Research · source analysis

Which sources show up around AI recommendations.

Every citation in every captured answer was recorded and classified. This page reports what appeared, how it was classified, how often it appeared, and what that does and does not mean.

Market
Jersey City, NJ · residential (anonymized)
Provider
ChatGPT (OpenAI)
Sample
64 questions × 4 runs · 256 valid captured answers
Captured
August 2026 · last updated 14 September 2026

Retrieved, mentioned, cited — three different things

Retrieved
Pages the provider fetched while composing an answer. We cannot see these. They are never reported here, and any figure claiming to measure them would be a guess.
Mentioned
A source named in the answer text without a link — for example, “according to a local business journal”. Recorded, but not counted in citation frequency.
Cited
A source the answer links or attributes explicitly, with a resolvable URL. This is the only category counted in the figures below.

Conflating these three is the most common way citation figures in this category become meaningless. Only explicit citations are counted below.

How sources are classified

Listing portals
National property search sites carrying agent and team profiles alongside inventory. Classified by domain, regardless of which page on the domain was cited.
Brokerage sites
Brokerage-owned domains, including franchise and office subdomains. Agent profile pages hosted by a brokerage are counted here, not as first-party.
Local publications
Local news outlets, business journals and city magazines. Regional editions of national outlets are counted here when the cited page is local coverage.
Community forums
Discussion sites and question-and-answer threads where the content is user-contributed rather than editorial.
Video
Video platforms, counted when the cited URL is the video or channel itself rather than an article embedding it.
First-party sites
The observed team's own domain. Kept separate from brokerage sites precisely because teams assume their own site is doing work that the data may not support.

Classification is by domain, applied consistently across the whole captured set and fixed before counting begins. A domain belongs to exactly one class, so no citation is counted twice. Where a domain could plausibly sit in two classes, it is assigned to the one describing who controls the content.

Why one answer can cite several sources

A single answer frequently names several teams and supports them with several different links. That means citations cannot be counted per answer without discarding most of the data.

So the unit here is one citation instance, not one answer. A single answer can cite several sources, so citations are counted per citation, not per answer. The consequence: citation counts on this page can exceed the number of answers captured, and that is expected rather than an error.

Source classes, by how often they appeared

Listing portals100
Brokerage sites44
Local publications22
Community forums12
Video7
The team's own site0

Indexed frequency (listing portals = 100). Unit: Indexed source-appearance frequency (most frequent class = 100). Numerator: Citations belonging to that source class within the captured answers. Denominator: Citations belonging to the most frequent class (listing portals). Market: Jersey City, NJ · residential (anonymized). Provider: ChatGPT (OpenAI). Capture date: August 2026. Sample size: 64 buyer and seller questions × 4 per question = 256 valid captured answers. Multi-source rule: A single answer can cite several sources, so citations are counted per citation, not per answer. Observed association only; no causal claim is made.

MeasuredCounted from the captured answer corpus for this market edition.

Cited domains recorded

Jersey City, NJ · residential (anonymized) · ChatGPT (OpenAI) · captured August 2026 · 64 buyer and seller questions × 4 per question · 256 valid captured answers
DomainCitations recorded
zillow.com696
realtor.com680
compass.com304
building websites40
homes.com40
the team's own site0

Counts are citation instances recorded across the captured set, not unique pages. The team's own site was not cited in the captured benchmark responses.

The consistent pattern in this market is that a small number of listing portals and brokerage properties account for most of the citation volume, while the team's own website appears in none of it.

DirectionalCoverage on the sources that appear around answers is a plausible prerequisite for being represented in them. That is a reasoned reading of the evidence, not a demonstrated mechanism.

Limitations

What this cannot prove

  • One market, one provider, one capture window.
  • Citation behavior varies between runs and changes as models and indexes change.
  • A source appearing in an answer does not prove it caused the recommendation.
  • Sources retrieved but not cited are invisible to this method and are not reported.
  • Entities are anonymized; the underlying answers are preserved and available for inspection on request.

How to cite this page

How to cite this page

Recommended First, AI Citation Sources in Real Estate, Jersey City, NJ, captured August 2026, methodology v1.0.

Please reproduce the indexed-frequency note with any figure taken from this page, so the unit and denominator travel with it.

See who AI recommends in your market — and where you stand.

One client per market. Start by telling me where you work.