Real Estate Brand Perception in LLM-Generated Agent and Brokerage Recommendations
AI systems now recommend agents from their own websites.

A buyer looking for an agent in 2026 no longer scrolls a directory. They ask ChatGPT or Gemini a single question and get two or three names back, and that short list, not a search results page, is where the real estate relationship now begins.
AI Recommendation Engines and the Real Estate Buyer Journey
Ask Perplexity or Claude "best real estate agent in Park Slope" and the model doesn't hand back a list of links to click through. By early 2025, LLM-generated responses already showed up on a growing share of search results pages, and by early 2026 that share had roughly doubled in many verticals. For the agents and brokerages named in that synthesis, the effect compounds. More citations lead to more clicks, more clicks lead to more inquiries, and more inquiries feed back into the signals that earn the next citation. For everyone left out, the opposite happens: they vanish from the buyer's first consideration set before the buyer has any idea they existed.
Real estate makes this asymmetry sharper than most industries, because buyers in 2026 pick the agent first and the brokerage second. So if this sounds like an abstract brand awareness concern, it isn't: it's a direct pipeline problem sitting at the feet of individual agents and teams. It also lands at an unusual moment for the industry's corporate structure. Brokerage ownership is consolidating into fewer, larger entities, but at the same time, the unit of AI recommendation is shrinking down to the individual agent. That combination makes investment in an agent's own identity more urgent, not less, because the corporate brand above an agent's name is doing less of the work it once did to get that agent recommended.
What LLMs Do When Recommending Real Estate Professionals

When you look at how these models actually generate real estate recommendations, you find a narrow, specific pattern, not a fuzzy one. When the Omnie Eclipse 2026 study looked at ChatGPT's real estate recommendations, it found that most citations came from the agent's own website and their Google Business Profile. Major portals, including Zillow, Realtor.com, Redfin, and Trulia, were cited zero times across the citations the study analyzed. So a decade of conventional wisdom about where agents should put their marketing effort needs to be reordered. Years spent building out a profile on a major portal generate close to no return once a buyer moves the search into an AI system, because the two surfaces actually feeding the model are the agent's own site and their Google Business Profile.
What the model says about a recommended agent is not vague praise. The justifications attached to these recommendations are countable and checkable facts, not marketing language. Review counts appear as justification in a majority of answers. Years in business appear in close to half. Sales volume, stated in dollars, appears in close to half as well, and awards or rankings appear in a large share of responses. The exact mix shifts depending on how the question is asked: a plain "best real estate agents" query tends to surface individual agents, while a "best luxury real estate agents" query shifts the results toward teams and firms, and the same brand can perform very differently depending on the precise wording a buyer types in.
Brand search volume is the strongest predictor of whether a name gets cited. Backlinks, the currency of a decade of SEO strategy, show a weak or neutral relationship to AI citation. Domain authority's correlation with AI citation fell sharply between 2024 and 2025. Independent of brand search volume, backlinks, and domain authority, pages carrying complete schema markup, specifically Article schema, FAQPage schema, and Organization schema, are substantially more likely to get cited than otherwise identical pages without it. Taken together, these patterns describe a system hunting for verifiable, structured facts about a real person or firm, pulled from a narrow set of owned sources, rather than a system rewarding the broad web of links and mentions that classic SEO was built to accumulate.
Why rankings and traditional SEO authority no longer predict who gets recommended
The single most counterintuitive finding in this research is how far AI citation has drifted from organic search rank. A large share of AI Overview citations now come from pages that don't even appear in the top ten of organic results, and a substantial share come from pages ranked beyond position 100. A brand can be essentially invisible in classic Google search and still get named by an AI system, and the reverse holds too: strong organic rank is no guarantee of ever getting mentioned. The gap has widened quickly. The proportion of AI citations pulled from top-ranked pages dropped sharply between mid-2025 and 2026, and domain authority's correlation with AI citation kept declining year over year. A metric the SEO industry spent roughly a decade teaching itself to optimize for is losing its predictive power at exactly the layer where buyers are now starting their search.
So if a brokerage or agent assumes their existing SEO work already covers this ground, they will feel the practical consequence. A brand with real investment in technical SEO, a strong backlink profile, and high domain authority is not automatically protected from AI invisibility, and it may have no way of finding that out, because the rank tracking tools built for classic search don't measure AI citation at all. Google's own stated position is that optimizing for generative AI search is still SEO, built on the same foundations as always. That holds true within a single engine. But if a brand wants to show up consistently across ChatGPT, Claude, Gemini, and Perplexity at once, it needs deliberate entity-building work that classic SEO never asked for and does not produce on its own. The SEO investment an agent already made is not wasted, but it was built to answer a different question than the one AI systems are now asking.
How three named real estate brands earned AI placement
Each case shows a distinct route into AI visibility, and the differences matter because agents and brokerages start from very different positions with very different resources available to them.
Carolwood Estates, the Beverly Hills brokerage, built its AI visibility through press authority. LLMs weight trusted, independent publications heavily when they synthesize an answer, so earned coverage from named outlets works as that kind of third-party signal. The lesson generalizes beyond luxury real estate: credibility that other parties grant carries more weight into AI training and retrieval than credibility a brand claims about itself. The Agency, the brokerage that gained national visibility through Netflix's "Buying Beverly Hills," illustrates a related version of the same mechanism: a well-defined entity in a narrow category is more legible to an LLM than a generalist with a larger overall footprint.
SERHANT., the New York brokerage built around Ryan Serhant's personal brand, shows a second path. SERHANT. scored 68 on the EPR Luxury Real Estate Brand Authority Index for the first quarter of 2026, with the highest social-to-earned media conversion rate of any brand in that index. So the case lines up with the broader research finding: named individuals, not corporate brand names, dominate the recommendation slots AI systems generate across most query types.
Nate Clark, an Austin agent, shows you what the path looks like when you start with almost nothing. FlyDragon's campaign for Clark began with close to no digital footprint at all, and after testing several different angles, the campaign narrowed his public identity down to one specialty: probate real estate in Austin. Short YouTube videos answered the specific questions families facing probate actually ask, so the content came out structured and narrowly focused, and an AI model could extract and cite it cleanly. Within four to six weeks, Clark ranked first for probate-agent queries across the major AI platforms, which shows that a sharply defined niche can outrun a broad but thin brand presence in how fast AI recommendation systems pick someone up. A narrowly defined entity reads more clearly to an LLM than a generalist with a wider but shallower footprint.
Without a framework built to measure the difference, a brand in that position has no way to tell mention volume apart from genuine AI trust, which is the exact problem the next section takes up.
Why most real estate brands can't track how AI systems represent them
Coldwell Banker's situation points to a gap that has nothing to do with marketing effort and everything to do with measurement. A new empirical discipline has begun forming around how to quantify AI brand perception: not whether a brand ranks in classic search, but how large language models semantically represent that brand across different models and different kinds of queries. Tugtekin published the AI Perception Index 2026 on SSRN in February 2026, and it lays out the first framework built specifically for this, using something called the Perception Control Framework v2 to run standardized prompts against brands across GPT-4o and Claude Sonnet. The study found a 36-point gap on its Master Perception Index between dominant brands and emerging ones, and it found substantial drift in how the same brand gets perceived from one AI system to the next.
That cross-model drift is the finding that matters most for anyone trying to act on this. A brand can come across well in ChatGPT's recommendations and poorly in Claude's or Gemini's, for the identical question asked by the identical buyer, and no classic analytics tool currently in wide use measures that discrepancy. Researchers working in this space now track five specific metrics to get a reading on AI brand health: Brand Mention Rate, Share of Voice in AI responses, Recommendation Rate (how often the AI explicitly endorses a brand rather than just naming it), Trust Score (which combines how often a brand is cited with the emotional tone of that citation), and Sentiment Score. Visibility within this system is heavily concentrated at the top: the category leader in a given market captures a disproportionate share of all AI visibility, with a steep drop-off for whoever is in second or third place. That concentration is why measuring early and acting on it carries far more value than trying to correct course after a competitor has already locked in the top AI slots.
A platform like Evident is built to answer this measurement gap directly. It scores a brand across the algorithmic, AI-system, and human evaluation dimensions that shape how a real estate brand gets perceived and recommended, so agents and brokerages get a clear read on their AI presence that no single-channel tool currently gives them. Measurement has to come before any attempt at improvement, and at the moment, most real estate brands have no instrument at all pointed at this part of their business.
The concrete signal stack agents and brokerages can act on to improve AI placement

Before any of the work below makes sense, a brand needs to know its current standing, because improving a score nobody has measured is guesswork dressed up as strategy. Once that baseline exists, you can build the signals in a clear order of priority.
Entity clarity comes first, because it is the condition everything else depends on. An LLM has to be able to construct one coherent, internally consistent identity for an agent from what it finds scattered across the web. That requires the same name, the same phone number, and the same brokerage affiliation showing up everywhere, a Google Business Profile listed under the correct primary category with a description that names specific specialties and service areas, and no contradictory details sitting on different platforms.
Review depth comes second: AI systems weigh what a review says over the star rating sitting on top of it. AI systems read the actual text of a review, not just its score: a review describing a specific transaction type, a named neighborhood, or a particular client situation teaches the model something concrete about who the agent serves and how, while a generic five-star review teaches it nothing. Whitespark's 2026 Local Search Ranking Factors study places high numerical ratings, review quantity, and review recency among the strongest local ranking signals, and those same signals feed directly into AI citation logic, which is part of why AI-cited agents earn 35 to 91 percent more clicks per impression than agents who aren't cited.
Schema completeness comes third. Pages carrying Article schema with full metadata, FAQPage schema wrapped around genuine Q&A content, and Organization schema that clarifies the entity behind the page are substantially more likely to get cited than equivalent pages with no structured markup at all.
Content concentration comes fourth, and the Nate Clark case is the clearest demonstration of why it outperforms content volume. If content answers the exact questions a specific kind of buyer asks, it beats a broad library of generic posts, because LLMs extract and cite specific, directly usable answers, not general brand storytelling.
Third-party citation breadth comes fifth, and it functions less as a foundation than as a multiplier on everything built before it. An agent who wants to show up across multiple AI engines at once needs a wide enough footprint of outside citation, built from earned press, local media mentions, and structured data sitting on third-party directories that the models actually read, not portal listings, so that each engine can independently arrive at the same conclusion about who this person is. In luxury and high-value segments specifically, earned press from named publications occupies its own tier above this, the Carolwood Estates model, because those publications carry a kind of independent authority that LLMs weight differently from anything a brand publishes about itself.
What brokerage consolidation means for agent-level AI perception strategy
The research across every section of this piece points the same direction: LLMs recommend named people and specific, well-defined entities, not holding companies. An agent operating under a newly consolidated corporate umbrella inherits none of the AI visibility automatically. The entity an LLM constructs, cites, and eventually recommends is built from that agent's own website, their Google Business Profile, the text of their reviews, the structured data on their pages, and the independent press that names them directly; the logo above their listing this year carries little weight. As the brokerage layer keeps consolidating into fewer, larger companies, the agents who treat their own identity as an asset they build and measure deliberately will be the ones a buyer's first question actually returns.


