Short answer: GEO for real estate is the work of getting an agent, team or brokerage named, and described correctly, when a buyer or seller asks an AI assistant who to call. In our Denver sample, collected on 7 September 2026, five assistants answered 20 questions about choosing a real estate agent. Every answer to a direct request for an agent named someone. But the answers agreed with each other less than in any of the 12 industries in our benchmark. No agent or brokerage was named by all five assistants. The sources were dominated by “top agents in Denver” ranking and referral platforms, and by one individual agent’s blog. One uncorroborated ranking page supplied agent names to four of the five assistants. Below we go through the data and turn it into a playbook for agents, teams and brokerages.

The raw tables are on our real estate agents industry page. This article explains what sits behind them. It covers why the brokerage brand often gets the credit for an agent’s work, why referral platforms matter so much here, what the single-source problem looks like in practice, and what an agent can publish that assistants will actually use.

  • Ask for an agent, get names; ask anything else, mostly don’t. All 35 answers to the seven direct requests named at least one agent, team or brokerage. Only 15 of the 65 answers to the other thirteen questions did.
  • The assistants barely agree. 127 providers were named and none by all five assistants. Two assistants typically had about one name in fourteen in common; ChatGPT and Perplexity had practically none.
  • Ranking and referral platforms carry a lot of weight. HomeLight, Clever, FastExpert, EffectiveAgents, AgentPronto and a single AskDoss page together supplied roughly one citation in seven.
  • One page can put names into four assistants. A “Best Real Estate Agents in Denver 2026” page was cited in 20 answers by Perplexity, Gemini, Claude and Grok, and we found no second source for the individual agents it named. By 22 October the page returned HTTP 410 (gone).
  • Agents’ own content works, but credit often goes to the brokerage. One Denver agent’s blog was cited 31 times, the second most-cited domain overall, yet the answers named his business once. “Compass” was named in 13 answers.
  • Questions about checking an agent’s record were answered with other states’ rules. On the disciplinary-history question, the Texas Real Estate Commission was cited 6 times, more than Colorado’s regulator.

What we measured in Denver, and how far one city sample goes

We generated 20 questions that a home buyer or seller might put to an AI assistant when choosing an agent. None of them named an agent or brokerage, so every name in the answers was the assistant’s own choice. Each question went to five assistants with web search available: ChatGPT (gpt-5-mini), Perplexity (sonar), Gemini (gemini-3.5-flash), Claude (claude-haiku-4-5) and Grok (grok-4.20-non-reasoning). Denver, Colorado was the sample city for local questions, and 13 of the 20 questions mentioned it. The run completed on 7 September 2026, and all 100 question-and-assistant cells returned an answer.

A separate parser read each answer in full and recorded the agents, teams, brokerages and services it named, whether it explicitly recommended them, the answer’s form and every source cited in the answer text. Retrieval candidates that were read but not cited are not counted. Across 100 answers, 127 distinct providers were named, and 913 citations pointed to 378 distinct domains.

What this supports

It supports conclusions about how assistants answer real estate questions. It shows which questions produce names and what kinds of sources they draw on. It shows how individual agents, teams and brokerages are represented, and how much the assistants agree. These patterns are consistent across the seven “find me an agent” questions, which cover selling, investment property, a fast sale before a job transfer, a downtown condo, luxury homes, relocation and a $600k budget.

What it does not support

It is not a ranking of Denver agents, and nothing here assesses anyone’s service. The most-named provider, Compass, appeared in 13 of 100 answers; most names appeared once. Answers change between runs. A different city would bring different agents, and possibly a different balance between referral platforms, portals and agents’ own sites. Real estate is also a business of individuals inside brokerages, so who counts as “the provider” is itself a judgement the assistants make differently, as you will see. We name businesses below only as the data shows them. Where we could not verify an individual agent, we do not name them.

Finding 1: names come with a direct request for an agent and a place

As in every local industry we have measured, the intent of the question decides whether a name appears at all.

SHARE OF ANSWERS THAT NAMED AT LEAST ONE AGENT OR BROKERAGE · DENVER · 7 SEP 2026Asks for an agent7 questions · 35 of 35 answers100%Comparison2 questions · 6 of 10 answers60%Evaluation3 questions · 3 of 15 answers20%Research4 questions · 3 of 20 answers15%Trust and verification4 questions · 3 of 20 answers15%
Figure 1. Where GEO for real estate can win a mention: share of answers that named at least one agent, team, brokerage or agent service, by the intent of the question. All 100 cells completed. Source: InsightWonder industry benchmark, real estate agents, Denver, Colorado sample, collected 7 September 2026.

All seven direct requests produced names in all five assistants: 35 of 35 answers. The comparison questions were the next most likely to name someone (6 of 10 answers). One of them was “I’m a first-time homebuyer with a $400k budget in Denver. What type of agent should I work with?”, a question about a type that most assistants answered with examples. Trust, research and evaluation questions rarely named anyone. Every answer to the questions about red flags, agent interviews and checking disciplinary history named nobody.

The city matters as much as the intent

The 13 questions that mentioned Denver produced named providers in 45 of 65 answers; the seven without a city did so in 5 of 35. Gemini was the exception that proves the rule. It named providers in 15 of its 20 answers, including generic questions such as “Should I use an agent from a large national firm or a local independent agent?”, where it listed national franchises (Keller Williams, RE/MAX, Coldwell Banker, Century 21) as examples.

What buyers and sellers actually ask for

The questions that produced names were situations, not “best agent in Denver”. Someone who knows my neighbourhood to sell my house. An investment-property specialist. A fast sale because I am leaving in 60 days. A condo near downtown. Luxury homes above $2 million. Help understanding areas when relocating. Good deals at $600k. Each is a niche, and each answer drew on a different mix of sources.

Situation (Denver sample)Most-named providers (assistants naming them)What got citedWhat an agent can build
Selling, wants neighbourhood knowledgeKeller Williams Integrity Real Estate (3); Elite Home Partners (2)A brokerage team site, HomeLight and EffectiveAgents lists, Zillow agent reviewsNeighbourhood pages with your own sales in that area, dated
Investment property specialistThe FI Team, Pinnacle Real Estate Advisors, Envision Advisors (3 each)The firms’ own sites, BiggerPockets, commercial listing sitesAn investment page with property types and the services you provide
Fast sale before a job moveHomeLight (2); iBuyer and cash-offer servicesHomeLight, Clever, FastExpert “top agents” pagesA relocation-sale page with your timeline and process
Condo near downtownUsaj Realty (3); several teams (2 each)Usaj Realty’s own blog (9 citations on this question)A downtown condo page with buildings, HOA issues, your track record
Luxury above $2 millionCompass (4); LIV Sotheby’s International Realty (3)Individual luxury agents’ websitesAgent site that states the brokerage and the agent as one entity
Relocating, needs area helpCorken + Company (3); Corcoran Perry & Co (2)Corken’s relocation guide; city and travel guidesA relocation guide to areas, commutes and price bands
Good deals at $600kRE/MAX Professionals and several individual agents (3 each)One “best agents 2026” ranking page, other agent listsA price-band page with your own recent results

Finding 2: the assistants agree less on real estate than on anything else we measured

Real estate is the least consistent of the 12 industries in our benchmark, and the only one in which no provider was named by all five assistants. In our Chicago law-firm and Austin dentist samples, five providers each reached all five. In Denver, none did. The most any provider reached was four assistants. Of the 127 names, most appeared in a single answer.

How little overlap there was

For each question, we compared the names that each pair of assistants produced. On average, two assistants had about one name in fourteen in common. The closest pair, Claude and Grok, shared about one name in seven. ChatGPT and Perplexity shared practically none: across the seven direct requests they almost never named the same agent. That is not a quirk of one question; it is the pattern across the whole run.

Why real estate splits so much

Three things push the assistants apart. First, there is no single obvious unit to name. An answer can name an agent, a team, a brokerage office or a national franchise, and the assistants choose differently. Second, the pool of candidates is large: a city like Denver has far more agents than law firms or dental practices of comparable visibility. Third, the sources are different for each assistant, as Finding 3 shows: Perplexity leans on ranking platforms and Reddit, while ChatGPT never cited Reddit or the AskDoss page at all.

AVERAGE CITATIONS PER ANSWER · 20 REAL ESTATE QUESTIONS · 7 SEP 2026Perplexitysonar19.5Grokgrok-4.20-non-reasoning9.35ChatGPTgpt-5-mini7.25Geminigemini-3.5-flash5.8Claudeclaude-haiku-4-53.75
Figure 4. Average number of sources cited per answer, by assistant, across all 20 real estate questions. Source: InsightWonder industry benchmark, real estate agents, Denver, Colorado sample, collected 7 September 2026.

What “good” looks like when nobody agrees

Low agreement changes what a realistic goal is. In law and dentistry, a handful of providers reached all five assistants on their strongest question. In Denver real estate, the strongest brokerages and teams reached four, and most named agents reached one. So a sensible first target for an agent is not “named everywhere”. It is to be named by two or three assistants on the one or two niches you work in, and to be the cited source on a related research question. Reaching that means your name, your niche and your numbers appear in enough independent places that different assistants, searching in different ways, keep landing on you. A spike in one assistant on one run is not that; consistency across assistants and runs is.

What this means for measurement

With overlap this low, a single “AI visibility” number for an agent is close to meaningless. Measure each assistant separately, for each niche. An agent can be named by Gemini on every downtown condo question and never by ChatGPT, and averaging the two produces a number that describes neither. Count your mentions as an individual and as part of your team or brokerage separately, and decide in advance which one you are trying to grow.

Assistant (model)Answers naming ≥1 providerProviders per answerCitations per answerAnswers pointing to directoriesNotable habit
ChatGPT (gpt-5-mini)9 of 202.27.254Never a ranked list; cited the Colorado regulator most (8 times); no Reddit, no AskDoss
Perplexity (sonar)9 of 201.619.52Most citations; 20 of the 27 Reddit citations and half of HomeLight’s
Gemini (gemini-3.5-flash)15 of 203.555.84Named providers even on generic questions, often brokerages
Claude (claude-haiku-4-5)8 of 201.553.752Fewest citations per answer; no Reddit
Grok (grok-4.20-non-reasoning)9 of 201.759.355Cited the AskDoss page most often (6 times)

Finding 3: ranking and referral platforms supply a large share of the evidence

The most-cited sources for real estate questions were not portals like Zillow or the agents themselves. They were “top agents in Denver” pages run by agent-ranking and referral platforms.

MOST-CITED DOMAINS · CITATIONS OUT OF 913 · ALL FIVE ASSISTANTShomelight.com“Top agents in Denver” + referral32rickjanson.comone Denver agent’s blog31reddit.comforum threads27listwithclever.com“Top agents in Denver” + referral22askdoss.comone “best agents 2026” page20fastexpert.com“Top agents in Denver” list18linkedin.comprofiles and posts18compass.combrokerage agent profiles17dre.colorado.govColorado Division of Real Estate16effectiveagents.com“Top realtors in Denver” list15usajrealty.coma brokerage’s own blog15Agent or brokerage siteAgent ranking or referral platformOther source
Figure 2. The eleven most-cited domains across 913 citations, by kind: agents’ and brokerages’ own sites, agent-ranking and referral platforms, and other sources. Source: InsightWonder industry benchmark, real estate agents, Denver, Colorado sample, collected 7 September 2026.

HomeLight was the most-cited domain (32 citations, half of them from Perplexity), mostly its “Top Real Estate Agents in Denver, CO” page. Clever (listwithclever.com, 22), FastExpert (18), EffectiveAgents (15), AgentPronto (10) and Expertise.com (7) followed with pages of the same kind. Together with the single AskDoss page, these platforms supplied roughly one citation in seven. Zillow’s agent reviews (13), Realtor.com (9) and Homes.com (7) added a smaller portal layer. Overall, about one citation in four had “best” or “top” in its title or URL.

Why these pages are cited so often

They are written in the exact form of the question: “top agents in Denver”, with names, numbers and short descriptions. Many are generated from transaction data or reviews the platform collects, and some are built around a referral model in which the platform introduces the buyer or seller to an agent. An assistant looking for a list of Denver agents finds a list of Denver agents. That makes them efficient evidence for an assistant and an important channel for an agent. It also means the description of you in the answer may be the platform’s description, based on whatever data it holds.

Reddit and LinkedIn: the two people-shaped sources

Two other sources stand out because they describe agents through people rather than rankings. Reddit was cited 27 times across 14 of the 20 questions, 20 of them by Perplexity. The threads were the kind a buyer would search for themselves: “Denver Real Estate Agent recommendation?” and recommendations for first-time buyers. ChatGPT and Claude did not cite Reddit at all in this run, so a strong reputation in local threads reaches buyers mainly through Perplexity, with occasional citations from Gemini and Grok.

LinkedIn was cited 18 times, and its pattern is more useful than it looks. On “What certifications or credentials should I look for in a Denver real estate agent?”, 7 of the citations were individual agents’ LinkedIn profiles, used as examples of designations in practice: profiles whose headlines list credentials such as CRS, GRI or SRES. If you hold designations, list them consistently on your LinkedIn profile, your site and your brokerage profile, with a line on what each one means for a client. The assistants are already using such profiles to explain what to look for.

What to do about them

Find out which platforms appear in answers for your niches, and what each says about you. Where you have a profile, make sure the name, brokerage, service areas, specialisms and contact details match your own site exactly. Where a platform runs a referral model, understand the terms before you join; being listed there is a business decision, not a technical fix. Our guide to third-party corroboration covers why several independent sources that agree carry more weight than any one of them.

Finding 4: one uncorroborated page shaped four assistants’ answers

The most important single observation in the Denver data is about one page.

One ranking page“Best Real Estate Agents inDenver 2026”, askdoss.com(single URL)Cited 20 timesIn 20 answers, on 10 of the20 questions, by Perplexity,Gemini, Claude and GrokNames repeatedIndividual agents, their pricebands and deal counts, quotedas fact in the answersNo second sourceWe found no other cited pageabout those agents in thoseanswers; ChatGPT never cited itPage now goneOn 22 October 2026 the URLreturned HTTP 410 (gone);the answers had already used itOne uncorroborated page can decide who four assistants name, until something replaces it
Figure 3. How a single ranking page on askdoss.com supplied individual agent names to four assistants. Citation counts are from the benchmark; the HTTP 410 response is our own check of the URL on 22 October 2026. Source: InsightWonder industry benchmark, real estate agents, Denver, Colorado sample, collected 7 September 2026.

A page titled “Best Real Estate Agents in Denver 2026” on askdoss.com was cited 20 times, once in each of 20 answers, on 10 of the 20 questions. The answers came from Perplexity (5), Gemini (5), Claude (4) and Grok (6); ChatGPT never cited it. Through it, a cluster of individual agent names entered the answers, each with specific claims: a price band, an average number of days on market, a count of condo transactions. Several of these names appeared in four assistants’ answers. In those answers, we found no second cited source about those agents. We did not verify whether they are licensed Colorado agents, and so we do not name them here. When we tried to open the page on 22 October 2026, the server returned HTTP 410, meaning the page has been removed.

Why this matters for every agent

The lesson is not about one website. In a category where assistants agree so little, a single page that answers the question in its own words can decide who gets named, across several assistants at once. This cuts both ways. Agents who are absent from such a page lose those answers to whoever it names, whether or not that is deserved. And a buyer who acts on such an answer may be acting on a source that no longer exists. The assistants did not flag that the agents rested on one source.

What an agent can do

  • Be the corroborated option. An agent whose name, licence and track record appear consistently on their own site, their brokerage profile, a portal profile and at least one ranking platform is far easier for an assistant to confirm than a name on a single list.
  • Publish your own checkable numbers. Single-source pages win because they offer specifics. Give the specifics yourself: sales by neighbourhood and price band, dated and verifiable against MLS records, with your licence number on the page.
  • Monitor the sources, not just the names. When a page like this appears in your niche, you will see it in the citations before you understand why the names changed.

Finding 5: agents’ own content works, but the credit often goes to the brokerage

Individual agents’ own sites did well as sources. The second most-cited domain in the whole run was rickjanson.com, the site of Rick Janson, a Denver agent with Compass. It had 31 citations: His guides on choosing an agent (“How to Find the Best Real Estate Agent in Denver” and others) were cited 13 times on the question about what to look for in an agent, and 10 times on the buyer’s agent versus seller’s agent question. Sallie Simmons’s site (14 citations), The Schlichter Team’s homepage (12) and Usaj Realty’s blog (15) followed the same pattern. In total, when a provider was named, its own site was cited in the same answer a little over four times in ten. That is about the same rate as in our law-firm sample, and lower than for dentists.

FOUR PAGES DESCRIBE THE SAME AGENT; THE ANSWER NAMES ONE ENTITYAgent’s own site and blogrickjanson.com: 31 citationsBrokerage profile pagecompass.com/agents/…: 17 citationsPortal profile and reviewszillow.com: 13 citationsRanking and referral platformsHomeLight, Clever, FastExpert …Assistantpicks anameBrokerage brand“Compass”: 13 answers,8 questionsThe individual agentRick Janson LuxuryProperties: named once
Figure 5. Four kinds of page describe the same agent, but the answer names one entity, often the brokerage. Counts are citations and named answers from the Denver data. Source: InsightWonder industry benchmark, real estate agents, Denver, Colorado sample, collected 7 September 2026.

Cited is not the same as named

Those 31 citations of rickjanson.com produced one mention of his business (“Rick Janson Luxury Properties”, once, by Gemini). His guides were cited on research questions, which rarely name anyone, so his content shaped the advice without putting his name in it. The luxury question shows the same split from the other side. Its answers cited individual luxury agents’ own websites, but the names the answers gave were mostly brokerages: Compass (four assistants), LIV Sotheby’s International Realty (three), Douglas Elliman and Coldwell Banker Global Luxury. “Compass” was named in 13 answers across 8 questions, more than any other provider, while Compass agent profile pages were cited 17 times. Variants such as “Compass Denver” and “Compass Real Estate” added more.

How to make the credit land on you

  • On every page of your own site, state who you are and where you work in one sentence an assistant can lift: “[Name] is a [city] real estate agent with [Brokerage], licensed in Colorado (licence number).”
  • Put your name in the title of your niche pages (“Downtown Denver condos with [Name], [Brokerage]”), not just the topic. A guide titled only by topic will be cited for the topic.
  • Use the same form of your name and your team name everywhere: site, brokerage profile, Zillow, Realtor.com, ranking platforms, LinkedIn. If your team has a name, decide whether the team or the agent is the public entity and stay consistent.
  • Add RealEstateAgent structured data for yourself with your name, brokerage, service area and sameAs links to those profiles. It removes ambiguity; it will not get you named on its own. Our guide to entities and sameAs has the details.

If you run a brokerage

The brokerage is the entity that most often gets the credit, which makes its site an asset for its agents rather than a competitor to them. Compass agent profile pages were cited 17 times in this run, more than most individual agents’ own sites. A brokerage can make those profiles work harder for its agents without much effort. Make each agent profile crawlable as plain HTML. Include the agent’s licence number, service areas and specialisms in text, not only in a photo caption. Link each profile to the agent’s own site and niche pages, and use the same name format as the agent uses everywhere else. Office pages should name the agents who actually work in each neighbourhood or niche, so that an answer naming the brokerage for “downtown condos” can be traced to a person. A profile page that is only a headshot, a phone number and a contact form gives an assistant a brand to name and nobody to recommend.

Self-declared “best realtor” pages

Usaj Realty’s posts titled “Best Realtor in Denver for Condo Buyers” and “Downtown Denver Condo Expert” were cited 9 times on the condo question, and three assistants named the firm. A page that declares itself the best gets repeated. That is a temptation, and a risk. The claim is the firm’s own, and an assistant that repeats it gives it an authority it did not earn independently. Advertising rules for licensees and the REALTOR® Code of Ethics require that claims in advertising be truthful. A page built on specific, checkable facts (buildings worked in, number of condo sales, years in the downtown market) gives an assistant the same material without the risk.

The GEO for real estate playbook for agents and teams

Here is the order we would work in for an individual agent or a small team. A brokerage can apply the same steps per office and per niche.

01Pick nichesNeighbourhood, price band,property type, situation(relocation, fast sale).02BaselineFive assistants, Denver inevery question, same day.Log names and sources.03Niche pagesOne page per niche withyour own checkablenumbers and dates.04One entityAgent name, team andbrokerage written thesame way everywhere.05Platform profilesZillow, Realtor.com, thebrokerage page, referralsites you actually use.06Re-measureMonthly, same questions.Per assistant. Check whothe sources say you are.Each round’s lost niches become the next round’s content brief
Figure 6. The order of work in a GEO programme for a real estate agent or team; a brokerage repeats it per office and niche. Source: the findings in this article.

Step 1: choose your niches

List the situations you want clients from. Use the seven in the table above as a starting point: neighbourhood selling, investment, fast sales, condos, luxury, relocation and price bands. Pick two or three where you have a real track record. For each, write questions as a buyer or seller would ask them, with the city or neighbourhood in every question. Add one trust question and one process question per niche for the citation side. Never put your own name in a question. Our guide to building a prompt set covers balance and size.

Step 2: measure a baseline

Run every question in each of the five assistants on the same day. Record whether you, your team or your brokerage was named, who else was named, and which pages were cited. Note which ranking platforms and portals appear, because those are the profiles you will audit in step 5, and note any single page that seems to drive several assistants’ names. Because the overlap is so low, run each question at least twice; a single run will mislead you about where you stand.

Step 3: publish one page per niche, with your own numbers

For each niche, publish a page that answers the situation directly. A downtown condo page should cover the buildings, HOA issues buyers ask about and how you handle inspections. It should also give your own sales in that segment, dated, with the source of the figures. A relocation page should cover areas, commute patterns and price bands, written from your actual experience. These pages compete with ranking platforms on their own ground: specifics. The difference is that they are verifiable and they are yours.

A worked outline: a downtown condo page

To make step 3 concrete, here is the outline we would use for the downtown condo niche, the question where a brokerage blog was cited 9 times in this run. Each part answers something buyers ask, and each gives an assistant a specific, attributable sentence to lift.

  1. Title and first paragraph: “Buying a condo in downtown Denver with [Name], [Brokerage]”, then two sentences on who you help (first-time buyers, investors, downsizers) and in which neighbourhoods.
  2. Your record in this segment: how many downtown condo purchases you have handled and over what period, with the source (your MLS history) and the date you last updated the figure.
  3. Buildings and areas: the buildings and neighbourhoods you know well, with one or two factual notes each, such as age of the building or the parking and storage situation.
  4. HOA and financing issues: what buyers should ask about HOA reserves, special assessments and rental restrictions, and how those affect financing. This is the section most likely to be quoted in research answers.
  5. How you work with buyers: the written buyer agreement you will ask them to sign before touring, what it covers and how your compensation is set.
  6. Questions buyers ask you: five to eight short, self-contained answers.
  7. Who you are: licence number, brokerage, designations, links to your profiles on the brokerage site, Zillow and Realtor.com.

Step 4: make yourself one entity

Write your name, your team name and your brokerage the same way on every page and profile. Decide which entity is public, you or the team, and make the other clearly subordinate (“[Name], part of [Team] at [Brokerage]”). Link each profile to the others.

Step 5: claim and correct the platform profiles

Check the ranking and referral platforms that appear in answers for your niches, plus Zillow and Realtor.com. Correct your service areas, specialisms and brokerage, and link to your niche pages where the profile allows it. Ask past clients for reviews in the same neutral way every time, on more than one platform.

Step 6: write the process content buyers and sellers need now

Since the NAR settlement’s MLS policy changes took effect on August 17, 2024, offers of compensation can no longer be published on an MLS. MLS participants working with a buyer must also have a written agreement with that buyer before the buyer tours a home. Assistants answering “buyer’s agent vs seller’s agent” or “is it worth paying a higher commission” need current explanations, and in this run the commission question drew sources from as far away as Australia and Singapore. A plain, dated explainer of how buyer agreements and commissions work in Colorado, written by a local agent, fills a real gap.

Step 7: technical basics

Make sure AI search crawlers can fetch your site; IDX plugins and agent-website platforms sometimes sit behind bot protection that blocks them. Keep niche pages as text in the HTML, not only inside a listings widget. If your sold-listing history only appears inside an embedded widget, summarise it in text too.

Step 8: re-measure monthly, sources first

Re-run the same questions monthly. Look first at the sources: is your niche page cited, has a new ranking page appeared, has a platform changed what it says about you? Then look at the names. With this much variation between assistants and runs, act on changes that hold across two runs, not on one.

Licensing, advertising and fair housing guardrails in GEO for real estate

Real estate licensees work under advertising rules and fair housing law that apply to every page an assistant might quote.

Truthful, checkable claims

State licensing rules and the REALTOR® Code of Ethics require advertising to be truthful and not misleading. That applies to “best agent” language, to sales statistics and to claims of specialisation. If you publish numbers, say where they come from and when, and keep them current. An assistant will repeat the number long after you have forgotten it is on the page.

Reviews

The FTC’s final rule on fake reviews and testimonials, announced on August 14, 2024, bans fake reviews and reviews that misstate the writer’s experience. It also bans compensation conditioned on a positive or negative review, undisclosed reviews by insiders and company-controlled review sites presented as independent. It lets the FTC seek civil penalties against knowing violators. Ask every client the same way and never reward a particular rating.

Fair housing in neighbourhood content

Neighbourhood pages are the pages most likely to be cited for relocation and “which area” questions, and they are where fair housing problems arise. Describe neighbourhoods by facts a buyer can check: housing stock, price ranges, commute times, amenities, schools by name with links to official data. Do not describe who lives there or who a neighbourhood is “good for” in terms of protected characteristics. An assistant quoting your page will quote that sentence too.

ContentWeak or risky versionVersion that is safer and more citable
Self-description“Denver’s #1 agent”“[Name], [Brokerage], licensed in Colorado since [year]; [number] sales in [area] since [year], per MLS records”
Niche page“We know downtown condos”Buildings worked in, HOA issues, your dated sales in that segment
Neighbourhood guide“Perfect for young families”Housing types, price ranges, commute times, named schools linked to official data
Commission pagePre-2024 explanation of MLS compensation offersCurrent, dated explanation of buyer agreements and how compensation is negotiated
ReviewsGift card for a five-star reviewThe same neutral request to every client, on more than one platform

How to measure GEO for real estate without fooling yourself

The Denver data makes the standard measurement traps worse, because the assistants disagree so much and the entity question is so open.

Decide what counts as “you”

Before you count anything, decide whether a mention of your brokerage counts, whether your team name counts, and how you treat variants. In this run “Compass”, “Compass Denver” and “Compass Real Estate” all appeared, and so did team names at Compass. Apply the same rule to every competitor, or your share of voice will be inflated.

Never put your own name in a question

“Is [Agent] good for downtown condos?” will mention you every time and measure nothing. All 20 questions here were generated without any provider name for that reason.

Keep the assistants and the runs separate

Perplexity cited 19.5 sources per answer and Claude 3.75. ChatGPT never cited Reddit or the AskDoss page, while Perplexity accounted for most of the Reddit citations. Average these and you describe no real assistant. Repeat runs, because with overlap this low, one run is a sample of one.

Check the sources behind the names

When a competitor suddenly appears across several assistants, look at the citations. In this run, the most consistent agreement on individual agents came from one page that no longer exists. That is not a competitor you need to copy; it is a source you need to outlast.

Watch the out-of-state answers

On “How do I check if an agent has any disciplinary history or complaints?”, the Texas Real Estate Commission was cited 6 times. That is more than the Colorado Division of Real Estate on that question, alongside California, Arizona and Florida regulators. On the track-record question, sources from Michigan, Indiana, Australia and New Zealand appeared. A Colorado agent who explains how to check a Colorado licence and complaint history gives the assistants a local source they currently lack.

Follow the client past the answer

In Yext’s 2026 survey of people searching for local businesses, only 5% said they act directly on an AI answer. After an AI recommendation, 53% search Google or Bing to verify, 49% visit the business’s website, 42% click the sources the AI cited and 28% look for reviews. For an agent, the niche page that got cited and the profiles that corroborate it are the places a client checks next. On Google’s side, local results depend mainly on relevance, distance and prominence, with prominence including links and reviews. Our guide to local GEO covers it in more detail.

Sources