When a buyer asks an AI model to recommend a buyer's agent in a top-ten metro, the result reflects months of accumulated citation signals. Not advertising spend. The agents who show up on that shortlist did specific work to get there. This is what that work looks like, and why most of the agents in any given market will never be on the list.

What the new buyer query looks like

The buyer used to open Zillow, draw a polygon on the map, and start scrolling. They still do that, but it is no longer the first step. The first step is asking an AI model a question that sounds like the question they would ask a friend.

"Who's the best realtor in Austin for first-time buyers under 600k." "Fastest selling agents in 78704." "I'm relocating to Charlotte from the Bay Area, who should I talk to about South End." "Best buyer's agent in Scottsdale who actually knows the new construction landscape."

These prompts return three to five named agents with a sentence of reasoning attached. The buyer reads the reasoning, picks one or two, and reaches out. Zillow Premier Agent positions and Realtor.com paid placements do not appear in this flow. The buyer never sees them, because the buyer never went to Zillow first.

If you are not on that AI shortlist in your metro, the buyer who would have been your client is now sitting in another agent's listing presentation.

Why ad spend doesn't move the needle

Every agent we talk to has the same instinct when they hear about AI search: "How much do I have to spend to be in the answer." The honest answer is that you cannot buy your way in. AI models are built on earned signals, not paid placements. Three to five thousand dollars a month on Zillow leads will not produce a single AI citation.

Models are looking for a different stack of evidence: clear entity data, transaction record, neighborhood depth, third-party reviews, local press mentions, and topical authority. None of those are available for purchase in a way that the model can detect. They are accumulated through real work, over time, in public.

This is the most important reframe in the entire shift. The agents who spent the 2010s buying leads are now optimizing for the wrong system. The agents who spent the 2010s building actual neighborhood reputation and producing content are the ones the models already understand.

The realtor GEO playbook

Five pillars move an agent into AI-recommended status in a competitive metro. None of them are optional. Skip one and the citation case stays incomplete.

1. Entity clarity

The model needs to know exactly who you are. That means your full legal name, license number, brokerage affiliation, MLS areas served, contact information, and team structure (if any) marked up in RealEstateAgent schema on your site. Consistent NAP (name, address, phone) across every directory you appear in. A Google Business Profile that matches. A Zillow profile that matches. A Realtor.com profile that matches.

If three sources say you are Jane Smith with Compass and one says you are Jane M. Smith with Keller Williams from two years ago, the model treats you as an ambiguous entity and demotes you in citation candidates. Cleaning this up is the cheapest, highest-ROI work in the entire stack.

2. Neighborhood depth

"I love this town" content is invisible to AI. Models cannot extract specific claims from generic affection. What gets cited is depth: school boundary explainers, HOA rule summaries, micro-neighborhood character notes, walkability and commute realities, what the market actually did in 78704 last quarter versus what it did in 78759.

Pick three to five neighborhoods you actually work in. Produce real content on each. Not 400-word blog posts written by an outsourced content mill. Specific, factual, written from the position of someone who has walked the block at 10pm and again at 7am.

3. Transaction record signals

Your sold listings are the most underused authority asset you own. Each sold listing should live on your site with structured data: list price, sale price, days on market, price per square foot, neighborhood, beds and baths, and a short summary of what made the transaction notable.

Aggregate that into market analysis pages: "What buyers paid per square foot in Eastside Austin in Q4 2025," with your own transactions as primary sources. This is the rare type of content where you, the agent, are an authoritative primary source that no national portal can match. The model knows the difference.

4. Off-domain authority

Models learn about you from everywhere, not just your site. A complete Zillow profile with reviews and a clear bio. A complete Realtor.com profile. Local press mentions when you sell a notable property or comment on a market trend. A Reddit presence in your city's real estate subreddit that is helpful rather than promotional. A Nextdoor presence in the neighborhoods you serve. The occasional podcast guest spot on a regional show.

Every one of these is a signal. None of them individually moves the needle. All of them together create the entity halo that produces citations.

5. Specialization signal

"I sell homes in Austin" is functionally invisible. The model has no specific prompt where you are the obvious answer, because the prompt itself is generic and the city already has 12,000 generalist agents. Specialization solves this. First-time buyers under 600k. Luxury new construction over 2M. Investor-focused multifamily. Relocation specialist for tech transfers. Historic home renovation specialist. Pick a lane and earn it through content, transactions, and reviews.

Agents tell us they worry about narrowing their addressable market. The reality is the opposite. A first-time buyer specialist who gets cited in 60 percent of relevant prompts in Austin captures more buyers than a generalist who gets cited in zero.

The competitive metro reality. In top-ten metros, three to five agents are already locking in default citation status in most categories. Most of them are not the biggest producers in the market. They are the agents who, deliberately or accidentally, built the citation stack first. Once the model trusts an agent for "first-time buyers in Austin," it takes meaningful work to displace them. The 60-to-90 day window to claim a slot is now, not in 2027.

The 60-to-90 day window

In our work with agents and small teams in competitive metros, initial citations begin appearing 60 to 90 days after the GEO fundamentals are in place. By month six, citations stabilize. By month nine to twelve, the model has a clear preference set and the agents in that set get cited by default.

The implication for an agent reading this in early 2026 is straightforward. The work you start this quarter shows up in AI shortlists by summer. The work you start in fall 2026 shows up in 2027 against agents who already have a year of compounding signals. The cost of waiting is not zero.

What to do first

Start with entity clarity. It is the cheapest pillar to fix and the foundation for everything else. Pull your name, brokerage, and license consistency across your site, Google Business Profile, Zillow, Realtor.com, and any directories you appear in. Add RealEstateAgent schema to your site. Make sure your areas served list matches what your bio actually says.

Then run the AI Visibility Scan. It will tell you, against the same signals models actually use, where you stand and where the gaps are. The full GEO playbook is a six-month build. The scan is the diagnostic that tells you which pillar to attack first.