AI Is Sending Their Buyers to Someone Else’s Website
Ask Microsoft Copilot about a $3 million branded residence in Fort Lauderdale. You get a confident, detailed answer. Pricing. HOA fees. Rental rules. Five sources.
None of the five is the developer’s official website.
All five sit on domains built from the project’s name. Four of the five citations land on sites whose own footers say, in writing, that they are not the developer’s official website. The fifth borrows the developer’s domain name and bolts “condos” onto it.
Another lookalike footer goes further: “This website cannot be relied upon as correctly stating the representation of the developer.” Asked about a branded tower in Brickell, Copilot cited that site first.
Copilot did not show the footer. Copilot showed the link.
Copilot reads Bing. On Bing, a second lookalike outranks the developer for that building’s own name.
What we tested
From October 13 to October 15, 2026, we asked six AI answer engines the same questions about two South Florida branded residences: ChatGPT, Gemini, Google AI Mode, Microsoft Copilot, Perplexity and Claude. We also checked how three search indexes rank each property: Google, Bing and Brave.
Property A is a pre-construction, hotel-anchored project in Fort Lauderdale, with entry pricing reported around $3 million. Property B is a completed tower in Brickell, selling from about $600,000 to nearly $3 million. Both carry a global hospitality brand. We are not naming either. The problem is not theirs alone.
Nothing about it is local. Any branded residence with a name worth registering as a domain is exposed, in Miami, Manhattan, Austin or anywhere else a hotel brand puts its name on a condo tower.
One run per question per engine. We saved every answer and its sources. Copies available on request.
Three ways to lose came out of it. Your buyers. Your facts. Your spot on the shortlist.
The lookalikes
Property A drew at least six lookalike domains across our captures. Property B drew at least five. Same brand name, different combinations. Residences. Condos. The city. A .net. A city domain extension.
We opened four of them. Every one had a lead form waiting for the buyer and a disclaimer in the footer. Two name Miami brokerages. One names a realty group. One names nobody.
When a broker registers a buyer with the developer’s sales team and the deal closes, the broker collects the buy-side commission. The developer pays it. Including on a buyer who started by asking an AI engine about the developer’s own building.
Google put both developers first. The AI answers cited lookalikes anyway.
What the engines did
Asked about Property A by name:
- Copilot cited five sources. Five lookalikes. Not one from the developer’s official site.
- ChatGPT cited fifteen links. Four were the developer’s. Six came from five different lookalike domains. It found the developer’s website four times, cited a lookalike for the entry price, and used a lookalike as its only source for the sales fact sheet.
- Google AI Mode cited one developer page and one lookalike.
- Perplexity cited the developer and the trade press. No lookalikes.
- Claude cited trade press, a brokerage and listing sites. No lookalikes. No developer either.
- Gemini cited the developer.
Asked about Property B by name:
- Google AI Mode cited ten sources. A lookalike first. The developer dead last.
- Copilot cited two sources. The lookalike first. It also showed two map listings at the same address. The first carries the brand name, calls the building an apartment complex, and links to the lookalike. The developer’s listing came second.
- ChatGPT cited two developer pages and two lookalikes.
- Perplexity, Gemini and Claude cited no lookalikes.
Three of six engines put lookalikes in their answers for both properties. For Property A, Copilot cited nothing else.
Indexed on Google is not enough
The AI engines do not all read the same index. Microsoft says in its own documentation that Copilot grounds its web answers in Bing.
On Bing, a lookalike ranks first for Property B’s own name. The developer is second and third. Bing’s AI summary cites the lookalike before the developer.
On Brave, another independent index, Property A’s official site ranks first. A lookalike ranks third. Another ranks eighth. The first one sits above the lead developer’s own project page at fifth.
Ranking first does not keep the lookalikes out. ChatGPT found Property A’s official site four times and cited lookalikes anyway.
Brave also decided Property A was in Brazil. One word in the property’s name matches a Brazilian state. Brave localized the search there and put Brazil’s +55 country code in front of the sales gallery’s Florida phone number. Your sales gallery. Now in South America.
A developer who checks Google, sees the official site on top, and goes back to lunch has checked one index out of three. The developer cannot assume the buyer is on Google.
The math
Miami developers have typically paid buy-side brokers 5% of the contract price on pre-construction sales. During the 2025 slowdown, some raised it to 7%, according to a May 25, 2025 report by Miami condo market analyst Peter Zalewski.
We don’t know Property A’s actual terms. Apply the published range to a $3 million unit and the number is $150,000 to $210,000.
Per closed sale.
We measured citations, not closings. The commission lands when a buyer who could have come direct gets registered by someone else instead. How often that happens is a number the developer has to go find. Checking Google alone won’t show it. You can’t plug a leak you never look for.
The second bill: your facts
Copilot’s Property A answer listed resort units from about $2 million, private residences at $4.4 million to $5 million, HOA dues of about $2.00 per square foot, and a six-month rental minimum. Every figure was cited to a lookalike. Not one to the developer.
Across the six engines, the basic facts did not agree. These are not opinions. They are basic project facts.
| Fact | What the engines said |
|---|---|
| Property A delivery | 2029 or 2030 |
| Property A hotel keys | 197 or 200 |
| Property A lowest starting price quoted | About $2 million or about $3 million |
| Property B unit count | 420 or 498 |
The AI has already handed the buyer its numbers. The developer doesn’t control which ones.
The third bill: the shortlist
We also asked each engine for its recommended new luxury condo developments in each city.
Property A made three of six lists. Gemini left it off and showed a map card for the hotel the project replaces.
Property B made zero of six. It is completed and selling, and some engines may read “new” as pre-construction only. Six engines answered a buyer’s question about Miami luxury condos. Property B was in none of the answers.
An engine can describe a building correctly when asked by name and never put it on a shortlist. Recognized is not recommended.
The AI warned the buyer. No developer was in the answer.
Asked about Property B by name, Claude never cited a lookalike. It never cited the developer either. It built its answer from trade press, listing sites and a press release from 2024. The developer’s live website, ranked first on Google, didn’t make the cut.
Asked separately to recommend new luxury condo developments in Miami, Claude listed eight projects, cited brokers and marketing sites, and closed with a warning to the buyer: “Sources are broker/marketing sites with sales interests.”
The AI warned the buyer that its sources had sales interests. Not one of those sources was a developer.
What developers should do
- Ask the engines. Put your project’s name into ChatGPT, Gemini, Google AI Mode, Copilot, Perplexity and Claude. Write down every domain each one cites. That list is your real competitive set.
- Check all three indexes. Google, Bing and Brave. Search your own building’s name. If a lookalike outranks you anywhere, find out which AI engines are reading that index.
- Secure the obvious domains before launch. Brand plus residences. Brand plus city. The .net. The local domain extension. Get them before someone else does.
- Find out who owns the second map listing. If a listing at your address carries your brand and links to someone else’s website, find out who controls it and whether it meets the platform’s rules. AI engines pull map data into their answers.
- Publish the facts in plain text. Pricing ranges, HOA, rental policy, delivery date, unit count. On the developer’s own site, dated, where an engine can read and cite it. ChatGPT found one developer’s site four times and still used a lookalike as its only source for the sales fact sheet. Put the numbers where the engine can find them.
- Match the footers to your registration log. Three of the four lookalikes we opened name the brokerage or realty group behind them. Check every name against the project’s broker registrations. Then add one question to the buyer intake sheet: which website did you use first?
- Bring in the brand. The hotel company that licensed its name has a stake in who speaks for it. Ask its counsel about third-party domains built on that name.
The footer is still there
Every lookalike we checked says, in writing, that it is not the developer.
Not one AI answer we captured passed that along.
Find out who AI is citing for your building. The AGR AI Visibility Report checks ChatGPT, Gemini and Google AI Mode in real time and shows whether the links in their answers lead to your official website, a broker, or someone else. Run your property’s AI Visibility Report.
