82.6% of the Houston real estate agents AI names appear in exactly one of 40 areas

How this Houston AI visibility study was run

CiteVantage put 500 real-estate queries covering 40 Houston areas to four AI surfaces between 18 and 22 September 2026, producing 2,000 answers, 1,940 of which passed the geo-clean filter. Every business named and every domain cited was recorded. The full row-level data, the query set and the analysis code are published under CC BY 4.0.

We put 500 queries about Houston real estate, covering 40 named Houston areas, to four AI surfaces in September 2026: ChatGPT, Gemini (API-based), the Gemini web app and Perplexity. That is 2,000 answers, of which 1,940 passed the geo-clean filter after 60 were excluded for answering about the wrong place. Fieldwork ran from 18 to 22 September 2026. For each answer we recorded every business or agent named and every domain cited. Names were extracted heuristically, and every structural percentage is published beside a strict-filter sensitivity check. The study is released under CC BY 4.0 with the row-level data, the query set and the analysis code published alongside it.

Queries

500, across 40 Houston areas and 8 intent tiers, plus metro-wide questions

Surfaces

ChatGPT, Gemini (API-based), Gemini (web), Perplexity

Answers

2,000; 1,940 counted

Excluded

60 answers (3%) about the wrong place

Recorded

Firms named and domains cited, per answer

Fieldwork

18 to 22 September 2026

Validation

400 names hand-labelled, blind to outcome

Extraction

Heuristic; strict filter and a hand-labelled sample published

Licence

CC BY 4.0, reuse with attribution

Prior work

The 150-query study, on which domains engines cite

Your area is the whole market

Across 500 Houston AI queries in September 2026, 84.7% of the extracted names appeared in exactly one of 40 areas, 4,100 of 4,841, and a hand-labelled sample puts the figure for real agents, teams and brokerages at 82.6%. Only 741 names reached two or more areas and only 106 reached five or more.

The firms with genuine breadth are countable. James Krueger appeared in 31 of the 40 areas, Sotheby's International in 22, Coldwell Banker in 22, Better Homes and Gardens Gary Greene in 18, Compass in 18, Alfonso Parodi in 18, Perry Homes in 17 and Jennifer Yoingco in 16. Below that list the field thins fast: only 106 names reached five areas, and 4,100 never left one.

The thinnest fields were League City with 110 distinct names ever returned, Atascocita with 119, Cinco Ranch with 131 and West University with 132, tied with River Oaks, which drew 46 answers. The most crowded were Spring with 327 names, the generic Houston queries with 233 and Alief with 191.

Read the crowded figures with the question count beside them. Spring drew 94 geo-clean answers and the generic Houston queries drew 77, against 48 for most areas, so part of what looks like a denser field is simply more questions asked about that area. The four thinnest fields all sit on the standard 48 answers, which makes them comparable to each other.

For a working agent the practical shape is this: the area you are named in is, in roughly eight cases out of ten, the only area you are named in, and the field you are competing against in a thin area is closer to 110 names than to 50,000.

All 40 areas, with their firm and answer counts, are in the data appendix.

Four surfaces, four markets

On the 442 Houston queries every surface answered in September 2026, 87.2% of the names returned came from only one of the four AI surfaces, 8.5% from two, 3.1% from three, and 51 names, 1.2%, from all four. Hand-checked across all queries, 83.5% of real agents appeared on a single surface.

The surfaces are not even working from lists of the same size. On that controlled subset, ChatGPT named 830 distinct firms, the Gemini web app 866, Gemini API-based 1,378 and Perplexity 1,983.

Overlap between any two of them is small. Gemini API-based and the Gemini web app shared the most, 268 firms, 13.6% of their combined list. ChatGPT and the Gemini web app shared 164, 10.7%. ChatGPT and Gemini API-based shared 182, 9.0%. Gemini API-based and Perplexity shared 197, 6.2%. ChatGPT and Perplexity shared 133, 5.0%. The Gemini web app and Perplexity shared 120, 4.4%, the lowest pair in the study.

The most-mentioned of the firms every surface named are James Krueger (153 mentions), Monica Foster (106), Jamie McMartin (78), Jay Thieme (62), Christy Buck (62), Jill Smith (55), Better Homes and Gardens Gary Greene (55), Cody Scurlock (54), Michele Harmon (50) and Houston Properties Team (48). That is the head of a list of 55 names returned by all four on the all-query basis, out of 4,841 extracted.

Most named firms are named once

A hand-labelled sample puts 56.4% of the real Houston agents, teams and brokerages that AI surfaces named across 500 queries in September 2026 at exactly one mention (95% interval 50.9 to 61.8). As extracted, including names that are not businesses, the figure is 68.0%, and the median name has one mention.

Concentration is real without being extreme. The Gini coefficient across 11,100 mentions is 0.491. The top 1% of firms, 48 of them, take 17.3% of all mentions, and the top 10%, 484 firms, take 47.4%.

What separates the head from the tail is breadth rather than volume. The top 50 firms average 9.3 areas, 3.42 surfaces and 7.02 intents. Everyone else averages 1.3 areas, 1.15 surfaces and 1.42 intents. A firm in the head is present in several different conversations at once, while a firm in the tail exists in one answer to one question.

National brands are a small share of the whole. 149 brand and franchise names took 728 mentions, 6.6% of the total, and the other 4,692 names took 10,372, 93.4%, most of them individual agents and local teams. The brands convert their mentions into breadth more often: 17 of 149 brands reached five or more areas, 11.4%, against 89 of the 4,692 other names, 1.9%.

Same engine, two surfaces

On 479 Houston queries that both Gemini surfaces answered in September 2026, the API-based surface named 7.8 firms per answer and the Gemini web app named 3.9. They named 3,739 and 1,866 firms respectively, shared only 352, and on roughly 57% of queries they named nobody in common. Any claim about Gemini visibility has to say which surface.

Across the 479 Houston queries that Gemini API-based and the Gemini web app both answered in September 2026, the API-based surface named 7.8 firms per answer against 3.9 on the web app, and the two surfaces shared only 352 of the firms they named, 6.7% of their combined list. API-based named 3,739 firms in total, the web app 1,866.

Query by query the divergence is starker. The two surfaces shared at least one firm on 189 of 440 queries, 43.0%, which leaves roughly 57% of queries where the same engine, asked the same question through two surfaces, returned no firm in common at all. Both surfaces answered the same 479 queries in the same five-day window, so the gap is a property of the surfaces rather than of the sample.

The web app also behaves like a local search product in a way the API-based surface does not. Google Business Profile markers, star ratings and opening hours, appeared in 88 of 479 Gemini web answers against 27 of 479 API-based answers.

The consequence for anyone measuring is direct: a report saying an agent is visible in Gemini, or invisible in Gemini, is close to meaningless unless it names the surface it measured. Both are Gemini, and they return different Houston markets.

Queries

479 answered by both Gemini surfaces

Firms per answer

7.8 API-based, 3.9 web

Firms named

3,739 API-based, 1,866 web

Shared

352 firms, 6.7% of the union

No firm in common

251 of 440 queries (57%)

Business Profile markers

88 web, 27 API-based, of 479

Change the question, change the list

In 500 Houston queries across four AI surfaces in September 2026, 94.0% of the firms named in new-build answers were absent from the generic discovery answer for the same area. Investment queries displaced 76.1% of names, luxury 73.4%, condo 59.3%, sell 58.7%, first-time 56.9% and relocation 52.2%.

The new-build number is the one to look at twice. Almost the entire list of firms an AI surface names for a new-construction question in a Houston area is a list of firms it does not name when asked that same area's general question about the best agent. New-build answers also run long, naming 8.2 firms per answer against 5.6 for discovery, and 80.4% of the firms in them sit on a single surface.

Builders are visible in a way agents are not. Perry Homes reached 17 of 40 areas, placing it among the eight widest-reaching names in the study, without being a brokerage at all.

Sections 3 and 7 read together give the addressable unit of AI visibility in Houston: an area crossed with an intent, and this study covers 40 areas and eight intent tiers.

What did not separate the winners

Three common explanations for AI visibility were tested against the 500-query Houston data from September 2026 and none of them separated the named from the unnamed: presence in agent referral networks sits 2.5 points above a 42.4% base rate, a 17x spread in production produced a 1.17x spread in mentions, and website size ran in the wrong direction.

Three explanations for why the AI surfaces name one Houston firm and skip another were tested against the 500-query September 2026 data, and none of them separated the winners. Presence in agent referral networks came first: across the top 50 firms, 44.9% of the answers naming them also cited a referral network, against a base rate of 42.4%, a deviation of 2.5 points. Answers that cite a referral network name more firms, so any firm lands in one at close to the base rate by chance.

Production came second. Agent-level research collected alongside the study found a 17x spread in closings across four brands producing a 1.17x spread in mentions. Alfonso Parodi closed 10 transactions in the trailing 12 months according to his US News real estate agent profile, read 23 September 2026, and was named 48 times; Cody Scurlock closed 87 in the same period according to his US News profile, read the same day, with 390 lifetime sales according to his HAR bio, and was named 54 times. The closing counts are secondary sources rather than our measurement; the mention counts are ours.

Website size came third, and ran backwards. Jennifer Yoingco's site carries 244 blog posts according to its own post sitemap, counted 23 September 2026, and she was named 49 times; Jay Thieme had no site reachable at the time of research and was named 62 times.

What the answers do lean on is a thin band of intermediaries. Of 7,576 citations, 29.8% went to portals (Zillow 9.1%, HAR 8.6%, Realtor.com 6.3%) and 13.6% to agent referral networks (FastExpert 4.2%, HomeLight 3.4%, EffectiveAgents 3.3%, AgentPronto 1.3%), 43.4% to intermediaries in total. The mix moves by surface: ChatGPT cited portals in 47.8% of its citations and referral networks in 10.8%, Gemini API-based 27.1% and 17.4%, Perplexity 29.4% and 10.2%. Gemini web answers in this dataset do not carry cited URLs, so the split is reported for three surfaces.

The engines are mostly not reading the agent's own website

In the 500-query Houston study of four AI surfaces in September 2026, a firm's own domain appears in the citations of an answer naming it between 8.4% and 10.4% of the time, and the rate is flat from the 50 most-named firms down to the 3,291 named once. Portals and agent referral networks supply the answer, and the firm's own site is a minority of the mechanism.

Across 1,940 geo-clean answers from four AI surfaces about Houston in September 2026, a firm's own domain appeared in the citations of an answer naming it between 8.4% and 10.4% of the time, depending on the tier, with no meaningful gradient between the head and the tail. The top 50 firms sit at 8.8% per answer and 34.0% ever. Firms ranked 51 to 500 sit at 10.4% per answer and 20.2% ever. Firms named twice or more sit at 8.7% and 14.5%. Firms named exactly once sit at 8.4% and 8.4%.

This is the finding that costs us money to publish. We sell AI visibility work, and a large part of what any agency in this category sells is work on the client's own website. The Houston data says the agent's own site is a minority of the mechanism behind an answer that names them, showing up in roughly one such answer in ten. We publish it anyway, at full strength, with the rows behind it.

Read it carefully in both directions. It says the surfaces assemble a Houston answer from portals and agent referral networks first, so being absent from those sources is expensive. It also has a real limit: the measurement records which domains an answer cited, and a surface can read a site during retrieval without citing it, so the 8.4% to 10.4% band measures citation and acts as a floor on influence. What it removes is the assumption that publishing more on your own domain is the lever that gets you named.

Nobody named, and the wrong place

7.4% of the 1,940 geo-clean Houston answers collected across four AI surfaces in September 2026 named no firm at all, ranging from 11.8% on Perplexity to 1.9% on Gemini API-based. Separately, 60 of 2,000 answers, 3.0%, answered about the wrong place and were excluded from every figure on this page.

Of the 1,940 geo-clean answers to 500 Houston queries across four AI surfaces in September 2026, 143 named no business or agent at all, 7.4%. The rate splits hard by surface: Perplexity returned no name on 11.8% of its answers, ChatGPT on 9.3%, the Gemini web app on 6.4% and Gemini API-based on 1.9%. By intent, the metro-wide questions were likeliest to return nobody at 13.0% and luxury questions least likely at 2.6%.

Geography failed separately and more interestingly. 60 of the 2,000 answers, 3.0%, answered about somewhere other than Houston and were dropped before any other figure was computed. Queries carrying a state marker such as "TX" or "Texas" failed at 1.1%, 10 of 880. Queries without one failed at 4.5%, 50 of 1,120.

On ChatGPT these failures were stable rather than random. Of 37 ChatGPT geo failures re-asked two days later, 29 were still wrong, and 27 of those landed in the same wrong city as the first time; 7 corrected themselves and 1 was unclear.

ChatGPT also answers in a different register from the others. It reasoned rather than simply handing over names in 97.6% of its answers, and across all surfaces the median reasoning answer ran 2,688 characters.

How many Houston agents are ever named

Four AI surfaces answering 500 Houston real-estate queries in September 2026 returned 4,841 distinct names. A hand-labelled sample puts roughly 3,000 of them (95% interval 2,800 to 3,267) as real agents, teams and brokerages. Against the Houston Association of Realtors' own figure of more than 50,000 members, that is about 6.1% named at least once.

The denominator is HAR's own published figure, which is why we use it, and the point survives any plausible alternative. Whether the true count of active Houston agents is 30,000 or 50,000, the share of them that four AI surfaces will name across 500 questions stays small: about one in sixteen on HAR's figure, and about one in ten even on the lower one.

FlyDragon's national report reached a convergent number from a different method, finding 91% of agents effectively invisible in AI answers. Their cross-surface figure runs the same way: of the agents appearing on at least one AI surface, they report only 11% appearing on three or more. Our equivalent on the controlled subset is 4.3%, being 3.1% on three surfaces plus 1.2% on four. Same direction, same order of magnitude, ours stricter.

Does this hold in other metros?

This study measured one metro at depth: 500 queries, 40 Houston areas, four AI surfaces, September 2026. We have not measured any other metro, and we will not claim the Houston percentages hold elsewhere until we have. The shape it shares with a national study of cited domains is the part most likely to transfer.

Our 150-Query Real Estate AI Search Study found 83.0% of 1,297 cited domains reaching only one of five engines nationally, and this study finds the same pattern at business level in Houston. FlyDragon's market atlas covers many US metros, and we do not contradict their Houston head: their top-ranked Houston agent is also our top-ranked agent, and 20 of their 25 ranked Houston agents appear in our data, from a page stamped 23 July 2026, two months before our fieldwork.

On whether any of this reaches a buyer, the evidence cuts both ways. For: Pew Research Center (fieldwork March 2025, published July 2025, n=900, 68,879 searches) found users clicked a result on 8% of visits with an AI summary against 15% without one. Ahrefs measured a 34.5% lower click-through rate for the top result on keywords with an AI Overview in April 2025, and 58% lower when it repeated the measurement in February 2026. Bank of America's 2026 Homebuyer Insights Report (June 2026, n=2,000) found 20% of prospective buyers and homeowners had used AI tools for homebuying research in the past year.

Against, at full strength: Similarweb measured 1.13 billion AI referrals against 191 billion from Google in June 2025 across the top 1,000 websites, roughly 0.6%; the NAR 2025 Profile of Home Buyers and Sellers (4 November 2025, n=6,103) found 88% of buyers and 91% of sellers used an agent, and 91% of buyers would recommend theirs; and the portals now sit inside the surfaces themselves, Zillow in ChatGPT since October 2025 and Realtor.com since March 2026. No study anywhere isolates consumers using AI to choose an agent, so the stakes remain unmeasured.

What a Houston agent can do about this

The 500-query Houston data from September 2026 points to six moves: choose one area and one intent rather than the metro, measure each of the four AI surfaces separately, keep the Google Business Profile accurate for the Gemini web app, get present in the portals and agent referral networks the answers actually cite, write the state and metro into your copy, and remeasure.

Pick one area and one intent, and treat that as the whole target. A hand-checked 82.6% of named Houston agents hold exactly one area, and new-build answers exclude 94.0% of the names in the same area's generic answer, so "best realtor in Sugar Land" and "new construction in Sugar Land" are two separate contests. In a thin area such as League City the surfaces have named 110 distinct firms in total, which is a field you can picture.

Measure each surface on its own. 83.5% of real agents appear on one surface only, the widest overlap between any two surfaces is 13.6% of their combined list, and the two Gemini surfaces share nobody on roughly 57% of queries. A single screenshot from one surface tells you almost nothing about the other three.

Keep the Google Business Profile accurate and current, specifically for the Gemini web app, where star ratings and opening hours appeared in 88 of 479 answers against 27 on the API-based surface.

Be present where the answers are actually assembled. Portals and agent referral networks account for 43.4% of all citations, while a firm's own domain shows up in roughly one answer in ten that names it, so a complete and current presence on HAR, Zillow and Realtor.com is doing more of the visible work than another post on your own blog.

Write "Houston, TX" into the copy people quote you in. Queries carrying a state marker resolved to the wrong place 1.1% of the time against 4.5% without one.

Then remeasure. This is one run in one five-day window, and the answers move.

Method and limits

CiteVantage wrote 500 Houston real-estate queries across 40 areas and eight intent tiers, put each to four AI surfaces between 18 and 22 September 2026, and recorded every firm named and every domain cited in all 2,000 answers. 60 answers, 3.0%, were excluded for answering about the wrong place. Names were extracted heuristically and every figure is published with a strict-filter check.

Query design. The study covers 40 named Houston areas and eight intent tiers: discovery, sell, first-time, invest, relocate, luxury, new-build and condo, plus 20 metro-wide questions asked about Houston as a whole. 38 of the 40 areas carry 12 queries each, four discovery, two sell and one for each of the six remaining intents; Spring carries 24, and the generic Houston set is the 20 metro-wide questions. Queries were written in two shapes, 250 search-shaped and 250 conversational, so the same underlying question was asked both the way a person types it into a search box and the way a person asks a chatbot.

Surfaces. Four surfaces, each queried directly: ChatGPT through its web app, Perplexity through its web app, Gemini through its web app, and Gemini API-based. The two Gemini surfaces are reported separately throughout, because, as the Gemini section above shows, they return different Houston markets. Anywhere this page says "Gemini API-based" it means rows collected through the API, nothing more.

The geo-clean filter. Every answer was checked for whether it answered about Houston at all. 60 of 2,000 answers, 3.0%, did not, and those were removed before any other number was computed, so every percentage on this page runs off the 1,940 geo-clean answers. The failures are reported above, in the section on wrong-place answers.

Extraction is heuristic. Business and agent names were pulled from the answer text by a script. A script mistakes some phrases for names and misses some real ones, so every structural percentage is published three ways. The strict filter deletes 24.2% of names, real firms included, but still lets some non-businesses through, so it is a stress test rather than a floor. The headline figures come from a third method: 400 extracted names drawn at random and labelled by hand, blind to how often or where each was named, then re-weighted to the full list with bootstrapped 95% intervals. The labels are published so any of them can be disputed.

One run, one window. Fieldwork ran across five days, 18 to 22 September 2026. AI answers move between runs, so this is a snapshot rather than a ranking, though some of the movement is smaller than it looks: on ChatGPT, 29 of 37 retried geo failures were still wrong two days later. Anyone relying on these numbers for a decision about a specific firm should remeasure.

We sell this. CiteVantage sells AI visibility work to real-estate firms, so our commercial interest is direct and declared. The study was conducted independently, was not commissioned by anyone, has no vendor platform underneath it, and every row is published. The own-website finding above cuts against the easiest thing we could sell, and it is published at full strength.

Everything is published. The row-level data, the query set, the hand labels and the analysis, sensitivity and validation scripts all ship with the study, so every figure on this page can be recomputed rather than taken on trust.

Downloads: data, queries and code

• Row-level data, CSV: all 2,000 answers, one row each, with the firms named and the domains cited.
• Named-firm appendix, JSON: every one of the 4,841 canonical names with mention count, areas, surfaces and intents.
• The query set, JSON: the 500 queries as asked, with area, intent and shape, published alongside so the wording is not taken on trust.
• Analysis code: study500_findings.py, study500_market.py, study500_sensitivity.py, study500_validation.py and a README that recomputes the headline from the CSV.

All 40 areas

Press kit: cite, download and reuse this study

Everything here is free to republish under CC BY 4.0, including commercially, with attribution and a link back. The row-level data is published alongside the findings so any figure on this page can be recomputed rather than taken on trust. No permission request is needed and there is no embargo.

How to cite it

For journalists

Questions about this Houston AI visibility study