ChatGPT, Perplexity and Gemini name the same agent for 7.6% of Houston questions, and for 29.7% in Coastal South Florida

What we found

When a buyer or seller asks an AI engine for an agent, the three big engines often disagree about who to name, and the disagreement is much larger in Houston than in the other markets we have enough comparable questions for. On the same kind of question ("who is the best real estate agent in this town"), at least one person or team appeared in the first three names from all three engines for 7.6% of Houston questions. In Coastal South Florida the share was 29.7%, about 4 times higher.

The Houston figure rests on 721 questions, so it is tight. The other markets rest on fewer and their ranges are wider, which the section on agreement shows. We do not yet know why Houston is different, and the section on what we cannot say yet lists what we have not ruled out.

What the real estate AI visibility data holds

Each row is one round of questions put to all three engines. Questions that name a specific business are kept out of the comparisons, because an engine asked about a business will name it. The rules for removing answers are under how it was run.

1. 01

   Coastal South FloridaNov 2026. 200 questions, 600 answers counted of 600 collected.

   600
2. 02

   Coastal South FloridaOct 2026. 235 questions, 1,799 answers counted of 1,800 collected.

   1,799
3. 03

   Houston, TexasOct 2026. 50 questions, 150 answers counted of 150 collected.

   150
4. 04

   Houston, TexasSep 2026. 500 questions, 1,434 answers counted of 2,000 collected.

   1,434
5. 05

   Houston, Texas and Tampa Bay, FloridaOct 2026. 1,190 questions, 3,489 answers counted of 3,572 collected.

   3,489
6. 06

   Inland Southern CaliforniaOct 2026. 100 questions, 299 answers counted of 300 collected.

   299
7. 07

   New Jersey shore and South Florida buildersNov 2026. 200 questions, 599 answers counted of 600 collected.

   599
8. 08

   Suburban New YorkSep 2026. 100 questions, 295 answers counted of 300 collected.

   295
9. 09

   Tampa, Florida, property managementOct 2026. 100 questions, 303 answers counted of 303 collected.

   303
10. 10

    Tampa, Florida, property managementSep 2026. 100 questions, 257 answers counted of 343 collected.

    257

Do the engines agree on who to recommend?

For each question we took the first three people or teams each engine named, and checked whether any one name appears in all three lists. Brokerage brands and portals are left out of this measure. Using the first three names keeps a market from looking more agreeable only because its answers run longer.

Inland Southern California had only 18 plain questions (66.7%) and Tampa Bay, Florida had only 24 plain questions (4.2%), which is too few to compare, so they are left out of the chart. Only markets with at least 30 plain questions are charted.

Across all questions that do not name a business, the same pattern holds: Houston 6.5% (1,307 questions) against 30.1% in Coastal South Florida (743). Those question sets differ more between markets, which is why the chart uses the plain questions.

Which engines agree with each other

The same measure for each pair of engines: the share of questions where both name at least one of the same first three people or teams. ChatGPT and Gemini share a name most often in Houston and in Coastal South Florida. In Suburban New York the three pairs sit within 4 points of each other.

Houston, Texas

ChatGPT and Perplexity 22.7%; ChatGPT and Gemini 30.3%; Perplexity and Gemini 17.3%

Suburban New York

ChatGPT and Perplexity 51.2%; ChatGPT and Gemini 53.5%; Perplexity and Gemini 54.8%

Coastal South Florida

ChatGPT and Perplexity 50.7%; ChatGPT and Gemini 72.2%; Perplexity and Gemini 50.8%

Beyond agents: property managers and home builders

Two smaller sets cover other real estate businesses: 184 questions about property managers in one Florida city, and 190 about home builders and renovators on the New Jersey shore and in South Florida. They are single-region sets, so they show how the engines treat these businesses, not how the whole country looks.

The first difference is who gets named. In the agent answers, 48% of the names were read as individual people. For property managers it was 5.5%, and for builders 22%. A manager or builder is named as a company, so an owner choosing between them is choosing a brand, not a person.

The second is what the question asks. When a property owner described a situation (for example "My old Tampa rental has never been managed professionally. How do I bring in a manager?"), 13.2% of answers named no firm at all, against 0% when the question asked which company to use. For builders, the questions about building rules and costs named no firm in 6.7% of answers, while the questions asking who builds or renovates named one every time.

Property managers

509 answers, 4.08 firms per answer; all three engines named a common firm in 24.6% of questions (95% range 17.2% to 32.8%, 122 questions)

Home builders and renovators

569 answers, 7.58 firms per answer; common firm in 62.4% of questions (95% range 55.4% to 69.4%, 186 questions)

Agents in Houston

4,746 answers, 4.86 firms per answer; common firm in 6.4% of questions (95% range 5.1% to 7.7%, 1,284 questions)

Here a firm is any named business, so these shares are not the person-or-team shares in the chart above. The builder and property-manager sets were written around a few towns and one city, which may push agreement up, so treat the gap with Houston as a pointer and not a ranking.

The same question twice

Part of the disagreement is the engines disagreeing with themselves. We put a random 100 of the Houston questions to ChatGPT and Gemini again, twice, about two weeks after the first answers. For the average ChatGPT question, between 26% and 38% of the firms either of two runs named were named by both. For Gemini the range was 10% to 15%. One answer is a weak guide to who an engine will name next time, which is why this dataset counts across thousands of answers.

The full method and the repeat-run data are in the Houston report.

What we cannot say yet

The Houston gap is measured. Its cause is not. Three explanations are open, and this dataset cannot separate them yet.

• Houston has far more agents competing for each question, so there is less reason for three engines to land on the same one.
• The Houston questions cover dozens of neighbourhoods and suburbs, while the other sets concentrate on a few towns. The plain-question comparison narrows this but does not remove it.
• The rounds were run at different times and with different question sets, and the engines change.

More markets run with the same plain questions would settle it. That is what this page is built to take in.

How the real estate AI visibility data was run

Each question was put to ChatGPT, Perplexity and Gemini through their apps, each in a new conversation, with the answer text and cited sources saved. Names were extracted from the saved text by one shared method across all rounds, and a person or team counts only if the method read it as one. The Houston report describes how the extraction was checked against a labelled sample.

Counted

9,225 answers of 9,968 collected

Removed

94 answers because they did not match the question asked, and 70 because they were too short to read as an answer

Set aside

577 Gemini answers taken through the programming interface, which is a different surface from the app people use

Compared

8,959 counted answers to questions that do not name a business, across 7 markets

Licence

CC BY 4.0, reuse with attribution

Limits

• Rounds differ in date, question set and engine settings, so market comparisons are indicative until more markets are run with the same questions.
• Markets other than Houston and Tampa Bay come from audits CiteVantage ran for real estate businesses. They are shown only as region labels, and the questions in them were written around those businesses.
• Every answer comes from one capture setup. Other setups may give different numbers.
• The name extraction is a method, not a ground truth. It was checked on Houston and has not been separately validated for every market.
• The repeat check covers ChatGPT and Gemini on one 100-question sample. Perplexity was not repeated.

Download the data

real-estate-ai-visibility-markets.csv holds one row per market: answers counted, questions, the agreement share with its 95% range, the plain-question share, and the three engine pairs. A column marks which markets have enough plain questions to compare. The Houston and Tampa Bay firm-level data is in the Houston report.

Cite this

Free to republish under CC BY 4.0, including commercially, with attribution and a link back.

Questions about this dataset