By Your own website is a minority of what AI cites about you. In Sydney it was 35.0% of the sources three engines surfaced when recommending a real estate agent. In New York, 41.5%. The rest is third-party, and the third party in charge flips by city: New York runs on listing portals like Zillow, Sydney runs on agent directories like RateMyAgent. Same question, same three engines, opposite playbook.
We ran the study to answer one thing real estate agents keep asking us: when someone asks ChatGPT for an agent, where does the answer actually come from? We probed Sydney on 2026-05-22 and New York on 2026-05-21, across OpenAI, Perplexity, and Gemini, then classified every cited URL by its registered domain. Sydney returned 432 cited URLs. New York returned 424. This is what those 856 citations are made of.
- Your own website is never a majority. Sydney 35.0%, New York 41.5%. The other ~65% (Sydney) and ~58% (New York) is third-party and you don't fully control it.
- The dominant gatekeeper flips by city. Sydney's largest third-party type is agent directories at 31.0%. New York's is listing portals at 25.7%.
- Directories are about 4.8x more present in Sydney than New York (31.0% vs 6.4%). Portals are about 2x more present in New York than Sydney (25.7% vs 13.0%).
- The inversion holds on all three engines independently. Every engine cited more directories in Sydney and more portals in New York.
- The own-site share is not engine-stable. In New York it ran from 33% (Perplexity) to 54% (OpenAI), with Gemini at 43%.
- This is one city per country, around 430 citations each. Sydney is not all of Australia. New York is not all of the US.
Is your own website enough to get recommended by AI?
Your own website is not enough. It was 35.0% of the sources cited about Sydney agents and 41.5% of the sources cited about New York agents. A minority in both cities. The larger share, roughly 65% in Sydney and 58% in New York, came from somewhere you do not own.
That is the finding agents tend to resist most, so here is the full distribution. Each row is the share of the pooled three-engine citation set that fell into that source type, after we reduced every cited URL to its registered domain and classified it.
| Source type | Sydney % | New York % |
|---|---|---|
| Agency's own website | 35.0 | 41.5 |
| Agent directory (RateMyAgent, WhichRealEstateAgent, Top10/Top3, iREC, LocalAgentFinder) | 31.0 | 6.4 |
| Listing portal (Zillow, Realtor.com, Domain, realestate.com.au) | 13.0 | 25.7 |
| Franchise HQ (Compass, McGrath, Belle, LJ Hooker) | 7.2 | 9.2 |
| Other third-party (lenders, services, tech, education) | 5.3 | 7.5 |
| News / trade media | 3.0 | 3.5 |
| Google Maps / search links | 2.8 | 1.4 |
| Community (Reddit, YouTube) | 2.1 | 2.6 |
| Association / government (e.g. NAR) | 0.7 | 2.1 |
| (n cited URLs) | 432 | 424 |
A cited URL is a source the engine surfaced alongside its answer. It is not proof of why a given agent got named. We are reporting what the sources behind the answer are made of, not a ranking mechanism. A great website is necessary. It is also under half the picture in both cities, and the other half is held by third parties who decide what they publish about you. What actually makes AI cite a source is the deeper question underneath this one.
What sources does AI cite to recommend a real estate agent, and how do Sydney and New York differ?
AI cites a stack of third-party sources, and the one at the top of that stack is structurally different in each city. In Sydney, the dominant third-party type is the agent directory: RateMyAgent, WhichRealEstateAgent, the Top10 and Top3 ranking sites, iREC, LocalAgentFinder. They were 31.0% of Sydney citations. In New York, that directory layer barely registers at 6.4%. The New York gatekeeper is the listing portal: Zillow and Realtor.com, at 25.7%.
The two magnitudes are the story. Directories run about 4.8x higher in Sydney than New York (31.0% vs 6.4%). Portals run about 2x higher in New York than Sydney (25.7% vs 13.0%). In New York, portals plus franchise HQ together are around 35% of citations. In Sydney, the agent-directory layer at 31.0% is the single largest source type after the agent's own site.
The raw domain counts make it concrete. In Sydney, the most-cited domains were RateMyAgent (38), realestate.com.au (27), WhichRealEstateAgent (25), Top10RealEstateAgent (16), iREC (14), domain.com.au (13), Top3RealEstateAgents (12), Google Maps (12), LocalAgentFinder (10), and McGrath (9). Five of the top eight are directories. In New York, the top domains were Zillow (52), Realtor.com (38), Compass (22), deniroteam.com (15, an individual team site), FastExpert (15, a directory), and NAR (9). Two portals sit at the top, and the directory presence is one site, not four.
Same question, same three engines. In Sydney the gatekeeper is the directory. In New York it is the portal.
Want to know which gatekeeper controls your market? Run the same probe on your own patch →
This connects to what we found at the brand level. Our bias-versus-grounding map showed that a brand can be remembered by a model yet absent from what it retrieves. This study sits one layer down: not which brands get named, but which domains feed the answer. Read alongside our Sydney agent study, the directory dominance in Sydney is consistent across both datasets.
Does the pattern hold across ChatGPT, Perplexity, and Gemini?
The pattern holds on all three engines independently. Every engine cited more directories in Sydney and more portals in New York. This is the test that tells us the inversion is a market signal and not an artifact of one engine's quirks.
| Engine | Directory — Sydney | Directory — NYC | Portal — Sydney | Portal — NYC |
|---|---|---|---|---|
| Perplexity | 25.8 | 4.5 | 10.6 | 30.6 |
| OpenAI | 29.9 | 4.3 | 6.5 | 25.7 |
| Gemini | 34.5 | 8.2 | 16.6 | 22.7 |
Read across any row. Directory share is higher in Sydney than New York for all three. Portal share is higher in New York than Sydney for all three. The market gatekeeper is consistent regardless of which engine you ask.
The own-site percentage is a different story. It is not engine-stable. Across engines, Sydney own-site ran 34 to 39%. New York ran from 33% on Perplexity to 54% on OpenAI, with Gemini at 43%. So the "your website is a minority" headline is firm in aggregate, but the exact minority depends on which engine surfaced the answer. The directory-versus-portal signal is the durable one; the own-site number moves.
One engine-specific quirk is easy to misread, so flag it now. Google Maps links are small overall, around 1 to 3% of citations, but they spike on OpenAI, where they were about 16% of OpenAI's Sydney citations and barely appeared elsewhere. Those are model-generated navigation links, not retrieved web pages. We bucket them on their own and do not count them as sources AI pulled from the web.
What we did
We mapped every source three AI engines cited when recommending a real estate agent, in Sydney and in New York, then classified each cited domain by type. Three engines: OpenAI, Perplexity, Gemini. We held the engine set fixed across both cities to keep the comparison like-for-like.
Sydney covered 10 suburbs, probed on 2026-05-22. New York covered 10 neighborhoods, probed on 2026-05-21. Same pipeline, matched tier distribution (3 prestige, 2 inner, 3 middle, 2 outer), matched prompt wording, matched volume. Sydney yielded 432 cited URLs. New York yielded 424. Zero URLs were dropped. We reduced every cited URL to its registered domain, then classified it into a source-type taxonomy: portal, agent-directory, agency-first-party, franchise HQ, Google Maps-search, news, association-gov, other third-party, and community. The full domain-to-type mapping is published with the study so anyone can reproduce the classification.
We made one correction, and we will state it plainly, because it is a credibility signal, not a weakness. Our first pass defaulted unmatched domains into "agency first-party," which over-counted own-sites. On review we reclassified the domains that are not an agent's own site: agent-ranking and comparison sites such as RateMyAgent variants and the Top10 and Top3 style sites, listing portals, lenders, trade media, and a conveyancing firm. The corrected own-site shares, 35.0% in Sydney and 41.5% in New York, are lower than the first pass and are what we report here.
- Small, directional sample. Around 430 citations across 10 suburbs or neighborhoods per city, three engines pooled. Aggregate and per-engine claims are defensible. Per-tier (prestige/inner/middle/outer) splits are not, so we do not report them.
- One city per country. Sydney is not all of Australia, and New York is not all of the US. A second US metro would be needed before anyone could claim a "US pattern."
- Citation is not cause. A cited URL is a source surfaced alongside the answer, not proof of why an agent was named.
- Domain classification is judgment. The full taxonomy and per-domain mapping are published, and the own-site bucket was corrected after an initial over-count, as described above.
- Gatekeeper names are local. RateMyAgent and Zillow are examples, not universal. Transfer the category (directory vs portal) across markets, not the brand name.
One more thing the data carries that the headline does not. Some cited agency sites are out-of-market noise: the engines occasionally surface a non-local brokerage. We still count those as own-sites. They reflect retrieval noise, not local authority, and they are part of why the own-site number wobbles between engines.
What this means for real estate agents
Optimizing only your own website leaves the larger share of your AI footprint untouched in both cities. The lever that closes the gap is the third-party surface that dominates your market, and that surface is different in each city.
| Market | Dominant gatekeeper | Where to put the work |
|---|---|---|
| New York | Listing portals (25.7%) plus franchise HQ | Get your Zillow and Realtor.com profiles and franchise-HQ presence right. |
| Sydney | Agent directories (31.0%) | Get your RateMyAgent and WhichRealEstateAgent review profiles and Google Business presence right. |
Same goal, different surface. The first step is the audit that tells you which surface is cited in your market, because guessing wrong sends the budget to the channel that barely registers where you operate. Our free instant check tests the engines for your market; our cross-market local-business study shows the same gatekeeper logic outside real estate.
FAQ
What sources does AI use to recommend a real estate agent?
A mix, and the agent's own website is the largest single piece without being a majority: 35.0% of cited sources in Sydney, 41.5% in New York. The rest is third-party. The biggest third-party type was agent directories in Sydney (31.0%) and listing portals in New York (25.7%). Franchise HQ sites, news and trade media, community posts, and association pages fill out the long tail. Every figure here is from the pooled set of three engines: OpenAI, Perplexity, and Gemini.
Is my own website enough to get recommended by AI?
No. Across the three engines we tested, your own site was 35.0% of cited sources in Sydney and 41.5% in New York, and the per-engine New York range ran from 33% to 54%. A minority either way. The majority of your AI footprint sits on third-party domains you do not control.
Why does AI recommend different sources in Sydney than in New York?
Because the third-party layer that publishes about agents is structurally different in each city. Sydney has a dense agent-directory and review ecosystem: RateMyAgent, WhichRealEstateAgent, the Top10 and Top3 ranking sites, iREC, LocalAgentFinder. Those sites were 31.0% of Sydney citations and only 6.4% of New York citations, about 4.8x higher in Sydney. New York instead runs on listing portals, mainly Zillow and Realtor.com, at 25.7% versus 13.0% in Sydney, about 2x higher. The engines cite what the local web publishes, so the gatekeeper inverts with the market.
Does ChatGPT cite different sources than Perplexity or Gemini?
The directory-versus-portal pattern is consistent across all three. Every engine cited more directories in Sydney and more portals in New York. The own-site percentage is where they diverge: in New York it ran from 33% on Perplexity to 54% on OpenAI, with Gemini at 43%. One engine-specific quirk is that Google Maps links spiked on OpenAI, around 16% of its Sydney citations, and barely appeared elsewhere.
Should real estate agents focus on their website or on third-party profiles?
Both, weighted to your market. Your website earns you under half your AI footprint, so the third-party surface is where the rest is won: portal profiles in New York, directory and review profiles in Sydney. Audit which one is cited in your market before you spend.
Did you test Grok?
No. This study used OpenAI, Perplexity, and Gemini for a like-for-like comparison.
Sources
- Cited Research, Citation Provenance — Sydney vs New York, 2026-05-22. Sydney: 10 suburbs probed 2026-05-22; New York: 10 neighborhoods probed 2026-05-21; three engines (OpenAI, Perplexity, Gemini); 432 cited URLs (Sydney) and 424 (New York) classified by registered domain into a published source-type taxonomy.
- Cited Research, Australian Real Estate Agent AI Study, 21 April 2026. Related grounding-layer dataset for the AU series.
- Cited Research, Bias-vs-Grounding Map of Australian Real Estate, 26 April 2026. Sister study on parametric memory vs grounded retrieval.
- Source-type taxonomy and full per-domain classification published by Cited with this study (domain to source-type mapping), 2026-05-22.
We built Cited because no one was measuring what AI engines actually recommend. Our methodology is public, our data is first-party, and we practise on ourselves before we advise clients.
More about Cited →