How ChatGPT finds and recommends real estate agents, and how to get on the shortlist

There is no database of licensed agents inside ChatGPT. When someone types "best realtor near me," the model does what a research assistant would do: it expands the question into several searches, reads a handful of pages, and writes a summary. The searches usually include variations like "top rated real estate agent in [city]," "best luxury realtor [city]," and the name of a neighborhood. The summary is the shortlist.

That process explains the two things agents find most confusing. First, the answer changes between platforms and between sessions, because each system retrieves a different set of pages. Second, the sources it trusts are boring. They are the pages you already control.

Read those four rows together and a pattern appears. The model is not evaluating your reputation. It is assembling an answer from whatever is easy to retrieve and easy to verify. That helps a single agent with a clean profile and hurts a brokerage that relies on its brand.

The first structural fact about AI search in real estate is scarcity. In the 2026 census, ChatGPT named a median of six businesses per city, with a mean of 6.1 and a range of three to ten. Eight was the single most common answer shape at 31.4%, but 34.6% of answers listed five or fewer. Asking for luxury agents did not create more room. Those answers averaged 6.0 names against 6.3 for the generic query.

A Google results page gives you ten listings, then a second page, then a third. An AI answer gives you six slots and no page two. In a metro with thousands of active agents, that is the entire visible market.

The concentration is just as important as the scarcity. In the average US city, the most-cited agent accounted for 26.6% of all agent citations, and the top four agents took more than half. Seventh place is effectively absent.

The platforms also disagree. Of the agents identified in the three-platform study, only 2.4% appeared across ChatGPT, Perplexity and Gemini, while 88.7% appeared on just one. Nationally, 82.6% of cited agents were named by a single platform and only 1.9% by all four. There is no single AI ranking to win, so the work is to be consistently documented rather than optimized for one engine.

A language model cannot see your production volume unless someone writes it down in a place it can read. That explains most of the gap between the agents who deserve the shortlist and the agents who get it.

Local Falcon cross-referenced its AI citations against RealTrends Verified production rankings. Of 945 top individual agents across 100 cities, 499 published a website that the ranking lists, and only 106 of those were cited by any AI platform. That is 21.2%. Nearly nine in ten agents the AI recommended held no production ranking in their own city at all.

What separated the visible agents was not a better reputation. It was a page. Among the 25 most-cited agent websites, 18 carried a "best real estate agents in [city]" page. Those pages are a small minority of cited sites, yet they absorbed about a third of every citation an agent website received. Example domains in the study include pages titled "Top Real Estate Agents in Las Vegas 2026 Rankings" and "Best Real Estate Agents in Winston-Salem: Side-by-Side Comparison."

The pattern is uncomfortable but clear: the agent who publishes the ranking is almost always the agent named best on it, and the model treats the page as the authority on who is best. That is a documentation gap, not a loophole. Most agents have never written down, in one crawlable place, who the good agents in their market are and why.

Brokerage size does not transfer into AI visibility either. Compass led the study with 1,510 citations across 86 cities, more than the next two brands combined. Coldwell Banker followed with 567 citations in 82 cities and RE/MAX with 454 in 77. Keller Williams, the largest brokerage in the country by agent count, finished eighth with 76 citations across 35 cities. Shorewest drew 85 citations, all of them in Milwaukee.

Concentrated local presence produces more retrievable evidence than a national logo. A brokerage with 40 agents in one metro and a deep set of local pages can out-cite a brand with thousands of agents spread across the country.

Across 1,567 citations, 57.6% came from Google Business Profile or Maps and 40.5% came from the agent's own website. Together that is 98.1%. Local media and editorial coverage accounted for 1.4%, and real estate portals and directories for 0.5%. Zillow, Realtor.com, Redfin, Trulia and Homes.com were cited zero times.

That split favors individual agents, because both places are under your direct control. It also means most agents invest in the one channel AI search does not read. A polished portal profile with hundreds of reviews can inform the prose of an answer while never earning a link.

The Business Profile is the most important asset, and most agent profiles are thin. Set the primary category to Real estate agent and add Real estate agency as a secondary. List the cities and ZIP codes you actually serve. Add the brokerage name, license number, phone, website and service areas exactly as they appear on your site. Upload real photos of you and your listings, not stock images. Publish a weekly post, answer the Q&A section yourself, and keep hours current.

Reviews matter most over time. A profile with 12 reviews is a business card. A profile with 120 reviews, a 4.9 average and responses to every one is evidence. The model quotes review counts more than any other signal, and it can only quote what is visible.

Your site does not need to be large. It needs to be specific and readable. An About page with countable facts in body text, a service page for each specialty, neighborhood pages for the areas you actually farm, and one city-level page that compares local agents. Put the numbers in HTML, not in a JPEG of an award badge.

The 8ight.ai study of 1,630 local answers found that ChatGPT cited a government or licensing site in 25.1% of answers and the Better Business Bureau in 38.7%. Real estate is a licensed category, so state license lookups and brokerage rosters corroborate your record. If your name is spelled three ways and your brokerage changed twice without an update, the record is weaker than it should be.

When researchers searched every answer for the justification attached to each pick, the pattern was consistent: countable, verifiable facts. Review counts appeared in 52.7% of answers, years in business in 47.6%, sales volume in 45.2%, awards and rankings in 42.2%, star ratings in 40.7%, and transaction counts in 27.2%.

The phrasing in real answers is blunt: "$230.9 million in sales volume," "a 4.9 rating from 649 reviews," "over 3,100 sales in Charlotte." Those numbers live on Google Business Profiles and About pages. If your site says "passionate about helping families" and nothing else, the model has nothing to quote, so it quotes someone else.

• Lead with the number. "412 homes sold across 14 years" beats "extensive experience."
• Use the same figures everywhere. Site, profile, bio and press kit should agree, because disagreement reads as unverifiable.
• Put it in text. A number inside an image or a PDF is invisible to most retrieval.
• Be specific about place. "Sold 38 homes in Riverside and Avondale" is stronger than "serving the metro."
• Date it. "As of October 2026" keeps a stale number from becoming a wrong one.

This is the part most AI search advice skips. ChatGPT is not a traffic channel at the scale of Google, and pretending otherwise sets agents up for disappointment. In a 12-month analysis of 100 websites and more than 145 million sessions, ChatGPT drove 0.62% as many sessions as Google organic search. Including every AI assistant raised that to 0.73%.

The quality is different, though. Among 55 sites tracking conversions, 13.8% of ChatGPT visitors converted against 3.02% of Google organic visitors. A separate 95-site benchmark covering September 2025 to August 2026 found AI referrals equal to 2.6% of organic sessions, converting at 1.18 times the organic rate, and 1.51 times for B2C.

Put those numbers together and the conclusion is not a pivot to AI. AI search is a small, high-intent slice, and it depends on the same evidence that drives referrals. Google organic still produced roughly 27 times as many conversions as all AI assistants combined. A 5% improvement in organic performance would add more conversions than doubling AI referral traffic.

What has changed is how buyers research. A Realtor.com survey of 1,000 adults in the market found 82% use AI for housing market information, with ChatGPT at 67% and Gemini at 54%. Yet 62% still named real estate agents as the most trusted and accurate source, ahead of AI. Buyers research with AI and hire through trust, so the shortlist is the handoff point.

The tactics below are ordered by impact per hour, not by difficulty. Moves one through three produce most of the change. Moves four through eight build over a quarter.

1. Fix the Google Business Profile first. Correct categories, real service areas, license and brokerage in the description, current hours, ten or more photos, and a weekly post. This is the place behind 57.6% of citations, and it is free.
2. Put countable proof in body text. Five verified numbers on the About page, in the profile description, and in every bio. Volume, sides, years, review count, rating. Date the figures.
3. Publish a city comparison page. A transparent, updated page about the best agents in your market, with a method, a criteria table, and your own factual entry. This is the strongest single content asset in the data.
4. Build specialty pages. First-time buyers, luxury, relocation, VA loans, condos, new construction, and the neighborhoods you farm. Specialty queries produced agent names 87% of the time against 73% for generic queries.
5. Keep your identity consistent. One spelling of your name, one brokerage, one license number, one phone, across your site, profile, state license lookup, brokerage roster, LinkedIn, Yelp and BBB.
6. Ask for reviews on a schedule. Ask at closing and again at 30 days. Ask for specifics. Respond to every review. Aim for a steady monthly cadence rather than a one-time burst.
7. Earn third-party mentions. Local news, association pages, podcast transcripts, charity boards, co-listing partner sites. These are the corroborating sources that make your own claims believable.
8. Measure with a prompt panel. Ten buyer questions across five platforms, once a month, logged in a spreadsheet. There is no rank tracker for an answer that changes every 40 minutes.

Nothing here requires publishing daily, buying a tool, or restructuring your website. The work is documentation and consistency, which is unglamorous and therefore rare.

The city comparison page deserves its own section because it is the most counterintuitive finding in the research and the easiest to copy. Eighteen of the 25 most-cited agent websites carried a page ranking local agents, and those pages absorbed roughly a third of all citations to agent sites.

The reason is mechanical. A model answering "who is the best realtor in Scottsdale" needs a page that makes a claim about who is best. Most markets have no such page outside portal directories, which the model does not cite. If you publish the only careful comparison in your market, you become the source.

1. State the method up front. Production data, transaction sides, verified reviews, years in business, specialty fit. Say where the numbers came from and when.
2. Include 8 to 12 agents. A list that contains only you is not a comparison. A list of 50 is not a shortlist.
3. Use a comparison table. Name, brokerage, specialty, years, sides or volume, review count, service area. Scannable and quotable.
4. Write a short profile for each. Eighty to 120 words, factual, no disparagement. Yours can explain fit, not superiority.
5. Add a "how we chose" section. Criteria, exclusions, and the date of the last update.
6. Update it quarterly. A page dated 2026 that still lists 2023 numbers is a liability.

One practical note: your own entry should be factual and modest. "Sara Lin has sold 212 homes in Scottsdale since 2014 and holds a 4.9 rating from 168 Google reviews" is more persuasive to a model and a reader than "Scottsdale's #1 luxury expert."

Citations are mentions other people publish about you. Reviews are the most accessible form, because you can ask for them directly. Mentions are slower, because they depend on someone else deciding you are worth writing about. Both matter, and they do different jobs.

Ask every client at closing, then again at 30 days. Send one link, not a menu. Ask for specifics: what the agent did, which neighborhood, how long it took, what surprised them. A review that says "closed in 21 days in Avondale after two failed appraisals" is evidence. "Great agent!" is not. Respond to every review, including the poor ones, because the response is visible text too.

Put Google first. The Business Profile is the page ChatGPT cites. Zillow and Facebook reviews are supporting evidence that can appear in the prose of an answer, but they did not produce a single citation in the 2026 census.

Do not manufacture these. Astroturfed forum posts and paid placements without disclosure are the fastest way to look untrustworthy to a reader and, eventually, to a retrieval system. The goal is a record that holds up when someone checks it, which is what a good answer does.

Here is the part that should feel familiar. An AI recommendation behaves like a referral. It is a short list, assembled by someone else, justified by evidence, matched to a specific need. A buyer who asks a friend for a realtor gets three names. A buyer who asks ChatGPT gets six. Both are trust shortcuts built from what other people have said about you.

That is why the agents who win AI visibility tend to be the agents who are well known locally, not the ones with the largest ad budget. Co-listing relationships, agent-to-agent introductions and past-client referrals create the testimonials, partner mentions, case studies, and local knowledge that show up in content.

Agent-to-agent marketing also builds this record. When you send a listing to local agents, you are looking for a buyer and building the relationships that produce named mentions, reciprocal testimonials, and the local recognition that both buyers and models draw on. Blastrow supports that work: digital flyers and hyperlocal email campaigns create the agent relationships, and those relationships create the third-party evidence that AI answers cite.

AI visibility is not a switch. It is a documentation project you measure monthly. This sequence starts with the moves that produce citations and defers the ones that take a quarter to mature.

The first month should be dull. Most agents skip the profile cleanup because it is not creative, then wonder why a new content page did not change the answer. The model has to be able to identify you before it can recommend you.

A surprising amount of AI search advice is either ineffective or actively harmful. These are the ones worth skipping.

• llms.txt files. No major AI platform has confirmed using them to decide which businesses to recommend. They are cheap to add and irrelevant to this problem.
• Keyword-stuffed city pages. Fifty pages that swap the city name and repeat the same text give a model nothing to quote and readers nothing to trust.
• Bought badges and paid listicles. A "top 1%" graphic with no source is not evidence. Paid placements are usually excluded from retrieval or ignored by readers.
• AI-generated bios at scale. Generic, unverifiable text adds noise. The model rewards specificity, and generated filler is the opposite of specific.
• Chasing "the algorithm." There is no stable ranking to reverse-engineer. The answer changes between sessions and platforms.
• Buying reviews. It violates platform policy, it is detectable, and it destroys the only asset that makes the rest of the strategy work.

The common thread is that every shortcut tries to fake evidence. The system is built to summarize evidence. None of these shortcuts produces a page a model can cite.

You cannot rank-track an answer that changes every 40 minutes, but you can measure a trend. Build a small monthly panel and treat it like a brand survey, not a keyword report.

Pick ten questions a real buyer would ask: best realtor in [city], best luxury agent in [city], first-time buyer agent [city], best agent for [neighborhood], realtors who sell [property type]. Run them on ChatGPT, Gemini, Perplexity, Google AI Overviews and Copilot on the same day each month. Log whether you are named, who else is named, and which sources are cited.

Judge the program by conversations, not sessions. AI referral volume will look trivial next to organic, because it is. The useful question is whether the buyers who arrive that way are further along, and whether your name appears at all when someone compares you with five other agents.

Shortlist size, citation sources and proof signals are from Omni Eclipse's 2026 Real Estate AI Search Report, which captured 334 ChatGPT answers across 253 queries and 197 property markets between April and May 2026, then classified 1,567 source links. Agent citation rates, city concentration and best-of page findings are from Local Falcon's Realtor AI Visibility Index, published August 2026, covering 37,500 AI searches across the 100 largest US cities on five platforms. Cross-platform visibility and specialty-query figures are from withNotable's State of AI Search in Real Estate, Spring 2026, covering 240 queries in 10 US markets. Local-business citation behavior is from 8ight.ai's analysis of 1,630 AI answers, published September 2026. AI referral traffic and conversion figures are from Marie Haynes' 12-month analysis of 100 websites and 145 million sessions, reported by Search Engine Land in October 2026, and from The SEO Works' 95-site benchmark covering September 2025 to August 2026. Consumer AI adoption figures are from Realtor.com's October 2025 survey of 1,000 US adults who were buying, selling or had done so in the past two years. Interview and referral behavior is from NAR's 2025 Profile of Home Buyers and Sellers and NAR generational research. Figures are US-focused unless stated, and AI answers are non-deterministic, so single queries should not be treated as measurements.