Houston neighbourhoods, November 2026 edition
Three AI engines named a common Houston firm on only 6.8% of neighbourhood questions
By Abdul Subkhan, CiteVantagePublished 5 November 2026Fieldwork 13 October 2026 to 16 October 2026
ChatGPT, Perplexity and Gemini answered 450 questions about 15 Houston neighbourhoods (13 to 16 October 2026), and 1,339 answers were counted. On the 381 questions where all three named someone, they named a common firm on 6.8%. About 1,118 real firms were named, and 70.3% appear in only one neighbourhood.
1,339 answers, 15 neighbourhoods
6.8%
of 381 questions had one firm named by all three engines
- ChatGPT and Perplexity, no firm in common 72.7%
- ChatGPT and Gemini, no firm in common 58.1%
- Perplexity and Gemini, no firm in common 76.2%
-
6.8%
of 381 questions had a firm named by all three engines
-
70.3%
of real firms named appear in one of 15 neighbourhoods (labelled, interval 69.4 to 71.4)
-
27.8%
of an area's answers name its most-named local firm, in the median neighbourhood
Three engines agree on one firm for 6.8% of questions
On 381 Houston neighbourhood questions where ChatGPT, Perplexity and Gemini all named at least one firm (13 to 16 October 2026), the three engines named a common firm on 6.8%. On 37.3% of those questions, no two engines shared any firm at all. Each engine builds its own short list.
A firm returned by one engine says little about what the other two will return. The next section shows which pairs agree most.
6.8%
of the 381 questions that all three engines answered with names had a firm named by all three
Questions where a pair of engines shared no firm at all
-
ChatGPT and Perplexity 72.7% 384 questions both named a firm - ChatGPT and Gemini 58.1% 434 questions both named a firm
- Perplexity and Gemini 76.2% 382 questions both named a firm
Most firms are named in one neighbourhood only
Of the real firms named across 15 Houston neighbourhoods (1,339 counted answers, 13 to 16 October 2026), 70.3% appeared in exactly one neighbourhood and 70.5% on exactly one engine. As extracted, with non-businesses included, the one-area figure is 71.1%. 49.2% of real firms were named exactly once.
A firm named for The Heights is, in most cases, one no engine names for Alief or Copperfield. Breadth is rare and countable, and the downloadable named-firm file lists how many areas each firm reached.
ChatGPT and Perplexity share no firm on 72.7% of questions
On 384 neighbourhood questions that both ChatGPT and Perplexity answered with names (13 to 16 October 2026), the two shared no firm on 72.7%. For Perplexity and Gemini the figure is 76.2% of 382 questions, and for ChatGPT and Gemini 58.1% of 434.
ChatGPT and Gemini agree most, and Perplexity is the odd one out. Perplexity gives short answers, with 3.29 names per answer on average against 5.19 for ChatGPT and 5.23 for Gemini, so part of the gap is answer length.
The most-named local firm in each neighbourhood
Share of each neighbourhood's counted answers that name its leading firm. 1 of 15 neighbourhoods reach 40%; the spread runs from 18.8% in Alief to 41.6% in Copperfield.
About 1,118 real firms were named in 15 neighbourhoods
The three engines returned 1,173 distinct names across 15 Houston neighbourhoods (13 to 16 October 2026). A labelled sample puts about 1,118 of them as real individual Realtors, teams or brokerages (95% interval 1,088 to 1,144). About 4.7% of sampled names were not businesses, such as headings and place names.
The neighbourhood count sits inside the combined figure on the report page, where it is set against HAR's 50,000-plus members.
A small group of firms takes a large share of mentions
In the 15-neighbourhood study (1,339 counted answers, 13 to 16 October 2026), the 12 most-named firms, the top 1%, took 21.1% of all mentions and the top 10%, 117 firms, took 59.7%. The Gini coefficient is 0.677. The median neighbourhood's leading local firm appeared in 27.8% of answers.
Only 1 of 15 neighbourhoods has a local leader at 40% or more, in Copperfield at 41.6%. The lowest leader share is 18.8% in Alief. National brands were set aside when choosing each leader. Suburbs look different, and the report page has that contrast.
The most-named firms across all 15 neighbourhoods
| Firm | Mentions | Neighbourhoods | Engines |
|---|---|---|---|
| Corcoran Prestige Realty | 197 | 14 | 3 of 3 |
| James Krueger | 148 | 13 | 3 of 3 |
| Martha Turner Sotheby's International Realty | 145 | 13 | 3 of 3 |
| EXP Realty | 115 | 14 | 3 of 3 |
| Keller Williams Realty | 105 | 15 | 3 of 3 |
| Greenwood King Properties | 102 | 10 | 3 of 3 |
| Paige Martin | 101 | 14 | 3 of 3 |
| Found Realty Group | 100 | 12 | 3 of 3 |
| Jennifer Yoingco | 95 | 12 | 3 of 3 |
| Better Homes and Gardens Real Estate Gary Greene | 67 | 10 | 3 of 3 |
Where the answers come from
Of 1,597 sources cited by ChatGPT and Perplexity in the neighbourhood study (13 to 16 October 2026), 40.1% went to portals, 14.2% to agent referral networks and 33.4% to individual, team, builder or local sites. HAR's own site took 18.2% of citations and Zillow 10.8%. Gemini's web app showed almost no URLs.
A firm's own website is a minority of the mechanism. Across the tiers we measured, a firm's own domain appeared in between 3.4% and 5.3% of the answers naming it. We sell AI visibility work, and this finding argues against the easiest thing to sell, so we publish it at full strength. A model can read a page without citing it, so the figure is a floor on influence.
- har.com: 290 citations, 18.2%
- zillow.com: 172 citations, 10.8%
- agentpronto.com: 93 citations, 5.8%
- realtor.com: 74 citations, 4.6%
- homes.com: 57 citations, 3.6%
Change the wording and the list changes
Near-synonym questions about the same neighbourhood and engine returned lists with no firm in common 38.6% of the time for "best agent" against "top five" (44 pairs) and 48.8% for "best agent" against "most trusted" (41 pairs), in the 13 to 16 October 2026 fieldwork. "Best agent" against "top realtors" was 15.9%.
Intent moves the list too. Of the names returned for rental questions, 77.6% (125 names) were absent from the generic best-agent answers for the same neighbourhood, and for investor questions 75.1% (197 names).
Each engine answers in its own way
In the neighbourhood study (13 to 16 October 2026), ChatGPT named 5.19 firms per answer, Gemini 5.23 and Perplexity 3.29. Perplexity named nobody in 12.3% of its 446 answers, against 0.2% for ChatGPT and 1.1% for Gemini. Its median answer ran 390 characters, and ChatGPT's 2,727.
That is why we report each engine separately and treat pooled counts as a floor.
Method and limits for this study
This study asked 450 questions about 15 Houston neighbourhoods once each of ChatGPT, Perplexity and Gemini between 13 to 16 October 2026, collecting 1,350 answers and counting 1,339. 11 answers were removed because they did not match the question asked. Names are extracted by rules, and a strict filter removes 18.6%.
One run per question means a repeat would give different lists. Extraction is heuristic, and the labelled sample of names was sorted by an AI reviewer, with a person re-checking 31 labels and agreeing on agreed with every one. CiteVantage sells AI visibility work, and this is independent research commissioned by no one. The full method is on the report page.
- Questions
- 450, across 15 neighbourhoods and 30 question shapes
- Engines
- ChatGPT, Perplexity, Gemini; ChatGPT 446, Gemini 447, Perplexity 446 counted
- Answers
- 1,350 collected; 1,339 counted
- Removed
- 11 answers, because they did not match the question asked
- Names
- 1,173 distinct names as extracted, about 1,118 real firms (interval 1,088 to 1,144)
- Labels
- 286 names labelled by an AI reviewer, a person re-checked a sample (31 of 31 agreed)
- Fieldwork
- 13 October 2026 to 16 October 2026
- Licence
- CC BY 4.0, reuse with attribution
Download the neighbourhood data
The neighbourhood study ships 1,339 counted answers as a row-level file with every firm named and every domain cited, plus the named-firm list, the 450 questions and the validation sample, under CC BY 4.0. The files describe answers collected 13 to 16 October 2026. There is no list of firms that were not named.
Downloads: data, queries and code
- Row-level data, CSV: 1,350 answers, one row each.
- Named-firm appendix, JSON: every canonical name with mentions, neighbourhoods and engines.
- The query set, JSON: the 450 questions as asked.
How to cite it
Reference
CiteVantage (2026). Houston Neighbourhoods: AI Visibility, November 2026. https://citevantage.com/research/houston-real-estate-ai-visibility-november-2026/neighborhoods/
In a sentence
A CiteVantage study of 1,339 AI answers found that ChatGPT, Perplexity and Gemini named a common firm on only 6.8% of Houston neighbourhood questions (CiteVantage, 2026).
Questions about this report
Do ChatGPT, Perplexity and Gemini name the same Houston firms?
Rarely. They named a common firm on 6.8% of 381 neighbourhood questions.
The questions were asked between 13 to 16 October 2026. On 37.3% of them, no two engines shared any firm.
Which two engines agree most?
ChatGPT and Gemini, though they still share no firm on 58.1% of questions.
ChatGPT and Perplexity share none on 72.7% and Perplexity and Gemini on 76.2%. Perplexity's shorter answers account for part of its gap.
If a firm is named for one neighbourhood, will it be named for another?
Usually not. 70.3% of real firms appear in only one of 15 neighbourhoods.
That is a labelled estimate for the real firms named in 1,339 counted answers.
Does a firm's own website get it named?
The data cannot show that it does.
A firm's own domain appeared in 3.4% to 5.3% of the answers naming it, across the tiers we measured. Portals took 40.1% of the 1,597 citations. This describes citation only, because a model can read a site without citing it.
CiteVantage sells this service. Is the study self-serving?
We ran it and we sell this work, so treat the bias as declared.
This is independent research, commissioned by no one. The questions, every row, the scripts and the labels are published, so every figure can be recomputed.
Would a repeat give the same answers?
No. Each question was asked once per engine, so a repeat would give different lists.
The results describe the window of 13 to 16 October 2026.
