Why you rank on Google and disappear in ChatGPT

An agent showed me her site last month and she was, understandably, proud of it. Page one for her neighborhood. Clean design. Years of blog posts. Then we opened ChatGPT, asked it to recommend a listing agent in her area, and she was not in the answer. Neither was anyone else she competes with. The model named three people, and two of them are not even the strongest agents in that market.

That gap is not a ranking problem. Ranking and citation are two separate outcomes produced by two separate mechanics, and the second one has different rules than the first.

A search engine indexes you. A model cites you.

Google's job is to hand you a list of doors and let you decide which to open. Position matters because position is the whole product. An assistant has no list. It reads, decides, and returns one composed answer, usually with a handful of sources attached. There is no page two to fall back to. You are either in the answer or you are not part of the conversation.

So the question shifts. Not "how do I rank for luxury condos in River North" but "if a model has to name three people who know River North condos, what would make me an obviously safe one to name?" That is a question about evidence, not keywords.

Models are, at bottom, pattern machines that hate being wrong in public. They favor sources where a claim is stated plainly, attributed to a named person, and corroborated somewhere else. If your expertise only exists as a vibe across a beautifully designed homepage, there is nothing for a model to hold onto.

The three things that actually decide it

When I audit a site for this, I look at three things before anything else. They are not glamorous and none of them require a redesign.

1. Can a machine tell who you are? Your name, your market, your license, your brokerage, and your specialty need to appear as text in a consistent form across your site. Not inside a hero image. Not implied. If your specialty is written one way on your homepage, another way on your bio, and a third way on LinkedIn, you have given the model three weak signals instead of one strong one.
2. Do you answer questions, or do you market? Nearly every agent site is written in brochure voice. Brochure voice is unciteable. An assistant cannot quote "unparalleled service and local expertise" to a buyer asking about assessments in a specific building. It can quote a page that says what the assessment situation in that building has actually looked like over the last three years.
3. Is any of it verifiable elsewhere? A single site making claims about itself is a weak source. The same facts appearing on your brokerage profile, a local publication, a podcast description, a professional directory, and your own site form a pattern. Patterns are what get repeated back.

Notice that none of those are tricks. This is not a loophole business. Every workaround I have seen sold in this space is either something the models already discount or something that will embarrass the person using it in about six months.

Do the ten-minute version before you do anything else

Before you spend a dollar, get a baseline. Open ChatGPT, Perplexity, and Google's AI results, and run the questions your clients actually ask, in the words they actually use:

• Who is the best listing agent in [your specific neighborhood]?
• I am relocating to [your city] with two kids. Which agent should I call and why?
• What should I know before buying a condo in [a building you know cold]?
• Who specializes in [your actual specialty] in [your market]?

Write down every name and every source the answers give you. Run the same questions in a fresh session a week later, because these answers move. What you now have is a scoreboard, and most people have never looked at it. Even seeing that the models are confidently recommending someone who left the business two years ago tells you how much of this is still unclaimed.

What to publish instead of more blog posts

The standard advice is to post more content. That advice produced the ocean of interchangeable market-update posts nobody reads, and now it is producing the same ocean written faster by a model. More of that will not help you. It may hurt you, because a site full of generic text is a site with no distinguishing signal.

What works is narrower and harder to fake: the things you know that are not written down anywhere else. The building where financing gets complicated and why. What actually happens at the inspection stage on prewar stock in your area. Which streets flood. What a realistic timeline looks like for a specific kind of seller in a specific season. Write those as direct answers, in your own words, with your name on them.

Ten pages like that outperform a hundred market updates, because they are the only ten pages a model cannot assemble from everyone else's content.

Why this is worth your attention now

The reason to move on this is not that AI search is enormous today. In most markets it is still a modest slice of how buyers find people. The reason is that the slice is growing while it is uncontested, and the position you build in it compounds. Sources that get cited get cited more, the same way early domain authority worked, except this time almost nobody in real estate is playing.

That will not stay true. It is true right now. That is the whole opportunity, and it is a smaller job than most people assume: be findable, be specific, be corroborated, and be patient enough to let it accumulate.