How AI Search Engines Choose Who to Cite, and What It Means for Real Estate
Four platforms now decide what most buyers and sellers learn before they ever click a link: ChatGPT, Google Gemini including AI Overviews and AI Mode, Perplexity, and Claude. Most agents assume these work roughly the same way, and that showing up in one means showing up in all four. The measurement says otherwise, and in real estate the gap between those two beliefs is costing people listings.
The short answer
The four AI answer engines are four separate channels with four different source preferences. Only about 11% of the domains cited by ChatGPT are also cited by Perplexity. They pull from different indexes, they select sources differently, and a single strategy will not win all four. For real estate specifically, the research points somewhere surprising: the portals are not the citation surface. Your own website and your Google Business Profile are.
These are four channels, not one
The overlap numbers are the most useful thing to know about this category, because they invalidate the way most people talk about it.
Roughly 11% of domains cited by ChatGPT are also cited by Perplexity. Claude and ChatGPT overlap on about 13% of cited domains and about 4% of cited URLs. Even inside Google’s own house, AI Mode and AI Overviews cite the same URLs only 13.7% of the time across 540,000 query pairs measured by Ahrefs, despite reaching the same conclusion 86% of the time. Two surfaces, one company, one infrastructure, different citation sets.
This is why a single AI visibility score is close to useless. Averaging four channels that agree 11% of the time produces a number that hides the only actionable information you had. Ask for per engine reporting or do not bother measuring.
One more piece of context worth holding onto. Cloudflare Radar data for May 2026 shows all four AI chatbots combined sending roughly 0.29% of search referrals, against Google’s 87.6%. AI search today is an influence channel far more than a traffic channel. It shapes the shortlist before anyone visits a site. Judge it on whether you are named, not on sessions.
What each engine is actually used for
The engines diverge most in what people bring to them, which drives what kind of content gets pulled.
| Platform | Dominant mode | What the user wants |
|---|---|---|
| ChatGPT | Advisory | A recommendation they can act on |
| Gemini and AI Overviews | Ambient answer | The answer without leaving Google |
| Perplexity | Verification | A cited, checkable synthesis |
| Claude | Production | A finished work artifact |
OpenAI and the National Bureau of Economic Research analyzed 1.5 million consumer conversations through July 2025, the largest study of real consumer AI usage published so far. About 49% of messages are asking rather than doing. People treat ChatGPT as an advisor. That is exactly the posture of someone deciding which agent to call, and it is why ChatGPT matters disproportionately in a high consideration purchase.
Google’s signal is different but points the same direction. Queries in AI Mode run roughly three times longer than traditional Google searches, and nearly one in six use voice or images rather than typed text. These are not keyword lookups. They are the multi-condition questions a buyer used to have to break into four separate searches.
How each engine finds and chooses sources
All four use retrieval augmented generation. A search layer supplies documents and the model summarizes only what it received. The differences are whose index, how the question gets expanded, and what gets selected.
ChatGPT
ChatGPT runs three stacked layers: training data with no citations shown, a search index historically anchored to Bing, and live page fetching during a conversation. It breaks a question into sub questions, retrieves candidates for each, fetches pages, and selects citations by reading the page rather than by scoring domain authority.
The practical trap is rendering. ChatGPT’s fetchers do not reliably execute JavaScript. Reported parsing success is around 94% for static HTML with schema markup against roughly 23% for JavaScript rendered content. A site that looks perfect in a browser can return an empty page to the crawler. This is the single most common technical reason an agent is invisible.
It runs three crawlers controlled separately in robots.txt: GPTBot for training, OAI-SearchBot for the search index behind citations, and ChatGPT-User for on demand fetches.
Google Gemini, AI Overviews, and AI Mode
Google confirms that both AI Overviews and AI Mode use query fan out, issuing multiple related searches across subtopics at once and then synthesizing. The consequence is the number every agent should know: only about 14% of URLs cited in AI Mode also appear in the organic top 10. Moz’s 40,000 keyword study put 88% of AI Mode citations as coming from outside the top 10.
Volatility is extreme. Repeat the same query three times and URL consistency runs about 9.2%. Pages that rank for the hidden fan out sub questions are reported to be 161% more likely to be cited than pages ranking only for the visible query. Rank tracking measures a shrinking fraction of this. The unit of optimization is the question space around a topic, not a keyword.
Perplexity
Perplexity is the only one of the four that controls the whole stack. It brought search in house in April 2025 and runs a proprietary index with hybrid lexical and semantic retrieval, LLM reranking, and cross encoder rerankers on top. The model is instructed to use nothing beyond retrieved passages, cite every sentence, and decline when sources are thin, which is why its citation density is the highest of the four at 6.4 unique domains per answer.
The consequence is that Perplexity visibility is not downstream of Google or Bing rankings. It is its own ranking problem, with its own crawlers, PerplexityBot and Perplexity-User.
Claude
Claude runs no proprietary index. The public evidence points to Brave Search: Brave appears on Anthropic’s subprocessor list, and independent testing found 86.7% overlap between Claude’s citations and Brave’s top non sponsored organic results. Brave runs an independent index of roughly 40 billion pages, separate from Bing since 2023.
Anthropic’s Citations API generates attributions at the sentence and passage level rather than the document level. Claude is selecting passages, which is why cleanly chunked, self contained sections outperform a comprehensive but monolithic page. Its crawlers, formally documented in February 2026, are ClaudeBot for training, Claude-SearchBot for search quality, and Claude-User for on demand fetches. Claude-SearchBot is the visibility critical one.
Which websites each engine cites
Rank order of top cited sources through Q1 2026, synthesized across five independent studies including Peec AI’s 30 million source analysis, Semrush’s 325,000 prompt study, and SE Ranking’s 129,000 domain analysis.
| Platform | Top five cited sources, in order |
|---|---|
| ChatGPT | Wikipedia, Reddit, LinkedIn, Forbes, Medium |
| Perplexity | Reddit, LinkedIn, NIH, Microsoft, G2 |
| Google AI Mode | LinkedIn, YouTube, Reddit, Google properties, Wikipedia |
| Google AI Overviews | YouTube, Reddit, Forbes, LinkedIn, Wikipedia |
| Gemini app | Reddit, YouTube, Wikipedia, Medium, Forbes |
Read that table and the obvious conclusion is to go win Reddit. For three of the four engines that is defensible advice. For the fourth it is completely wrong.
Why Claude is the exception
OtterlyAI measured 379,321 Claude citations across 16,406 domains in June 2026. The distribution does not look like the table above at all.
| Source category | Share of citations |
|---|---|
| Brand, company, and product sites | 64.0% |
| News and media | 14.9% |
| Blogs and personal sites | 5.4% |
| Education | 2.7% |
| Community and forum | 1.3% |
| Social media | 0.9% |
Reddit was cited zero times. Not rarely. Zero, across 379,321 citations. First party brand content is 64% of what Claude cites, more than four times the next category, and no single brand dominates it. The top 10 domains account for just 9.5% of all citations, and nearly a third of cited domains appeared exactly once.
Several widely circulated 2026 summaries claim Reddit is the number one source across every major engine including Claude. Direct measurement says otherwise. The likely explanation is that consolidated indexes average across the engines that were actually measured and extrapolate to the one that was not. When a synthesis and a direct measurement disagree, trust the direct measurement. It is a good illustration of why per engine tracking beats an aggregate score.
What this means for real estate specifically
Three findings change the picture materially for anyone selling or listing property.
Luxury real estate has the lowest AI Overview trigger rate of any major US industry at approximately 0.14%, against 13% for health, 4.2% for finance, and 2.1% for retail, according to the 2026 Luxury Real Estate AI Discovery Report from Haute Residence and 5WPR, published April 2026. The same report found 82% of agents now use AI daily. Google’s AI layer barely fires on these queries, which makes ChatGPT, Perplexity, and Claude disproportionately important in this vertical relative to any other.
The portals are not being cited. Omni Eclipse analyzed 334 ChatGPT answers across 253 buying queries in 197 markets across the US, UK, and UAE during April and May 2026. Across 1,567 citations the study recorded zero links to Zillow, zero to Realtor.com, zero to Redfin, zero to Trulia, and zero to Homes.com. The portals appear inside the prose as descriptive evidence about an agent, as in a 5.0 rating on Zillow, but never as a source backing the recommendation. What did get cited was Google Business Profile and Maps at 57.6% and the agent’s own website at 40.5%. That is 98% of every citation recorded.
The field is nearly empty. Only about 8.4% of practicing US agents appear in any AI generated answer to high intent queries in their own market, while the top 1% capture 47% of all citation share, according to FlyDragon’s 2026 State of AI Search in Real Estate. In the same firm’s Q1 2026 survey of 4,180 buyers across 38 US metros, 67% reported using an AI tool as their primary research method before contacting an agent, up from 17% eighteen months earlier.
Put those together and the strategic read is unusually clean. Two thirds of buyers are running an AI search before they call anyone, nine out of ten agents are invisible in that search, and the portals you may be paying for are not the thing being cited. Your own website and your Google Business Profile are the citation surface. That aligns almost exactly with Claude’s brand dominant pattern and with ChatGPT’s observed behavior in this vertical, and it means portal spend and AI visibility are close to unrelated investments.
What to do about it
- Check crawler access first. The most common cause of invisibility is a robots.txt written before these bots existed. At minimum allow OAI-SearchBot, ChatGPT-User, PerplexityBot, Perplexity-User, Claude-SearchBot, and Claude-User. Treat GPTBot and ClaudeBot, the training crawlers, as a separate policy decision, because blocking those does not remove you from search results.
- Serve real HTML. If your site renders its content in JavaScript, assume the engines are seeing an empty page. Static markup with schema is the difference between roughly 94% and 23% parsing success.
- Fix your Google Business Profile before anything else. It is 57.6% of citations in the only study that has measured this vertical directly.
- Write pages that can be lifted intact. Sentence level citation on Claude, passage extraction on Google, and chunk level selection on ChatGPT all reward the same structure: clean headings, self contained sections, explicit definitions, real evidence.
- Report per engine and audit quarterly. ChatGPT’s Reddit citation share collapsed from about 60% of prompt responses to roughly 10% inside two weeks in September 2025, then partially recovered. This category moves in weeks, not years.
Frequently Asked Questions
Which AI search engine matters most for real estate?
ChatGPT, Perplexity, and Claude matter more in real estate than in almost any other industry, because Google’s AI Overviews barely trigger on these queries. Luxury real estate has the lowest AI Overview trigger rate of any major US industry at roughly 0.14%, compared with 13% for health and 4.2% for finance, per the 2026 Luxury Real Estate AI Discovery Report from Haute Residence and 5WPR. If you are choosing one engine to start with, start with ChatGPT, because its usage pattern is advisory and that is the posture of a buyer choosing an agent.
Does ChatGPT cite Zillow or Realtor.com?
No. Omni Eclipse analyzed 1,567 citations across 334 ChatGPT answers to real estate buying queries in 197 markets during April and May 2026 and recorded zero links to Zillow, Realtor.com, Redfin, Trulia, or Homes.com. Portals appear inside the text as descriptive evidence about an agent, such as a rating, but never as the cited source. Google Business Profile and Maps accounted for 57.6% of citations and the agent’s own website for 40.5%.
How many real estate agents show up in AI search results?
About 8.4% of practicing US agents appear in any AI generated answer to high intent queries in their own market, and the top 1% capture 47% of all citation share, according to FlyDragon’s 2026 State of AI Search in Real Estate. The field is close to empty, which is the opportunity. Roughly nine out of ten agents are not present in the answer at all.
Is one AI visibility score enough to measure this?
No. Only about 11% of domains cited by ChatGPT are also cited by Perplexity, and Google’s own AI Mode and AI Overviews cite the same URLs just 13.7% of the time. A single composite score averages away the only information you could have acted on. Measure and report per engine.
Do buyers actually use AI before contacting an agent?
Yes, and the shift has been fast. In a Q1 2026 survey of 4,180 buyers across 38 US metros, FlyDragon found 67% used an AI tool as their primary research method before contacting an agent, up from 17% eighteen months earlier. The initial shortlist is increasingly formed before any agent is contacted.
Should I block AI crawlers from my website?
Not the search crawlers. Blocking OAI-SearchBot, PerplexityBot, or Claude-SearchBot removes you from the results these tools generate. The training crawlers, GPTBot and ClaudeBot, are a separate decision, and blocking those does not remove you from search results. Note also that Brave, which appears to power Claude’s search, follows a reciprocal rule: if you block Googlebot, Brave generally will not crawl you either.
The takeaway
Treat these as four channels, because that is what the measurement says they are. In real estate the picture is unusually favorable: the AI layer on Google barely fires, the portals are not being cited, nine out of ten agents are absent, and the two things that do get cited are the two things you already control. That is a narrow window, and it closes as the field fills in.
Talk with us about where you currently appear, or see how this has played out for other teams.
