Every time our scanner asks an AI engine a home seller's question, the engine reports which pages it read. We logged those sources for months. Here is what 15,784 of them say about who gets quoted, and why.
Our scanner asks ChatGPT, Claude and Gemini the questions home sellers actually ask: who should list my home, who sells fastest, who handles an as-is sale, and so on. It runs 12 questions per agent scan and 50 per city study, across US markets.
The engines report their sources. Through mid-July 2026 we logged 4,172 answers that carried source citations: 15,784 citation events across 1,475 distinct websites. One cleanup note: Gemini returns most of its sources as Google redirect links that hide the underlying site, so those were excluded. What remains is attributable, countable, and repeatable.
72% of all citation events went to the big portals and agent directories. Realtor.com was cited 4,081 times, Zillow 3,757, then Homes.com, FastExpert, HomeLight and Redfin. No surprise: these sites hold structured data on millions of agents, and the engines lean on them for anything generic.
If the story ended there, the advice would be depressing: be a portal or be invisible. It does not end there.
Of the 1,475 distinct websites the engines quoted, 936 appeared exactly once in our entire dataset. That is 63%.
We looked at what those single-citation sites are. Agent websites. Team pages. Local market blogs. Small niche directories. Pages nobody would call an authority, quoted because they were the single best answer to one specific question in one market.
A page about flat-fee listing options in one city. A neighborhood guide written by the agent who actually farms it. One page, matched to one question, read into one answer.
The engines are not loyal to big brands. They cite whatever page answers the question best, and most questions are specific. You do not have to outrank Zillow at 'best agent in Dallas'. You have to publish the one page that answers the question the portals answer generically: the as-is sale, the divorce timeline, the 55+ downsize, the flat-fee option, in your specific market.
In our scans, questions about reviews and 'best overall' come back full of names. Questions about pricing models and special situations often come back empty. Those empty answers are open seats, and the first readable page tends to take them.
Dataset: 4,172 cited answers collected by the DoIAppear scanner through July 16, 2026, US markets, seller-intent questions, engines as listed. We count citation events per domain, not page views or traffic. The dataset covers the three engines named above and nothing else. Numbers will move as we keep scanning; we update this page when they move materially.
PropCite city leaderboards are built from the same scan data: they record which agents the engines name, per city, per month. No paid placement. We measure what AI actually says.
PropCite measures who AI names. To see your own results — and a step-by-step on what to fix — run the free AI-visibility check, right here on PropCite.
Run my free check →