Local Falcon ran 37,500 AI searches across the 100 largest US cities in August 2026, using three prompts buyers actually type: “best realtor near me,” “best real estate agent near me,” and “top rated real estate agent near me.” Among established agents and teams who publish a website, 91.5 percent were never named once. Not ranked low. Never named. The models returned 3,757 distinct agents nationwide, roughly 42 per city, in markets that carry thousands of licensed agents each.
The number that should reset your plan is a different one. Only 21.2 percent of the highest-volume individual agents in those cities got cited by any AI platform. Production does not buy the citation. Whatever the models are reading, it is not your closed volume.
Your Sales Record Is Invisible to the Model
Agents keep treating AI visibility like a reputation problem. Sell more, get known, get named. The August data kills that assumption. Four out of five top producers in the largest US markets are absent from the answer when a buyer asks who to hire.
The reason is mechanical. ChatGPT, Gemini, Grok and Google AI Overviews do not query your MLS. They fetch web pages, read what is on them, and summarize. A model cannot see that you closed 61 sides last year unless a page it retrieves says so in text. Your GCI lives in systems the retriever never touches. So the ranking you win is the ranking on whatever page it happens to pull, and for agent queries that page is almost never yours.
ChatGPT Reads the Directory Page, Not Your Profile
Here is the mechanism. ChatGPT draws 39.8 percent of its agent-recommendation citations from Zillow, 18.0 percent from realtor.com, and 7.3 percent from homelight.com. Portals dominate. That part most agents guessed.
The part almost nobody has acted on: of those Zillow citations, 86.9 percent point at a city-wide agent directory page, not at an individual agent’s profile. The model is not reading your profile and deciding you are good. It is fetching Zillow’s Agent Finder list for your city, reading the names near the top, and repeating them.
Work the math on one market. A metro like Charlotte carries north of 6,000 active licensees. The models named roughly 42 agents there. Zillow’s Agent Finder page for a city shows a couple dozen rows before a reader or a crawler has to paginate. Those numbers sit close enough together that they are effectively the same set. The visible rows on one list page are the shortlist for every AI answer in the market, and everyone below the fold is competing for scraps.
That changes the target completely. You are not optimizing a profile page so a model will like it. You are trying to place inside a list that a model retrieves. This is closer to 2011 local SEO than to anything people call AI strategy. Rank in the directory, get named. Miss the visible rows, stay invisible no matter how good your profile copy reads.
The Five-Day Directory Push
This is a week of work, not a quarter. Block one hour a day.
Day 1. Fix what Zillow knows about you. Claim the profile if you have not. Then use the MLS ID button in your Zillow profile settings, which exists specifically to pull in missing listings and past sales. An unlinked MLS ID is the single most common reason a producing agent shows a thin sales record on the directory row. Set your service areas to the city name buyers type, not only the suburb where you live. If the query is “best realtor in Mesa,” the Mesa list is the one that gets fetched.
Day 2. Push review volume and recency. Reviews are the sort signal sitting right on the directory row, in front of the retriever, in text. Pull your last 20 closed clients and ask 10 of them by name this week. Target five new reviews inside 14 days. Send the direct review link, not a request to go find your page. Recency matters as much as count. A wall of 2023 reviews reads as a stale agent to both the sorter and the buyer.
Day 3. Claim realtor.com. It is 18 percent of the citation pool and it is free. Claiming auto-pulls your NAR certifications and designations. Add past sales, ratings, and reviews. Twenty minutes of work for the second-largest source in the pool.
Day 4. Check HomeLight. It runs on closed transaction data, so your record may already be there and wrong. Verify the transaction count and the ZIP coverage. Correct it if it undercounts you.
Day 5. Make your own site the corroborating source. Build one real city page, not a template farm. Put your name, city, brokerage, transaction count and specialties in body text. Then add a Person plus RealEstateAgent schema block with a sameAs array pointing at your Zillow, realtor.com and HomeLight profile URLs. That array is how a model reconciles that four pages describe one person. Site work does not win the discovery query, but it decides whether you survive the second look.
Then test it. Open ChatGPT in a logged-out window and run: “I am buying a first home in [your city] this fall, budget [your median price]. Name five agents and say why, then list the sources you used.” Note the names and the sources. Save the output. Rerun it in 30 days and compare.
Two Studies Disagree, and the Gap Is the Lesson
Be honest about the conflict here. A separate 2026 analysis reviewed 1,567 ChatGPT citations and found 98 percent came from the agent’s own website and Google Business Profile, with Zillow, realtor.com, Redfin and Trulia cited zero times. That is the opposite conclusion, from a real dataset.
Both findings hold, because they answer different prompts. Discovery queries with no name attached, like “best realtor near me,” pull city-level list pages, and portals own those. Verification queries with a name attached, like “is [agent name] a good agent in [city],” pull that agent’s own site and Google Business Profile because those are the pages about that person. Buyers run both, in that order. So fund both, in that order. Directory placement gets you into the shortlist. Your site and GBP decide whether you survive the check the buyer runs before calling.
For a baseline, Roomvu shipped a free tool called Found on August 21 that scans an agent’s footprint in about 60 seconds and returns a scorecard on profile consistency, review volume and search visibility. Use it to find the gaps, then verify anything it flags by hand. Any vendor grading your visibility on their own scale has a product to sell you at the end of the scan.
The Signal to Watch by October 1
Zillow AI Mode launched March 25, 2026 and answers buyer questions inside Zillow, then hands the buyer to a local agent. Every referral it makes runs on your Zillow profile. Zillow now supplies the largest share of citations to outside models and routes buyers internally with the same data. One profile feeds both paths, which is the strongest argument for spending Day 1 well.
Watch one number over the next 30 days: whether the portal share of AI citations moves. If Zillow’s 39.8 percent climbs, the directory row is the whole game and profile hygiene becomes table stakes. If it falls as models weight first-party pages more heavily, the site layer you built on Day 5 starts paying. Run the prompt above monthly and keep the outputs in a file. Six months of saved answers will tell you more about your market than any national study.
Atlas Unchained builds the profile, schema and city-page layer for solo agents and small teams who would rather list homes than audit directory rows. If you want the five-day push run for you, reply to this post or subscribe below and we will send the checklist and the test prompt.
