookalike pages. Observed: third-party pages built on hotel-named web addresses appeared in ChatGPT’s background search results in 27 answers. Check: search the hotel’s name with words such as hotel, resort and the destination, and look for sites built on the hotel’s name that the hotel does not control.
4. Measure presence, not only referrals. Observed: most answers that named a hotel carried no link to a hotel website. Check: whether the hotel appears in AI answers, for which questions, and where those answers link, using a fixed question set and repeated runs. Watch branded search and direct traffic beside the AI referral line, without assuming every change comes from AI.
5. Know the lists the engines cite. Observed: ChatGPT and Gemini sent most of their citation links to publishers, guides and ratings bodies. Check: which of those pages cover the hotel’s destination and category. Not tested: whether gaining coverage changes how often an engine names a hotel.
AGR’s free AI Visibility Report shows how ChatGPT, Gemini and Google AI Mode represent a property. AGR’s Knowledge Formation Optimization work corrects and strengthens the public record AI assistants draw from.
Methodology
Design. 100 hotel questions across 10 destinations, 10 questions per destination, asked three times each on ChatGPT, Google AI Mode and Gemini. 900 answers. The question set is the same one used in AGR’s September 27, 2026 study. The destinations were New York City, Greater Miami, Las Vegas, Chicago, the Los Angeles area, the Orlando area, Cancún and Costa Mujeres, Punta Cana, Los Cabos and Honolulu.
Questions. Eight shared intents per destination: general hotel discovery, luxury, romantic getaway, family vacation, boutique hotels, spas, on-site dining and exceptional service. Two local questions per destination, such as Walt Disney World and water parks in Orlando, and Broadway and Central Park in New York City.
Capture conditions. Consumer AI search experiences, logged out, US location, collected November 11, 2026, between 10:33 a.m. and 12:05 p.m. Eastern time. Each question and run produced one saved answer. Collection errors were retried until an answer was saved. No answers were selected or discarded.
Platform inclusion. Copilot, part of the September study, could not be collected and is excluded. Claude, Perplexity and other AI assistants were not tested.
Two link measures. A hotel website citation is a link to a hotel or hotel-brand website among the sources the engine attached to the answer. Any hotel website link adds hyperlinks to hotel or hotel-brand websites in the answer text, excluding images. Each distinct URL counts once per answer. The saved answers record that these links were present. The study did not verify how each link rendered on screen.
Classification. Each link was classified by domain: hotel or brand website, including official subdomains, hotel collections and portfolio sites; Google; Tripadvisor; OTA or booking site, including travel agencies and services that arrange stays; Yelp, Reddit or YouTube; or publisher and other site. Every domain classified as publisher or other was reviewed for official hotel websites.
Hotels named. Answers were matched against AGR’s verified hotel list and documented name variants for each destination from the September study. Hotels outside that list are not counted, so “named a hotel” figures are conservative.
Map pins and hotel cards. ChatGPT map pins were attributed by the data provider recorded for each pin. Google AI Mode hotel cards were checked for a hotel website in the card’s website field and for the destination of the card’s link field.
Background sources. ChatGPT’s full search-result list, which is not shown as citations, was used only for the lookalike-page observation.
Repeat consistency. For each engine and question, the hotels named in the three answers were compared. A group counts as changed if any answer named a different set of hotels. One Gemini answer ended after 75 characters and named no hotel. Its group is excluded from the 292 of 299 figure. Including it, 293 of 300 saved groups changed.
Limits. One collection day. Results can change when questions are repeated. The study measures hotels named and links present, not clicks, traffic or bookings. A link to a hotel or hotel-brand website was counted at the answer level, not matched to each hotel named. The November Gemini collection recorded no Google Travel cards, so Gemini cannot be compared with the September study on that point. September and November used different collection methods, so any comparison of link shares between them is directional only.
Prior work
This study reuses the question set from AGR’s September 27, 2026 study, Which Hotels Do AI Engines Recommend? A Study of 10 Destinations. That study assessed 1,177 answers to 100 travel questions across ChatGPT, Google AI Mode, Gemini and Copilot, with three repetitions per question and engine, and recorded 6,110 featured hotel appearances. It measured which hotels led each destination and how leaders varied by engine and traveler intent.
AI Hotel Recommendations: Who Gets the Link?, published September 28, 2026, examined the links in the September answers. ChatGPT cited a hotel or brand website in 31.6% of its answers and publishers in 74.6%. Google AI Mode used Google viewer links in 81.0%, and Gemini linked to Google Travel in 91.2%. That study used a different collection method and link definition, so its figures are not directly comparable with this one.
On November 2, 2026, AGR repeated the September questions across all four engines. Does AI Recommend the Same Hotels Five Weeks Later? found that seven of 10 destinations kept the same leading hotel, while individual recommendation lists frequently changed.
The AGR Luxury Hotel AI Visibility Index, captured July 29, 2026, recorded 824 recommendation slots in 180 answers across six US luxury markets. Five hotels per market, on average, captured half of the slots. It used a different question set, markets and protocol, and its data are not pooled with this study.
How to cite this study
Andrew Paul, Americas Great Resorts. The AGR AI Hotel Link Study. Published November 11, 2026. Fieldwork: November 11, 2026. Permanent study page.
Suggested short citation: “In AGR’s November 11, 2026 study of 900 AI hotel answers, 60% of answers that named a hotel contained no link to any hotel website.”
Alternative short citation: “In AGR’s November 2026 study, Tripadvisor was the recorded provider for 95% of ChatGPT’s hotel map pins.”
The answer dataset and notes, covering all 900 answers with engine, destination, question, run, hotels named and link destinations, are available upon request. Journalists and researchers may reproduce the exhibits above with attribution. Dataset and media inquiries: info@americasgreatresorts.net.
Structured findings (JSON)
The AGR AI Hotel Link Study is published by Americas Great Resorts, operating in independent luxury hospitality since 1993. No hotel paid to appear in this study, and inclusion cannot be purchased. Cite the fieldwork date with its findings.
