How AI Decides Which Land Sites to Cite

You rank first on Google for hunting land for sale in your county. Then you ask ChatGPT the same question, and your brokerage never comes up. It names two competitors and a marketplace you do not control.

When an AI tool points to a website as its source, that site has been cited. AI citations are decided after ranking, not by it. Your Google ranking isn’t wasted work here, because it’s how you get considered in the first place. But it is not the finish line.

This article explains that gap and what it means for your brokerage pages. You will see the four steps an AI tool takes before it names anyone, what the research says works, what the AI companies themselves say you can ignore, and what all of it means for the listings and county pages you already have.

Ranking well gets you considered. It does not decide which page gets named.

Your website has hundreds of pages. Google ranks pages, not sites, so when you rank first for hunting land in your county, one specific page holds that spot.

Think of an AI tool as making two key decisions, one after the other. First, it decides which websites it trusts on the subject, and your rankings do most of that work. Then it looks for the page that best answers the exact question in front of it. That may be a county page, a single listing, or an agent page. It is often not the page you rank first with.

So your best page gets you in the door. A different page has to answer the question. If that page is thin, your site is trusted, and you still do not get named.

Semrush’s 2025 study of 5,000 searches ran the same queries through four AI tools and found that the more top-10 rankings a site held, the more often those tools named it. But the tools frequently pointed to a different page from that site than the one ranking.

Read the first column and the tools look consistent, because they all lean on sites Google ranks well. Read the second and the gap opens up, since they keep choosing a different page. The same pattern shows in Ahrefs’ 2025 data on Google’s AI Overviews, where 76 percent of the pages named were already ranking in the top 10. So keep doing the search work that gets your site trusted, then make sure you have a page worth naming when a tool comes looking.

The four steps behind every AI citation

Google documents more of this than most brokers realize. Four things happen between a buyer typing a question and your name showing up in the answer.

Step 1: It searches for questions your buyer never asked

An AI tool does not run the one search it was given. It quietly writes several related searches, runs them at once, and builds a single answer from everything it finds. Google calls this query fan-out and describes it in Google’s guidance for website owners, using the example of a question about a lawn full of weeds that also triggers searches on herbicides, chemical-free removal, and prevention.

Now run that on land. A buyer asks which brokerages sell hunting land in a county. Behind the scenes, the tool may also be searching the deer and turkey in that county, whether tracts have creek frontage, recent prices per acre, whether mineral rights convey, and who lists there.

Your site is not competing to answer one question. It is competing to answer five or six, and a different page may be the best answer to each. That is why the shape of your county and state pages matters more than how polished any single page is.

REALSTACK builds location pages around this thinking, and we call them Collections. They run from state to region to county to city. Each is built around your live inventory, with a title tied to the current listing count, breadcrumbs, the area’s actual listings, filters, and draw-on-map search. On top of that, a broker can write real local content at any level. The North Carolina state Collection includes the state’s regions, agriculture figures, wildlife, major rivers and lakes, and a summary table alongside the listings.

One caution. Google says building separate pages for every possible variation of how someone might search, including these behind-the-scenes searches, violates its spam policy on mass-produced content. The point isn’t to guess hidden questions and build a page for each. It is that a county page backed by real listings and genuine local knowledge already answers several of them honestly.

Step 2: It pulls out one fact, not your whole page

An AI tool does not read your listing the way a buyer does. It looks for one specific fact, pulls it out, and moves on. A fact in a labeled field comes out clean. A fact in the fourth sentence of a paragraph about sunsets often does not, because the tool has to work out what you meant.

This is why it matters that a listing is stored as data rather than as a story. REALSTACK captures every land listing as separate fields. Price, acreage with the plus-or-minus notation, and status each have their own field. So does property type, tagged as Hunting Land, Timberland, or Recreational Land. So do the county, the map coordinates, and the categories for wildlife, recreation, and possible use. Mineral, oil, and gas rights sit in their own field, next to easements and disclosures.

Schema.org publishes a shared vocabulary for describing this to machines, including a type built for property listings. You can manage the whole field set in one place with REALSTACK listings management.

Step 3: It checks whether you are a real business

An AI tool will not recommend a company it cannot pin down. Before naming your brokerage or an agent, it looks for consistent, checkable information about who you are and where you work. HousingWire’s guidance for residential agents makes the same point, and it matters more in land, where expertise is regional, and buyers know it.

A name in your footer gives a machine nothing to verify. A full agent page does. Every REALSTACK agent gets their own page and web address, with the office they work out of, the state they are licensed in, and a bio in their own voice about their actual land experience. The page carries any accreditation they hold, reviews with the reviewer’s name attached, and a live feed of that agent’s listings. Those pages also carry Person schema, which is a standard way of labeling a page so a machine understands it is looking at a real individual with a job, an office, and a license.

Step 4: It names a handful of sources and drops the rest

Most of what the tool finds never makes the answer. Semrush found Google’s AI Mode linked to about seven websites per answer, and AI Overviews to about three. You can clear the first three steps and still get left out.

Plenty of answers name nobody at all. In Ahrefs’ 2025 dataset, roughly a third of AI responses mentioned no business or person whatsoever. The slots are scarce, which is the argument for covering more of your market with real pages rather than perfecting one.

What actually works, and what you can safely ignore

A team of researchers at Princeton and elsewhere ran the first serious study on this, published in 2024. They built a test set of 10,000 questions, rewrote the same source pages nine different ways, and measured which versions AI tools quoted more often.

Three changes worked consistently. Adding real statistics, adding credible quotes, and citing your sources all raised how often a page got used, by up to 40 percent on the study’s main measure. Stuffing in keywords performed badly. Keep the headline number in perspective, since it measures how much of the answer came from that page in a controlled test, not what happens to your traffic.

The quieter finding matters more for land. The researchers reported that what works varies by subject, which is why they argued for tailoring the approach by field. Advice written for software companies is not calibrated for a buyer who wants to know about access, water, soil, and whether the minerals convey.

Now the part almost nobody selling this will tell you. Two of these companies publish guidance for site owners, and neither asks you to do anything exotic. OpenAI’s guidance for publishers is almost entirely about access, stating that any public website can appear in ChatGPT search as long as you don’t block its search crawler: no special format, no file to add, and no markup requirement.

Google goes further and publishes a list of things it says you can ignore for Google Search.

• You do not need a special AI text file. Files like llms.txt are heavily promoted. Google says its search systems don’t use them.
• You do not need to chop your pages into small pieces. Google says its systems handle several topics on one page fine.
• You do not need to rewrite your site in a special style for AI. These tools understand synonyms and meaning.
• Chasing mentions around the web is not the shortcut it looks like. Google says its quality and spam systems both feed its AI answers.
• You do not need special markup to appear. It is still worth having, because it helps machines read your pages and keeps you eligible for richer search results.

Notice that the two lists agree by omission, since the tactics being sold hardest are the ones neither company asks for. We will say this plainly, because brokers get pitched on it constantly. A large share of what gets sold as an AI optimization package is built on things these companies have publicly said they don’t need or use. When a vendor pitches you one, ask what the actual work is underneath the acronym.

What does work is harder to sell. Google asks for content built on real firsthand experience, and its own example contrasts a generic first-time homebuyer article with a specific account of a real inspection decision. That is where a land broker has the advantage. A broker who has walked the tract knows the access, the water, the soil, and what the ground is worth. That knowledge is the one thing an outsourced content mill cannot manufacture.

First, can an AI tool even open your pages?

None of the four steps happen if the software cannot reach your site. Google is clear: a page must be indexed and eligible to appear with a text snippet before it can show up in any AI answer. OpenAI names two separate crawlers, one that gathers pages for ChatGPT search and one that collects training data, and blocking the wrong one quietly removes you from the answers.

REALSTACK sites are set up for this by default. Pages tell search engines to read and index them, using full-length text and large images. The file that controls which automated visitors are allowed does not block AI tools, so they are free to come in. Managed cloud website hosting with a content delivery network keeps pages fast enough to be crawled and served reliably.

Most brokers never check any of this, and a blocked crawler removes you from every AI answer without a warning.

What this means for each page on your site

Here is how those four steps land on the pages you already have.

Read that as a coverage problem more than a quality problem. Most land websites have one or two of these page types doing real work while the rest sit thin. A county page with no local writing can still rank respectably, and it still fails at the second step, because there is nothing on it worth lifting out.

The fix is unglamorous. Fill in the fields on every listing. Write real local content on the counties you actually work. Give every agent a page a machine can verify. REALSTACK map search and the Collections structure give you the framework. The ground knowledge has to come from you.

Frequently asked questions

Does ranking first on Google get you cited by AI?

It helps a great deal, but it does not guarantee it. Ahrefs found that 76 percent of the pages Google’s AI Overviews linked to were already ranking in the top 10. What your rankings buy you is a website these tools trust. Which page they name depends on the question, and it often isn’t the page ranking first.

How do you rank in ChatGPT?

There is no ranking inside ChatGPT, only being named as a source. OpenAI states that any public website can appear, as long as you are not blocking its search crawler. From there, build a site the tool already trusts on your subject, and keep your facts in clean fields rather than long paragraphs.

What is query fan-out?

Query fan-out is when an AI tool takes one question and runs several related searches at the same time. Google describes it using a lawn care question that also triggers searches on herbicides, chemical-free removal, and prevention. For a land site, one buyer question can put five or six of your pages in play at once.

Do land brokerage websites need special markup to be cited by AI?

No. Google states that structured data is not required for its AI answers and that there is no special markup you need to add. OpenAI’s guidance for publishers does not ask for any either. It is still worth having, because it helps machines read your listings and agent pages accurately.

Where this leaves your brokerage

You cannot control what these tools decide. You can control everything they read. All four steps run on things you already own, which are the way your listings are stored, whether your agents look like real people to a machine, what you know about your counties, and whether the software can get to any of it.

You can measure it too, starting with your analytics setup. OpenAI tags every link ChatGPT sends you with a marker your analytics can read, so those visits can be separated instead of disappearing into your totals.

REALSTACK builds the structure underneath it all, across your listings, your agents, and your county pages. You add the ground knowledge nobody can fake. If you want your site built to be read and named by these tools, our team will walk you through what needs to change. The full series starts with The Land Broker’s Guide to AI Search.

Sources

• Semrush, How Google’s AI Mode Compares to Traditional Search and Other LLMs
• Ahrefs, 76% of AI Overview Citations Pull From the Top 10
• Ahrefs, Are AI Mode and AI Overviews Just Different Versions of the Same Answer?
• Google Search Central, Optimizing for Generative AI Features
• Google Search Central, AI Features and Your Website
• OpenAI, Publishers and Developers FAQ
• GEO: Generative Engine Optimization, Aggarwal et al., 2024
• Schema.org, RealEstateListing
• HousingWire, How agents can get cited by ChatGPT and AI search overviews
• Search Engine Land, Query fan-out guide