One of our real estate clients, a brokerage in the Caribbean, gets a steady stream of buyers from a channel most agents have never checked: ChatGPT. In one recent 15 day window, ChatGPT sent 77 visitors to the site. AI assistants fetched its pages about 1,400 times in the same stretch to answer buyer questions in live conversations. This is not a prediction about the future. It is in the logs.
The numbers, measured two ways
We verified this twice because we did not believe it the first time. Google Analytics counted 64 sessions referred from chatgpt.com out of 4,109 total sessions in two weeks. That is 15.6 AI-referred sessions per 1,000, or about 1 in 64 visits starting inside an AI chat. Our server logs, filtered to OpenAI’s published bot IP ranges, agreed within about 15 percent. Two independent measurement systems, same story.
A detail worth knowing: when we audited those logs, roughly 3 of every 4 visits claiming to be an OpenAI bot were fake scanners spoofing the name. If you look at raw user agents you will badly overcount. Verify by IP before you trust any bot number.
Also in the logs: Anthropic’s crawler (the company behind Claude) hit the site 2,461 times in the same window, more than OpenAI did. And Perplexity referred exactly zero visitors across our entire client portfolio. The AI referral game today is ChatGPT first, Claude building an index, everyone else a rounding error.
Why this site, out of thousands of brokerages?
One page per buyer intent
The site has a separate page for each property type a buyer might ask about: condos, beachfront homes, land, commercial. When someone asks an AI “where can I find a beachfront condo in this area,” there is a page whose whole job is answering exactly that. The AI does not have to infer. It quotes.
Plain facts, stated plainly
Prices, locations, and what is included, written in ordinary sentences. No “luxury redefined” copywriting. Language models recommend what they can quote with confidence, and they cannot quote a vibe.
A blog that answers real buyer questions
The most-fetched pages in the AI logs are blog posts answering practical questions about buying in the region: residency, financing, fees. Boring to write, precious to an AI assembling an answer for a real buyer.
Nothing blocking the readers
Fast WordPress site, clean structure, no login walls, no crawler blocks. AI cannot recommend what it cannot read. We wrote about why the platform underneath matters in WordPress site or AI-built website.
Why this matters beyond real estate
The same pattern shows up across our portfolio: two tiny home communities we market get AI-referred leads, a fitness studio gets them, a cottage rental community gets them from three different AI assistants. The sites that win share the same trait: specific pages with plain facts that answer one question each.
One warning: do not let a new channel become a crutch. AI referral traffic can change with a model update, the same way rankings change with an algorithm update. The answer is the same as it has always been: never depend on one channel, and own the asset that all channels point at.
Want a second set of eyes on your ad account or your lead flow? We will look at what you have and tell you what we would do. No pitch deck, no pressure.
