Your next guest is typing "where should we stay in Gatlinburg with the kids" into ChatGPT, and getting back three names, not ten blue links. If one of them is you, that is a booking. If not, there is no Page 2 to save you. We analyzed thousands of real AI answers across 20 markets and four engines to find out who gets named and why. This guide is what we found, and what to do about it. Each section hands off to a full article for the deep version.
AI search is what happens when a guest asks ChatGPT, Perplexity, Gemini, or Google AI Overviews where to stay. Instead of a ranked list of links, they get a direct answer that names a handful of sites. No scrolling. No Page 2.
The mechanics decide everything that follows. The AI does not read your site like a person does. It cuts each page into pieces, then picks the one piece that best answers the question. A page built as one long block gives it nothing to pick. A page built in clear sections hands it a dozen candidates.
That is why we call this the next wave of SEO. The habits that won Google rankings for a decade are not the habits that win AI answers. Our vacation rental SEO research holds the data behind this guide and gets refreshed as we run new studies.
We ran roughly 3,200 real guest questions through four AI engines, across 20 US markets, five times each, and classified about 24,500 citations. Three numbers tell the story, and you need all three at once.
So the category is winning while most individual brands sit on the bench. The question was never whether AI recommends property managers. It is whether AI recommends you or the operator down the street. Where property managers already win in AI search walks through the full picture. Everything below is about becoming the one it picks.
Here is the part that surprises people who grew up on classic SEO. In our data, backlinks did not separate the operators AI cited from the ones it skipped. Domain rank looked the same on both sides. Generic schema markup was identical: one type each, cited or not. Site maturity was a wash too, with about seven in ten operators carrying any schema at all whether AI cited them or not.
The most telling case runs the other direction. The national brands with the most links of anyone in our dataset were cited least in local answers. Links come with being an established company. The engine is rewarding something else.
Backlinks still matter for classic Google rankings. That is a separate game, and it is still worth playing. But if your AI visibility plan is "build authority and add schema," you are pulling levers that do not move this machine. Why backlinks do not get vacation rentals cited has the full breakdown.
Two page-level exceptions earn their keep, and we cover them below: review markup and FAQ blocks.
The strongest site-level separator we found is breadth of local content. Cited operators publish real pages for the guest types, areas, and locations in their market. Think "cabins for families," a downtown neighborhood guide, a page for group trips.
On our largest panel of 1,181 operators, cited operators carry a median of five area guides against one for non-cited operators, nine guest-type pages against one, and about two and a half times as many pages overall. On the directory-sourced panel, the non-cited operator's median is zero guides and zero guest-type pages. Most operators AI skips are running a brochure, not a library.
Citation rate rises in a clean line with guide count. Operators with zero or one guide were cited 32% of the time. One to five guides, 38%. Five to fifteen, 48%. Fifteen to forty, 52%. Forty or more, 60%. The directory panel shows the same curve at lower base rates, from 15% to 50%.
An obvious objection is that big sites get cited more and also happen to have more guides. We checked. Inside every site-size bucket, having three or more guides roughly doubles citation rate. On the largest sites the jump is biggest of all, from 26% to 57%. If size were the real driver, that is the opposite of what you would see.
Enough to be a real local library, and then you can stop. In our data, content works as a threshold, not a treadmill. A set of local pages moves a property manager from invisible to visible. Past that line, stacking more pages buys nothing.
Here is the evidence. Among operators AI already cites, the link between page count and how often they get cited is roughly zero. The most-cited operators in a market do not have meaningfully more content than lightly-cited ones. Once you are over the line, how often you appear comes down to your market and who else shows up.
We have not pinned the threshold to a single number, and anyone who quotes you one is guessing. What the curve shows is that the biggest single jump comes between "one to five guides" and "five to fifteen," which is where a site stops looking like a brochure and starts looking like a resource. A practical target for one market:
That is a finite project. Cover your guest types, your areas, your locations, and your properties, and you are done. You are not signing up to publish forever.
First, content is close to necessary, but it is not sufficient. We found operators who built a full library and still are not cited. In a crowded market, a library gets you into the draw without guaranteeing a pick.
Second, the threshold is local. A national portfolio with more than 1,200 templated guest pages across the country never crosses the line in any single market, because none of those pages is about one place in depth. Only pages that go deep on one place count toward that market's line.
Most of this guide is about the "where should I stay" question. Guests ask AI a lot more than that, and the earlier questions turn out to be even better ground for property managers.
We ran a second study of 80 trip-planning questions across the same 20 markets, four engines, three runs each, and classified 13,417 citations. The questions were things like "things to do in Broken Bow" and "help me plan three days in Asheville," with no mention of lodging at all.
So the guide you write for "things to do" is not just a breadth signal that helps your other pages get picked. It wins citations directly, at the moment the guest is still deciding what their trip looks like.
Winning the planning answer does not hand you the lodging answer. The engine runs a fresh search on every question, so your guide getting cited for "things to do" does not make your family page more likely to be cited for "where to stay." Build both. Judge the guide by its own citations on planning questions, and judge your guest-type pages by theirs.
One step further upstream, at "where should we go this fall," the guest has not picked a market yet. Here your visibility is gated by your market's. One destination in our set appears in about 20% of those answers; another appears in none. If AI never names your town, it cannot name you. That is a per-market number worth knowing before you plan anything.
When AI cites a property manager, it almost never sends the guest to the homepage. Across about 6,900 citations of operator sites, seven in ten pointed to an interior page. On questions with several requirements, nine in ten did. At the planning stage the homepage is 3% and the guide is 81%. The page that answered the question is the page the guest lands on. A site that is a homepage plus listings has no surface for most of the questions guests ask.
At the page level, structure is the strongest separator we measured. We compared 307 cited pages against 360 non-cited pages from the same operators' own sites, so this is not "good operators versus bad operators." It is "the page AI picked versus the page next to it."
| Separator | Cited pages | Non-cited pages |
|---|---|---|
| Headed sections (median) | 13 | 1 |
| FAQ or Q&A block | 28% | 2% |
| Review (AggregateRating) markup | 16% | 0% |
| Place name in headings (area guides) | 44% | 4% |
| Book-direct call to action | 86% | 63% |
The formula is short. Break each page into many headed sections, each answering one thing. Put your market in the headings. Add an FAQ. Mark up your reviews. Structure beats length.
Length is where people go wrong. Adding words to a single-block page does nothing. What you need is enough real material to fill a dozen sections, which is why cited area guides run about 2,400 words against 1,400 for non-cited guides on the same sites, and why a property page of a few hundred words does not get picked. The vacation rental page AI actually cites has the templates for guest-type pages, area guides, and property pages.
You cannot win every question, and the losing ones follow a pattern. Operators do best on what we call the qualified middle: questions with one or two requirements, like "best cabins for families in Broken Bow." On those, operator sites take roughly half the citations (51% at one requirement, 48% at two). Pile on a third and fourth requirement (pet-friendly, hot tub, EV charger) and the answer flips to Airbnb and Vrbo, with operators falling to about 27%. That is a filter question, and filters belong to databases.
Questions about who the guest is beat questions about what the property has. At matched length, questions about the guest (families, groups, couples) gave operators 65% of citations. Questions about a feature (hot tub, pet-friendly) gave them 46%. Look at it from the AI's side. "Which cabin suits a family reunion" is a judgment call, and AI answers judgment calls with prose pages that make a case. "Which cabin has an EV charger" is a database lookup, and the big booking sites own the database.
Build pages that make a recommendation for a kind of guest. Skip the amenity checklist pages, and skip pricing and deals pages entirely: out of 5,125 cited operator pages, exactly zero were pricing pages. Win the qualified middle instead of the long tail lays out which queries to target and which to leave alone.
There is a hidden reason guest-type pages matter even on plain questions. Gemini turns one prompt into about five searches behind the scenes, and in about half of broad prompts it adds a requirement the guest never stated. "Where should I stay in Asheville" becomes "best family resorts with pools in Asheville" inside the engine. Your family page is competing for that plain question whether the guest typed "family" or not.
Local depth beats scale in AI search, and the biggest brands in the country are the proof. We looked at the 20 largest property-management companies in the country, the ones with the most units, the most links, and the most pages. Only seven of the twenty appear in local AI answers at all, and most of those appear as marketplaces rather than as operators. The portfolio with the most linking domains in our entire dataset is cited nowhere. A 30,000-unit brand with more than 1,200 templated guest pages is not cited in a single local market. Meanwhile a far smaller operator with five genuinely local pages is.
The reason is the threshold from earlier, which is local. A thousand generic pages spread across the country never cross the line in any one market, while five real pages about one place can. Why the 30,000-unit giants lose local AI search tells the whole story.
The same logic flips the conventional wisdom on small markets. Yes, the big booking sites take more share in small markets. But not because small-market sites are worse. We checked: they are just as well built as big-market sites. The gap is supply. Small markets cite 89% of the operators that exist there, against 77% in large markets, and spread citations across roughly 80 operators instead of 150. Fewer operators show up, so AI defaults to an OTA.
Flip that around and it is the best per-operator odds in the industry. If you run in a smaller market, the field is uncrowded and the door is open. Airbnb only wins by default when you do not show up. AI search visibility in small markets has the numbers.
This is also why managers of growing portfolios should own this play directly. A property management website that treats each market as its own content home is the version of "get bigger" that AI rewards.
None of them alone. The engines diverge a lot. Gemini cites operators most, at 53% of its citations. Perplexity cites them least, at 37%, and leans on community sources like Reddit. ChatGPT leans hardest toward the big booking sites. Google AI Overviews simply did not appear on roughly a third of our pilot questions, where Google served a hotels pack or plain results instead.
They all select content the same basic way, by lifting well-structured passages from pages they can read. So page structure is the one lever that works across every engine. Community presence is a secondary signal worth knowing about: 72% of cited operators have some Reddit footprint in their market against 39% of non-cited, though we cannot yet say that presence causes citations.
Work this in order. Each step is ranked by how strongly the factor separates cited operators from non-cited ones, not by how easy it is.
Days 1 to 30: build the pages AI has nothing to pick from yet.
Days 31 to 60: make every page quotable.
Days 61 to 90: fill out the library and fix the misses.
That sequence tracks exactly what separates cited operators from invisible ones. It also happens to make a better direct booking website for the humans who land on it. If you want the per-market version, the Sightline Report is the same playbook run against your market, your competitors, and your site.
AI search for vacation rentals is when a traveler asks an AI tool like ChatGPT, Gemini, Perplexity, or Google AI Overviews where to stay, and the tool answers with a short list of recommended properties and websites instead of a page of links. The AI builds that answer by pulling passages from pages it can read and quote, which is why page structure and local content decide who gets named.
No. The 90% figure describes property managers as a category: about nine in ten AI answers include at least one operator site. Any single operator appears in a small slice of answers, typically a few percent even in their own market. The 90% means the door is open, not that everyone walks through it.
Not much. Backlinks did not separate the vacation rental sites AI cited from the ones it skipped, and the national brands with the most links were cited least in local answers. Local content is what separated the winners. Backlinks still help classic Google rankings, so they are not wasted effort, but they are not the lever for AI visibility.
Generic schema markup does not separate cited vacation rental sites from non-cited ones; both groups carry about the same markup. The one exception is review markup: 16% of cited pages mark up their reviews with AggregateRating schema, while non-cited pages almost never do. Mark up your reviews, and treat the rest of schema as housekeeping rather than a growth lever.
Enough local pages to cross the visibility threshold: a page for each main guest type, a guide for each area, and a page for each location, plus property pages with real depth. Citation rate climbs from about a third of operators with one guide or fewer to about half with five to fifteen guides. Past the threshold, adding more generic pages does not increase how often AI cites you.
Usually not the homepage. Across about 6,900 AI citations of property manager sites, seven in ten pointed to an interior page such as a guest-type page, an area guide, or a property page. On questions with several requirements, nine in ten did. The page that answered the guest's question is the page they land on.
None of them alone. Gemini cites operators most, ChatGPT leans hardest toward the big booking sites, and Perplexity pulls from community sources. All of them select content the same basic way, by lifting well-structured passages, so page structure is the one lever that works across every engine.
No. Classic vacation rental SEO rewards authority signals like backlinks over time. AI search selects the best-structured, most specific answer to the question asked, and the two reward different things. They overlap in the basics: real content, clean pages, local focus.
Your guests are already asking AI where to stay, and AI is already answering with property managers.
Get discovered. Convert the right guests.
See how CraftedStays builds sites that AI can read