Cloned ChatGPT Offers Are Blowing Up Home Deals

Two unrepresented buyers sent the same ChatGPT-written offer, each about 5 percent under asking, to one Hudson Valley listing. The wording matched. The structure matched. The number matched.

Hillary Kolos, the Compass agent on that file, ran the prompt herself. The bot came back about 5 percent light, in the same tidy sections. That is the housing story of 2026. People are not swapping agents for software so much as dropping a chatbot into live deals and watching the deals buckle.

Why Two Offers Arrived in the Same Words

Kolos is used to unrepresented buyers. Those notes, she said, are usually a couple of sentences in an email, not a packet with headings and gentle sales language. Both of these were overly organized. Both landed at almost the same dollar figure.

Megan Brenn-White, founder of Compass’s Upstate Curious team in New York, said the seller could not really accept terms that strange. Unrepresented is legal. The strategy the bot handed them was not strong. When an offer reads that odd on day one, she said, inspection and negotiation tend to go the same way, and sellers stop wanting those bids.

The sellers did take one of them. Routine inspection items then hit, the kind an agent often talks a buyer through. The buyer walked. They lost the $1,000 they had spent on the inspection. The house went back on the market.

Brenn-White said her 23-agent team, which expects about 450 deals this year, noticed chatbot clients about six months before that file. Only in the last two or three months, she said, did it become an industry-size problem. The workaround they trust is ugly and simple. They feed every question they can think of into the same bots before a listing goes live, so they can guess what a buyer will be told.

On October 13, Bay Area realtor Yu-En Tsai wrote that clients now arrive with chatbot answers that take a long time to unwind. More text, he said, has not made the choice easier. In some people it splits trust with the professional in the room. That is the same pattern Kolos hit on paper: two strangers, one model, one weak bid, twice.

72 Percent Would Use AI on a House Deal

Comfort is not the same as handing over the offer. LendingTree commissioned QuestionPro to poll 2,001 U.S. adults ages 18 to 80 from July 2 to 8, 2026. In that survey, 72 percent would use AI for a home task. Another 28 percent said they would not use it for any housing task at all.

WHERE AMERICANS SAY THEY WOULD USE AI

The 37 percent who would let a model run a purchase still sit behind the 41 percent who would not. Looking to 2036, 36 percent expect AI to handle most or nearly all of a sale or purchase. Sixty-four percent expect only some tasks, or almost none.

When the question is who they trust on the offer price, a real estate agent leads at 25 percent and a person’s own research follows at 23 percent. Both beat AI. The top fear is bad or misleading information, at 22 percent, then data security at 16 percent.

Matt Schulz, LendingTree’s chief consumer finance analyst, called the bot a starting point. “AI should be a starting point, not the final word,” he said. “A mortgage or real estate professional can help explain trade-offs, local market realities, loan rules and risks that a chatbot or online tool may miss.”

Veterans United Home Loans, in a separate survey of prospective buyers, found 53 percent would be comfortable purchasing a home with no direct human involvement. Only 25 percent in that poll said they would be very comfortable. The gap between “I would try a bot” and “I would trust it on the number” is where Kolos’s twin emails live.

In the past 12 months, 34 percent of LendingTree’s adults had used an online home search site. AI tools sat at 15 percent, tied with agent websites. Among people who had already used AI on housing, the live jobs were neighborhood research at 38 percent, home search at 35 percent, and value estimates at 33 percent. Writing listing copy was last, at 13 percent.

ChatGPT Talked Both Sides Off a $50 Million Penthouse

Ryan Serhant, founder of the brokerage SERHANT, walked through a New York City penthouse that had finally reached a deal sheet at $50 million flat. The buyer, at the last hour, asked ChatGPT if $50 million was too much. The model said yes. The buyer’s broker called to kill the deal.

Serhant told the seller. The seller asked the inverse question: with the buyer gone because the bot said 50 was too high, was 50 too low? The same system said it was. Both sides had a tidy answer that matched the way they had asked.

They know the history of the internet. They don’t know path forward.

Ryan Serhant, CEO of SERHANT, at Brainstorm Tech

He put the contract back together with off-market context the model could not scrape. Ravi Kantha, a broker at the same firm, said the tools “sort of tell you what you want to hear sometimes.” Prompt the buyer as a cautious shopper and the model hunts for reasons the price is rich. Prompt the seller as someone leaving money on the table and it hunts for upside.

That is the expensive version of Kolos’s 5 percent haircut. At $50 million the people could still reconvene. At a Hudson Valley list price, the cloned bid just looks cheap and weird, and the seller moves on.

Cooper City Closed at $954,800 Without an Agent

Robert Levine, a Cooper City, Florida, father of three, did close. He and his wife started asking ChatGPT about selling on a drive to North Carolina, then let it steer pricing, paint, marketing, the listing day, and a draft contract. A lawyer reviewed the paperwork. He showed the four-bedroom, three-bath house himself.

The listing went live on Tuesday, March 3, 2026. Fifteen showings followed. Five offers arrived inside 72 hours. By Sunday morning the house was under contract at $954,800. Levine said that beat what local agents had told him by $100,000, and that skipping a listing agent saved about 3 percent, or about $28,000.

Ines Hegedus-Garcia, managing partner at Avanti Way East Miami, where the buyer’s agent works, pulled 31 non-waterfront family sales in the area. Levine’s home sold for $358.41 a square foot, above a $339.20 average. She still did not buy the idea that the bot had nailed the price. The lot, the updates, and the condition were stronger than the wider sample, she said, and the photos did not show it. The buyer’s agent thought it was more of a fixer until they walked in.

Five offers, in her view, were the moment to call for highest and best, not to pick a contract and stop. “You did not capitalize on that multiple offer scenario,” she said. Joel Berner, a housing economist, called Cooper City a buyer’s market in February, with inventory up 5.1 percent and median list prices down 14.6 percent from a year earlier. Immediate competition, he said, was rare leverage. Hegedus-Garcia put the cost of skipping representation, after subtracting Levine’s claimed $28,000 commission save, between $75,000 and $225,000.

Levine also kept walking contract questions back through ChatGPT, then to the buyer’s agent, who had to say some of it did not apply in Florida. Dan Weisman, director of innovation strategy at the National Association of Realtors, said chatbots can help with research and are not a substitute for someone with a fiduciary duty. The viral version of this file is a five-day win. The buyer-side file is a house that looked worse online than it was, and a seller who never ran the bid-up.

A $30,000 Roof Turns Into a $70,000 Fight

Pricing is only the first break. Brenn-White’s buyers now run inspection reports through chatbots, then arrive with repair numbers no local crew has quoted.

People are dropping out of deals or asking for wild numbers. Like a roof that we would get a quote from a contractor for let’s say $30,000, AI was telling them would be $70,000.

Megan Brenn-White, founder, Upstate Curious team at Compass

If her team says they have seen that roof done for less, the client answers with the printout. Trust moves to the page that looks official. Kati Spaniak, a realtor in the Chicago suburbs, said a local agent can tell you foundation cracks in that town show up in 99 percent of the houses. The bot says “yes, that is an issue” and lists trades to call.

Shantih Moriarty, a Redfin agent in the Bay Area, has seen models invent cracks that are not in the report and name pipe types that are not in the house. Suzanne Goldklang, a licensed salesperson in New York, put the design problem in one line: a computer can boldface a guess and still be guessing. Region is the other miss. Ask for a Hudson Valley roofer and the model may hand you Albany. Ask for Bay Area labor and it may quote Michigan.

HOW CHATBOTS KEEP BREAKING LIVE DEALS

• Cloned bids: Two strangers using the same prompt both came in about 5 percent under ask, in matching language the seller did not want.
• Opposite advice: On the $50 million penthouse, ChatGPT told the buyer the price was high and told the seller it was low.
• Phantom defects: Moriarty has watched models describe foundation cracks and piping that were not in the house.
• Inflated repairs: Brenn-White’s local roof quote sat near $30,000 while the bot told clients $70,000.
• Wrong-city crews: Agents say the models pick contractors and prices from the wrong metro, then buyers treat those figures as the ask.
• Walkaways: After the cloned Hudson Valley offer, routine inspection items ended the deal and burned $1,000.

Moriarty still sends clients into long condo documents with a bot, because CC&Rs can run 300 pages and nobody reads them in a week. A hundred questions might hide ten that matter and three nobody would have asked. Her rule is trust but verify, with an agent in the loop. Joey Santiago, buying in Southern California, used ChatGPT to sketch a down payment plan, then Grok and ChatGPT to test whether a queen bed or a 30- to 32-inch vanity would fit. He cross-checked before he offered. That is the narrow use that does not blow up a contract.

Buyers have a separate complaint aimed the other way. Agents now use the same class of tools to polish listing photos and spin walk-through videos, a habit shoppers call housefishing. New York’s Department of State has already warned that AI listing images can cross into deceptive ads. The ouroboros is real enough: people punish agents for synthetic photos, then paste the inspection PDF into a chat window and treat the reply as local fact.

The 2024 Commission Rules Opened This Door

Generic bots did not invent the unrepresented buyer. The 2024 NAR settlement changed the paperwork around that buyer, and the chat window filled the gap.

FROM COMMISSION RULES TO CHATBOT CLOSINGS

1. March 15, 2024: NAR announces a settlement that includes $418 million over about four years and new MLS rules, while denying wrongdoing.
2. August 17, 2024: Offers of buyer-broker pay disappear from MLS listings, and MLS participants working with a buyer need a written deal that states compensation before a tour.
3. November 26, 2024: A court grants final approval of the settlement.
4. May 2025: Homa, a licensed brokerage built around AI tools and in-house brokers, launches in Florida.
5. March 2026: Levine lists in Cooper City and Serhant’s penthouse nearly dies in the same chatbot week.
6. June 2026: Homa opens in Texas.
7. July 2 to 8, 2026: LendingTree fields the 2,001-adult survey that finds 72 percent open to one AI housing task and 41 percent unwilling to hand over the purchase.

NAR’s consumer explainer is blunt about the tour rule. Buyers who work with an MLS participant now sign a written agreement before they tour a home, and that paper has to state how the broker is paid. Compensation can still come from the seller, the buyer, or both. It just cannot sit as a field on the MLS.

Homa’s own explainer uses a $500,000 home to show the sting: a 2.5 percent buyer-broker fee is $12,500 on top of down payment and closing costs, unless the buyer contacts the listing agent directly and stays unrepresented. That is the hole ChatGPT is being asked to fill for free, and the hole a flat-fee brokerage is trying to productize.

Homa Rebates Up to 2 Percent and Still Uses Brokers

Arman Javaherian, a former Zillow senior director, co-founded Homa with CTO Federico Campeotto after watching buyers do the search work and still fund a full buyer-side commission. The company is a licensed brokerage. It does not send a naked chatbot to the negotiating table.

On paper, Homa takes a 1 percent fee and a $4,995 minimum, then credits the rest of whatever buyer-side compensation is in the deal. Its about page promises up to 2 percent of the purchase price back at closing, subject to the actual offer, the fee, lender caps, and local closing rules. In-house agents are paid per file, not as a share of a higher price. Showing specialists open doors. Licensed brokers still write and negotiate the offer.

Javaherian told interviewers he surveyed 1,000 people who had closed in the prior three years. Sixty-one percent said they had held back a question they wanted to ask a realtor, worried it would sound annoying or create extra work. That is the Instacart instinct: stay in the app, skip the small talk. He also said more than 100 homes had closed or come under contract on Homa. The site lists California, Florida, and Texas.

That model is the opposite of Levine’s ChatGPT tab and of Kolos’s twin emails. A purpose-built broker still has a license and a person who can be sued. A public chatbot will draft the offer, invent the pipe, and quote the wrong roof, then format it so it looks like a memo. Brenn-White’s team now spends listing prep on guessing those memos in advance. Serhant spent a day and a half talking two rich clients out of answers they had asked for.

The replacement story was always the wrong plot. The bot is already in the deal. It writes both sides a confident page, and someone still has to get them back in the same room.

Disclaimer: This article is news reporting and analysis of how buyers, sellers, and licensed brokers are using consumer chatbots and AI brokerages. It is informational only and is not a real estate, mortgage, legal, or investment recommendation. It does not tell you whether to hire an agent, sign a buyer-broker agreement, skip representation, or rely on a chatbot for pricing, inspections, or contracts. Talk with a licensed real estate broker, a real estate attorney, and a mortgage lender in your state before you act on any of this. Figures, product terms, and survey results reflect the named sources as of their publication dates and can change with new rules, new models, and new local prices.