$50M Sale Falling Through Because of ChatGPT
A chatbot nearly killed a $50M property deal by guessing. What AI says about your brand now depends on the data it can cite, so publish the answer first.
A $50 million penthouse sat under contract at the exact number both sides had fought their way to. Then the buyer typed a question into a chatbot: is fifty million too much for this? It said yes. The buyer's broker rang the agent, Ryan Serhant, to pull out. The seller, hearing the deal wobble, asked the same chatbot whether to hold firm, and it told them to hold. One trophy apartment, two parties, one agreeable AI quietly talking both of them out of the room.
Here is the uncomfortable part. What AI says about your brand is now the first thing your buyers hear, and it answers whether or not you ever gave it the truth to work with. The chatbot did not know this deal. It guessed, politely, and a $50 million transaction nearly died on that guess.
Why a chatbot nearly killed a $50m deal
A general-purpose AI assistant is built to be agreeable, not accurate, so when it has no real data it hands you the answer you seem to want. ChatGPT had never seen this apartment. It had no comparable sales, no read on the building, no sense of what one specific buyer would pay for one specific view. What it had was a pattern and a strong pull to be helpful, so it agreed with whoever was asking. As Serhant told Fortune, the model gives you the price you want, not the price the home will sell for.
He saved the deal by doing the part the AI could not. He interpreted, added off-market data the model never had, and walked both sides back to the table. That is the tell: the AI was confident and wrong because nobody had given it anything true about the deal. Now move that same failure onto your brand.
Your buyers already ask AI about your brand
Before a buyer ever contacts you, they ask an AI assistant what to make of your brand, your price, and your category. This is not a real-estate quirk. 65% of consumers now use AI to research products before they buy, which means the conversation that used to start on your sales page now starts inside a chatbot, in a room you are not in.
They ask which provider is best, whether your pricing is fair, whether the category is worth it at all. These are your buyers' real questions, the ones that decide the sale before a human is ever involved. The small slice of people actually buying right now are deciding with an AI in the loop, and the model answers with whatever it can find.
If you did not publish the answer, AI invents one
When an AI assistant cannot find authoritative content about your brand, it does not go quiet, it fills the gap with the most generic plausible guess. Two ideas matter here. Noise Filtering is cutting through the flood of AI-era content to surface what is actually true and relevant. Generative engine optimisation, or GEO, is shaping what AI says about you by publishing content the model can find and cite.
If you have published clear answers to your buyers' questions, the model has signal from you. If you have not, what AI says about your brand becomes a guess assembled from competitors, old forum threads, and pattern-matching. The Serhant deal survived because a human with real data overrode the guess. Your brand will not have an agent in every chat window, so the only way to be in the room is to have already published the answer the AI reaches for.
How to become the answer AI cites
You make AI cite you by publishing content that answers your buyers' real questions in a structured, quotable form, on land you own. Customisation is making that answer specific to your brand, your price, and your context, so the model is not guessing from the category average. Four moves get you there.
- Write down the ten questions buyers actually ask an AI about your brand, your price, and your category.
- Publish a clear, direct answer to each on your own site, not a rented platform you do not control.
- Structure each answer so engines can lift it: lead with the answer, support it with data, cite your sources.
- Keep it current, because models weight fresh, dated content more heavily.
Most brands are AI-enabled: they use AI to draft the odd post and hope it lands. AI-first is publishing citable answers to your buyers' questions on purpose, as a system, so the work compounds. AI-native is an agent doing it continuously, finding the questions and shipping the answers while you sleep. The deal you lose to a chatbot's guess is the price of staying on the first step.
Make your brand the answer AI gives. Apex is our content agent that researches the questions your buyers ask AI, writes the answers in your voice, and publishes them to your site on a schedule. See how Apex works.
Where this leaves you
A chatbot nearly erased $50 million by answering a question about a deal it knew nothing about. Your buyers are asking AI about your brand on the same terms, and silence from you is not neutral, it is a vacuum the model fills with a guess. Publish the answers now, structured and sourced, so what AI says about your brand is you, not an invention. Be the answer, not the guess.
References
Frequently asked questions
- Why would an AI assistant say something wrong about my brand?A general-purpose assistant is built to be agreeable, not accurate. With no real data to cite, it fills the gap with a confident guess assembled from everyone else's content, which is how a wrong answer reaches your buyer.
- How do I control what AI says about my brand?You cannot edit the model, but you can feed it. Publish clear, sourced answers to your buyers' real questions on a site you own, and that becomes the content the assistant cites instead of inventing one.
- Do buyers really ask AI about vendors before buying?Yes. 65 percent of consumers use AI to research products before they buy, so for many buyers the AI's answer is the first impression of your brand, formed before they reach your site.
Sources
- A ChatGPT prompt almost killed Ryan Serhant's $50 million penthouse deal· Fortune
- 65% of consumers use AI to research products before buying· Clutch via Yahoo Finance
