Back to ResourcesThe Bottom Line
AI agents like ChatGPT, Claude, Perplexity, and Google AI Overviews now read websites on behalf of buyers. They compare agents against each other in real time. They decide who to recommend.
67% of US homebuyers now use an AI tool as their primary research method before contacting an agent, according to FlyDragon's 12,400-response 2026 State of AI SEO in Real Estate survey across 192 metros. That's up from 17% eighteen months earlier. 61.3% of buyer-side searches now begin in an AI engine rather than Google.
Here's the catch from the same study. 91.6% of US agents are invisible to AI in their own metro. The top 1% capture 47% of citation share. In 71% of US metros, no single agent has more than 15% citation share. The dominant position is unclaimed.
Most agents are not optimizing for this yet. That's the window.
The Shift: Your Site Has Two Audiences Now
Your agent website has two audiences now. Most teams only think about the first.
Buyers and sellers want a site that looks professional, loads fast, and feels trustworthy. That hasn't changed.
AI agents need something different. They need information they can extract and quote: clean HTML, structured data, and content broken into the comparable, factual chunks LLMs use to assemble answers. A buyer might spend 30 seconds scanning your bio. ChatGPT can read your entire site in a single fetch, but only if your content is in the raw HTML.
The JavaScript Problem
Vercel's analysis of more than 500 million AI crawler fetches confirmed that GPTBot, ClaudeBot, and PerplexityBot do not execute JavaScript. They fetch the raw HTML, look at it once, and move on under a 1 to 5 second timeout. If your listing details, agent bios, or neighborhood guides only appear after a React or Vue render, AI sees a blank page.
This is the trap most agent websites built on Wix, Squarespace, or older IDX widgets fall into. The site looks beautiful in a browser. To AI it doesn't exist.
The fix is server-side rendering or static generation. Your content needs to be in the raw HTML before any JavaScript runs.
What's Different About Real Estate
Real estate is uniquely concentrated as an AI-search opportunity, and behaves differently from generic SEO.
Conductor's 2026 AEO/GEO Industry Benchmarks found real estate has the lowest AI Overview trigger rate of any consumer vertical, around 4.5%. The discovery surface that actually matters is conversational ChatGPT, Perplexity, and Gemini, not Google's AI Overview box.
Zillow's agent-discovery share fell from 41.2% to 33.8% year over year, the first decline since FlyDragon began tracking in 2024. The default AI frame for real estate has been portal-shaped (agents appear as line items inside Zillow profiles), but recommendation queries are where portals lose. ChatGPT and Perplexity look for people-as-entities, not directory listings, and weight third-party validation heavily.
Three places where portals reliably lose to a well-built agent site:
| Query type | Why portals lose |
|---|---|
| "Best realtor for first-time buyers in Squirrel Hill" | AI looks for the person, not the directory |
| "Who specializes in 1920s craftsman homes in Granite Bay" | Hyperlocal, sub-neighborhood context portals don't write |
| "Agent for a downsizing senior in a 55+ community in Sarasota" | Specialty queries portals don't index by |
The structural read: real estate is uniquely well-suited to conversational AI search. Low transaction frequency, high stakes, hyper-local nuance, and the buyer's need for explanation rather than listing dump. Compounding favors early movers. Per FlyDragon, agents who started AI-SEO work in early 2025 hold 5.7× the citation share of agents who started 12 months later, despite the latter spending more on average.
The 7-Point Framework
This is the playbook. Seven structural signals that move the needle in ChatGPT, Perplexity, Gemini, and Google AI search. Each is something a real estate team can ship over the next 90 days.
1. Server-Rendered HTML
The first signal is the most binary. AI crawlers see what's in the raw HTML response, not what your JavaScript builds after the fact.
Vercel's 500M-fetch crawler study confirmed every major AI crawler (GPTBot, OAI-SearchBot, ChatGPT-User, ClaudeBot, PerplexityBot) fetches HTML once and moves on without rendering. Alli AI's 24-million-request analysis from January through March 2026 found ChatGPT-User now generates 3.6× the volume of Googlebot on enabled sites. A hit from ChatGPT-User reliably indicates a real user inside ChatGPT clicking through. If the page is empty when JS is off, that user sees nothing.
What to do tomorrow: open your site, disable JavaScript in your browser, reload. If your bio, listings, neighborhood pages, or reviews disappear, you have a rendering problem. Move to Next.js, Astro, or a RESO Web API import that produces native HTML on your domain. iframe-embedded IDX widgets that pull listings from a third party leave AI seeing an empty shell on your most important pages.
2. Deep Real-Estate JSON-LD
The second signal tells AI engines what each page IS, in machine-readable form. Generic Article schema is table-stakes. The real lift comes from layered, real-estate-specific schemas: RealEstateAgent, Person, RealEstateListing, Place for neighborhoods, Review, Organization for the brokerage, and Event for open houses, all linked together with @id and sameAs.
Bing's December 2025 webmaster blog and Schema App's analysis both confirm AI engines parse RealEstateListing, Review, Person, Organization, and LocalBusiness schemas as primary signals when assembling answers. The schema work that used to be about Google rich snippets is now about AI extraction. Same code, different beneficiary.
What to do tomorrow: pick one agent and ship a RealEstateAgent block on their bio page that includes name, image, telephone, email, brokerage as worksFor, areaServed listing every neighborhood they sell, knowsAbout for specialties, aggregateRating from real reviews, and a sameAs array of every public profile. Validate with Google's Rich Results Test. Then layer Person for the human and link the two with @id.
3. Verified Entity Networks via sameAs
The third signal is the one most teams miss. AI grounding pipelines build "consensus" about whether an entity is real by cross-referencing names and details across the open web. Ahrefs' 75,000-brand analysis measured brand-mention frequency at r=0.664 with AI visibility, against backlinks at r=0.218. That's a 3× gap. Brand mentions and entity consistency are now the strongest correlate of AI citation we have outside the peer-reviewed literature.
For an agent, the consensus engine works like this: if the same name, photo, license number, brokerage, and phone appear identically on Zillow, Realtor.com, GBP, LinkedIn, Yelp, and the agent's own site, AI engines treat that agent as a high-confidence entity. If the details disagree across those profiles, AI hedges and recommends someone else.
What to do tomorrow: pick one agent. Pull up Zillow, Realtor.com, GBP, LinkedIn, Facebook, Yelp, and any team page on the brokerage site. Standardize the name spelling, headshot, brokerage name, license number, phone, and bio. Then list every URL in the agent's sameAs JSON-LD array on their site. Set a quarterly check.
4. Distributed On-Site Reviews
The fourth signal turns reviews into ranking content. DAC Group's January 2026 analysis found AI Overviews now quote individual Google Business Profile reviews verbatim inside answers. FlyDragon's data ranks distributed review presence as the second-largest predictor of AI citation among real estate agents.
Reviews scattered across Zillow, GBP, Realtor.com, and Yelp build trust for buyers but stay invisible to AI grounding when they live behind those sites' UIs. Mirroring them on-site as Review schema, with attribution back to the original source, is what gets them quoted.
What to do tomorrow: with reviewer permission, pull your top 20 Google and Zillow reviews onto a public reviews page on your site. Each review gets its own permalink, Review schema with author.Person, datePublished, reviewBody, and itemReviewed linked by @id to the agent's RealEstateAgent block. Owner responses on every review (AI extracts those too).
5. Hyperlocal Neighborhood Pages
The fifth signal is content most agent sites just don't have: 15 to 30 deep neighborhood pages per agent, with a fixed structure that mirrors across every neighborhood.
Why fixed structure matters: AI extraction loves comparable formats. When every neighborhood page has the same eight sections (Overview, Demographics, Schools, Transit, Dining, Housing Stock and Median Price, Best For, Trade-offs), ChatGPT can cleanly answer "compare Squirrel Hill and Shadyside for families" because the underlying data is structured the same way. Inconsistent structure kills the comparison.
Per FlyDragon, hyperlocal queries account for 71% of buyer-shortlist queries. Portals don't write this content because their economics require national breadth.
What to do tomorrow: pick the five neighborhoods where your team actually sells. Build the eight-section page for one of them. 1,500 to 3,000 words. Include the current quarter's median price, days on market, and inventory. Quote a named local mortgage broker, builder, or appraiser inside the body. Link to one adjacent neighborhood for AI comparison queries. Then template the rest.
6. Original Quarterly Market Reports With Named Expert Quotes
The sixth signal is the one with the strongest evidence behind it. Aggarwal et al., presented at ACM SIGKDD 2024 and known as the Princeton GEO study, tested nine optimization methods across 10,000 queries. Statistics Addition produced a +41% Position-Adjusted Word Count gain in AI citation. Quotation Addition produced +28%. Citing primary sources produced +30 to 40%. Lower-ranked pages saw a 115% lift. Keyword stuffing produced a negative effect.
This is the highest-quality empirical evidence in the GEO literature. Original numbers and named expert quotes are the most reliable lift documented in a peer-reviewed study to date.
Recency stacks on top: half of cited content in major AI surfaces is now less than 13 weeks old. SE Ranking's research found Gemini 3 replaced about 42% of previously-cited domains within a single update. Quarterly cadence isn't optional.
What to do tomorrow: commit to a quarterly market report per metro. Each report contains absorption rate, months of inventory, list-to-sales ratio, median DOM, and price-band momentum, pulled from your MLS. Quote a named local mortgage broker, builder, or county assessor inside the report. Permanent URL per metro per quarter. Archive previous quarters. Author byline visible at the top.
7. llms.txt and Forward-Compatibility
The seventh point is the one that gets the most hype and the least empirical support. Some commentators have hyped llms.txt as a major AI-search ranking lever. Current evidence suggests AI engines don't yet read it.
Here's what the research found:
| Source | Finding |
|---|---|
| John Mueller (Google), June 2025 | "FWIW no AI system currently uses llms.txt." Mueller compared it to the long-deprecated keywords meta tag. |
| SE Ranking 300,000-domain study (Mar 2026) | No correlation between llms.txt presence and AI citation frequency. Removing llms.txt as a feature improved their ML model's accuracy. |
| ALLMO.ai 94,000-URL citation analysis (Jan 2026) | No measurable citation uplift. |
| Rankability top-1,000 monitor | 0% adoption among the top 1,000 websites. Walmart implemented llms.txt in November 2025 and removed it by January 2026. |
We add llms.txt because (a) the cost is roughly an hour of work, (b) the standard could matter later if Google or OpenAI announces support, and (c) it signals AI-readiness to the humans evaluating your site. We do not lead with it as a ranking lever. Anyone selling llms.txt as the centerpiece of their AI-SEO offer is selling cargo culting.
Does This Actually Work?
A sophisticated client should ask this question. Here's what we have:
Live, named proof at one client. BBRG (our flagship Sacramento client) has been named unprompted in ChatGPT and Gemini responses for "real estate agents in Granite Bay" and "real estate agents in Grass Valley." Three separate callers told a BBRG agent that ChatGPT recommended him. Position 1.0 in Google Search Console for "selling a home in granite bay." We can show you the search-console screenshots.
The strongest evidence comes from a peer-reviewed study. The Princeton GEO study (ACM SIGKDD 2024) found +28% to +41% citation lift from adding original statistics and named expert quotes. That's the lift you can plan around. It is a controlled experiment with published methodology, not a vendor case study.
Large-N studies corroborate the signal architecture. Vercel's 500M-fetch crawler study established that AI crawlers don't render JS. Ahrefs' 75,000-brand analysis measured brand-mention correlation with AI visibility at r=0.664. SE Ranking's 18,767-keyword study found 43.5% of AI Overview citations come from outside the organic top 100, meaning entity authority can override traditional ranking. These are large-sample, methodology-disclosed studies.
The real-estate-specific stats are best-available, not definitive. FlyDragon's 12,400-response real estate survey is the largest published real-estate AI benchmark, and we cite it heavily. The publisher sells AI-SEO services, so the precise percentages should be read as best-available rather than peer-reviewed. The pattern (low agent visibility, top-1% concentration, declining Zillow share, conversational dominance over Google AI Overviews) is corroborated across multiple practitioner sources.
The honest summary: the signal architecture is durable, the precise numbers will move quarter to quarter, and one peer-reviewed study underwrites the strongest single tactic (original statistics and named quotes). The rest is large-sample correlation and live proof at one client.
The Window Is Closing
Two compounding dynamics are running right now.
First-movers compound. Per FlyDragon, agents who started AI-SEO work in early 2025 hold 5.7× the citation share of agents who started 12 months later, despite the latter spending more on average. Once an entity has the sameAs network, the review distribution, and 15 deep neighborhood pages with quarterly refreshes, it accumulates citations faster than late entrants can catch up. AI grounding systems weight established entity confidence heavily.
Citation volatility is now measured in weeks. SE Ranking's analysis of Gemini 3 (rolled out late 2025 to early 2026) found the new model replaced about 42% of previously-cited domains and surfaced 32% more sources per response. The agents positioned with deep entity infrastructure when each model update lands are the ones who get pulled in. The ones with empty agent profiles and no neighborhood content stay invisible.
The window isn't infinitely long. ChatGPT integrations from Zillow (Oct 2025), Redfin (Feb 2026), Realtor.com (Mar 2026), and Zumper concentrate listing-search inside the portals' apps. The agent-recommendation surface remains open in conversational ChatGPT, Perplexity, and Gemini. That's the surface to win.
The ONE Thing to Do Today
Open ChatGPT. Type: "Compare [your name] to [your top competitor's name] in [your zip code]." Read what comes back.
If ChatGPT recommends your competitor over you with specific reasons, those reasons are your starting point. If ChatGPT can't tell the two of you apart, you both have an entity problem. If ChatGPT doesn't know either of you, your zip is one of the 71% of US metros with no dominant agent above 15% citation share. The window is wide open.
Whatever happens, the fix is the same playbook above: server-rendered HTML, deep real-estate JSON-LD, a verified sameAs network, mirrored on-site reviews, hyperlocal neighborhood pages, and original quarterly market reports with named expert quotes. Then ship llms.txt for forward-compatibility, knowing it's not the lever that moves the citations.
The shift from "ranking for keywords" to "being worth citing" is the biggest change in real estate marketing in a decade. It's not about tricks. It's about being a real, well-documented, locally credible entity that AI engines can read, verify, and quote.
Want help setting your team up for this? See the Website + IDX tier or book a call. We map out exactly what to ship first based on your market and your competitive position in AI search today.
