Contents
Milestone Inc. built this benchmark from 1.38 million live AI answers across 1,600 Milestone hospitality customer locations. The AI Hospitality Alliance (AIHA), HEDNA and AVEN Hospitality join Milestone Inc. in releasing it to the industry, giving hoteliers a shared, evidence-based view of how often AI engines name or cite properties, which sources each engine trusts, and how that is shifting quarter to quarter.
Key Findings
A benchmark of observed AI visibility across 1,600 Milestone hospitality customer locations and five AI engines, July–September 2026. Figures describe this customer base, not the entire hospitality market. Hospitality covers 1,128 hotels and resorts and 472 RV parks, campgrounds and communities. RV parks are 30% of locations but 54% of answers. Apart from the 63% and 55% benchmarks, the findings below are for hotels and resorts only.
Across 1.38M AI answers, median location-level AI Visibility (named or cited) was 63%. The middle half of properties range from 42% to 79%.
On branded prompts, AI names the property in essentially every answer. On unbranded prompts, where engines choose among competing properties, the median falls to 55%.
Hotels and resorts only: median 54%.
A hotel’s own site appears in 45% of its answers, yet owned sites are only 8% of sources cited. Booking sites (21%), travel media and other businesses (14% each) and review sites (11%) make up most of the rest.
For the median hotel, 49% of unbranded prompt answers name or cite an OTA. Booking and review sites together account for 32% of hotel citations; booking sites alone, 21%.
The highest of any engine, about 1.4 to 2.5 times the chat engines (6–11%), with the fewest OTA citations (7%).
Brand-affiliated hotels were named in a median 47% of unbranded prompt answers, independents in 55%. The ranges overlap: the data did not establish a reliable difference between segments.
Hotel visibility rose from 65% in July to 70% in September, and 72% of hotel brands saw the rise.
Travelers Now Consult AI Before They Book
Hotel discovery is moving from search results to synthesized answers. A traveler planning a weekend in Santa Fe once compared a page of links. Increasingly, that traveler asks ChatGPT or reads the AI summary above Google’s results, and leaves with a short list of three names. Independent research indicates this already shapes how a significant share of travelers plan.
AI-referred visitors to travel sites still convert 28% less often than other visitors, according to Adobe, but that gap has narrowed sharply since late 2024.
AI Summaries Reduce Clicks to Websites
Pew Research Center found that Google users who encountered an AI summary were less likely to click on links to other websites than users who did not see one, and very rarely clicked on the sources cited in the summary.
Being Known by AI vs. Being Named or Cited by AI
AI Visibility measures the share of answers that name a property or cite its website. The most revealing distinction is between branded and unbranded prompts. Branded prompts supply the property’s name, so visibility is near-certain. Unbranded prompts are where engines decide which properties to name or cite, and where most of the variation between properties occurs.
The traveler has already chosen the property. AI names it in 100% of answers.
The traveler is still deciding. The median property is named or cited in only 55% of answers.
*The Juniper House in Asheville is a fictional name used to mask real property details. Results reflect real benchmark data.
Visibility Varies by Prompt Intent
Visibility depends heavily on the intent behind the prompt. Comparison prompts, which ask an engine to weigh competing hotels without naming the property itself, showed the highest visibility (75%), though on a small sample. Only unbranded prompts are counted in this chart.
When a traveler compares named hotels, AI often brings in other hotels from the same competitive set. Properties in that set tend to appear.
Depends on AI knowing where a property is and what is nearby. Properties with limited location information are named less often.
These prompts (“best boutique hotel…”) make up 73% of hospitality answers, yet median AI Visibility (named or cited) is only 55%.
From roughly 15% to 80%, with the top tenth near 95%. The gap suggests location content may be a differentiator and warrants further testing.
Specific “near X” content: distances to landmarks and the airport, neighborhood pages and accurate map listings.
Reviews and OTA listings that reflect what the property is “best for”: couples, pets, families, business.
Plainly stated facts on the website: parking, EV charging, pet fees and accessibility.
Different AI Engines Trust Different Sources
ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews run on different models and retrieval partners, and return different answers to the same prompt on the same day. For hotels and resorts, each engine has a distinct citation fingerprint: the mix of sources it relies on before naming a property.
Three patterns stand out. Booking sites are the most-cited source for hotels (21%), leading on Gemini, Perplexity and Claude. ChatGPT leans most on other local businesses’ sites (18%). Google AI Overviews draws 15% of its citations from the hotel’s own website, the highest of any engine, and cites booking sites least (7%).
How Each Engine Decides
Taken together, the fingerprints describe five distinct editorial models. Each draws on a different mix of sources, which may indicate where marketing effort is most visible.
Leans on other businesses’ sites: local partners, venues and attractions. It cites hotels’ own sites more than any other chat engine (11%) and booking sites least among them (14%).
The most booking-driven engine: 24% of citations from OTAs and 16% from travel media. It cites review sites least (6%).
OTAs (21%) plus reviews (16%). Its hotel visibility rose from 65% to 70% this quarter, the largest significant rise of any engine.
The broadest mix: OTAs (21%), travel media (17%, the highest of any engine) and reviews (16%, level with Perplexity).
Cites the hotel’s own site most (15%) and OTAs least (7%), and names the fewest hotels per answer. Inclusion points the traveler to the property, not an intermediary.
The five engines do not behave as a single channel. Gemini, Perplexity and Claude cite booking listings most. Claude and Perplexity also cite reviews more than the others, and Google AI Overviews cites owned pages most. Because engines differ, a single combined score can hide what drives visibility on each.
Booking and Review Sites Shape the AI Answer
This is one of the clearest patterns in the benchmark. AI engines build answers from the sources they retrieve and cite, and for hotels those sources are led by booking platforms, followed by review sites. Expedia, Tripadvisor and Booking.com are each cited in 25–29% of hotel answers. Owned websites are cited, but they form a small share of the evidence behind the answer.
Citing OTA Listings and Reviews Coincides with Higher Visibility
When an AI answer draws on a booking listing or a review page, Milestone observed higher AI Visibility for the hotel (named or cited): 65% vs 56% when no OTA listing is cited. The lift from review pages is smaller (+5 points). When the answer draws on a blog, guide or listicle, the hotel is named or cited less often, most of all with listicles (−10 points), as those articles tend to favor the hotels they already rank.
Most of What Shapes the Answer Sits Outside the Owned Website
For the median hotel, 45% of all answers link to the property’s own site. Yet owned sites represent only 8% of hotel citations. The remainder comes from booking sites, review sites, travel media, and other local businesses.
Being Named Means Competing with Other Hotels in the Same Answer
When engines name a property, sentiment is overwhelmingly favorable: for the median hotel, 94% of mentions are positive and fewer than 1% negative. The greater challenge is competition within the answer. Unbranded prompt answers typically name several properties, and the median hotel holds a 44% share of voice against the competitors it tracks.
Perplexity Named Hotels More Often This Quarter (65% to 70%), Google AI Overviews Less Often (57% to 52%)
A benchmark establishes position; trend tests show how that position is changing. Milestone tested 138 potential changes between July and September on identical prompts and retained only those meeting every criterion: statistically significant after correction, at least 2 points in size, and consistent across most of the market.
Google Is Shortening Its Lists
Google AI Overviews already returns the shortest answers of any engine, averaging 2.19 hotels per unbranded answer across the quarter, against 4.83 to 6.44 for the others. This quarter they contracted further: the number of hotels named in a typical unbranded prompt answer fell by more than a quarter (29%).
Google AI Overviews is listing fewer hotels per answer across the market. This is a market-wide pattern consistent with shorter answers; it does not rule out property-level content as a factor in any individual decline.
Perplexity now names 5.99 hotels per unbranded answer, second only to Claude (6.44). Monitoring limited to a single engine would not capture these differences.
No Clear Brand-Affiliation Advantage Was Established in This Sample
Brand-affiliated hotels were named or cited in 47% of unbranded prompt answers, independents in 55%. The ranges overlap, so the data did not establish a reliable difference between segments.
Independents (55%) and luxury and upper-upscale hotels (55%) had the same median; upscale and upper-midscale hotels were lowest (42%). Because the ranges overlap, these differences are directional only. One possible explanation, not tested here, is that some branded properties rely on brand-level pages with fewer property-specific facts.
Five Priorities for Hospitality Marketers
Each priority is grounded in the findings above and concerns measurement or content practice.
Map the questions travelers ask about properties like yours, including the follow-up queries AI engines generate. Track them as part of your visibility measurement.
Each engine relies on different sources, and a blended score can obscure engine-specific differences. Include Perplexity in the tracking strategy: its hotel visibility rose from 65% to 70% this quarter.
Booking and review sites account for 32% of hotel citations, with booking sites alone at 21%. Keep OTA listings complete and consistent, earn travel-media coverage and respond to reviews.
Publish factually accurate content on every key page to help engines represent the property accurately. Maintain a single source of truth and keep details consistent across channels. One proposed explanation is that clear, structured facts make it easier for AI engines to interpret a property’s information (Milestone, “From strings to things”); this benchmark does not test that effect directly.
The benchmark associates booking listings, and to a lesser extent review pages, with higher visibility, and blogs, guides and listicles with lower visibility. Test formats against specific prompt intents and engines before investing, rather than relying on citation frequency alone.
Research Methodology
This edition of the quarterly Milestone benchmark, published in November 2026, analyzes AI answers recorded for Milestone hospitality customers between July 1 and September 30, 2026. All answers were captured during routine monitoring; none were re-run or simulated for this report. Findings reflect this customer base rather than the hospitality market as a whole and are reported in aggregate.
Sample
*Some businesses own both hotels and RV parks, so business counts overlap. A location is one property, measured as its own profile. A business is the customer account that owns it. An answer is one engine’s reply to one prompt on one day: 1,381,203 answers with text in Q3 2026. RV parks are 30% of locations but 54% of answers because they run more often (a median of about 1,470 answers per location, against about 320 for hotels). Citation, domain, sentiment, share-of-voice and trend findings use hotels and resorts only (1,128 locations). Location benchmarks cover all 1,600 hospitality locations, with hotel-only figures shown alongside. Comparison results are Low confidence (fewer than 20 businesses). Prompts are in English; 86% of locations are in the US.
Engines and collection
Engines are queried through their APIs with live web search on. Each location runs daily, weekly or monthly, and each run asks every engine each prompt once. The median location was measured on 3 run dates with 38 prompts. 615 locations used all five engines and 968 used four, mostly without Claude.
Prompts and prompt types
Prompt libraries are drafted when a location is set up, from the property’s website, name and area; customers can edit them. A prompt is branded when it contains the property’s name or an alias; all others are unbranded (discovery) prompts, about 92% of answers. Every prompt carries equal weight, regardless of search volume.
Share of answers by prompt type. Results use unbranded prompts only. Local: 50% (412 locations, 56 businesses), hotels 56% (209, 54), high confidence. Comparison: 75% (326, 17), hotels 85% (39, 15), low confidence; excludes comparisons naming the property.
Metrics
What was excluded
Benchmark bands and publishing rules
One score per location for the quarter, read at the 25th, 50th (benchmark), 75th and 90th percentiles, with no business above 15% of a group’s weight. Medians carry 95% ranges from 400 redraws; overlapping ranges are not reported as differences. Groups need 25+ locations and 5+ businesses.
Trend tests
Milestone pre-registered 138 tests (101 hospitality), comparing July with September on identical prompts, engines and locations, with 2,000 redraws and a Benjamini-Hochberg correction at 10%. A trend must move 2+ points, in the same direction for 60%+ of businesses, and survive removing the largest. Model settings did not change, so shifts most likely come from the engines.
Limitations
Figures are directional. Prompts are written for each property’s own market, so rates likely run higher than a neutral sample. Results reflect API answers, not signed-in app sessions, and associations do not show cause. Accuracy, clicks and bookings are not measured; no customer is identifiable.
Sources
1,381,203 live AI answers collected for 1,600 Milestone hospitality customer locations, July to September 2026. All Milestone figures in this report come from this dataset.
Secondary research: Phocuswright (March and July 2026), Adobe Digital Insights (June 2026), Pew Research Center (July 2025 and June 2026).
Download the Full Report
Get the complete benchmark as a PDF, including all findings, charts and the full research methodology.