Is AI Replacing Realtors?

When Joey Santiago started to think about buying a house in the competitive Southern California market several years back, he went to ChatGPT for help. He wanted to become financially stable enough to afford a down payment, and though he says the advice he received was pretty generic, like recommendations to "save money" and "use a high yield savings account," he admits it was nice to have a framework.

That’s why earlier this year, when Santiago had finally scraped together what he says was "a fat down payment" thanks to ChatGPT’s advice, he went back to the artificial intelligence well for help envisioning how he could make potential properties his own. By taking measurements and providing pictures of a kitchen, guest room, or primary bath, he says, he could ask Grok and ChatGPT questions like: Could a queen bed fit in this space? What kinds of floor molding could I use in this room? What kind of 30- to 32-inch midcentury-modern vanity would work in this bathroom?

The results weren’t always foolproof, he says, and he always made sure to cross-check the chatbots’ responses before he made any offers or placed furniture orders. But he says using AI helped him figure out what he could do with a space, even if it was owner occupied, full of furniture, and had horrible carpet and popcorn ceilings.

Santiago’s not alone in believing this. More and more people are turning to AI for help navigating the home buying and selling process, leaning on chatbots for answers to questions and guidance in gaining a competitive and financial edge. Meanwhile, many homeseekers are calling out realtors’ use of AI to tweak listing photos or generate flashy videos, calling the practice deceptive "housefishing." (That’s just one way realtors are using AI, as Dwell has previously reported.) It’s like an ouroboros, and all along the way, AI’s rise is majorly impacting the housing market, with data centers causing market price fluctuations and contributing to the ecological issues that could change where humans can comfortably live in decades to come.

A recent LendingTree study said that 72 percent of all Americans are comfortable using AI for at least one task related to buying or selling a home, like looking for the best realtor in an area or asking for mortgage tips. (A Realtor.com survey from last year put that number closer to 85 percent.) Meanwhile, a study by Veterans United Home Loans found that 53 percent of prospective homebuyers would be comfortable buying a home without any direct human involvement, and, according to the LendingTree study, 37 percent of respondents would let AI manage their entire home purchase with minimal human input. Younger buyers, like those in Gen Z, are particularly prone to relying on AI, with about a third saying they use it to do much of their research about the process. The LendingTree study also found that 36 percent of respondents expect AI to be able to handle most or all of a home sale or purchase by 2036—a fact that’s increasingly concerning to realtors, many of whom are already dealing with a scourge of increasingly fraught transactions thanks to people asking AI for guidance.

Earlier this year, New York City realtor and reality TV personality Ryan Serhant said he’d had a $50 million deal blown up at the last minute after both the buyer and seller consulted ChatGPT to ask about whether the property was priced correctly. (ChatGPT reportedly told the seller to ask for more money, while it told the buyer the property was overpriced.) In Florida, a man went viral this spring for using ChatGPT to price, list, and sell his four-bed, three-bath house online, only to have area realtors point out in the press that the seller had probably left upward of $200,000 on the table thanks to shoddy listing photos and a too-low initial asking price, both of which are issues a human realtor likely would have caught and addressed.

Some buyers and sellers have taken to Reddit to ask others’ opinions on whether they should use AI in their processes, weighing the viability of using it to help with negotiations, identify potential local tax increases, and review FEMA flood maps. It makes sense in theory: About a third of U.S. adults already say they use AI large language models every day, so why shouldn’t they apply that practice to real estate? That was the logic of a number of people Dwell spoke with in early reporting for this story, most of whom didn’t think the way they used AI tools during the home searching or selling process was notable enough to warrant their comment. In replies, though, most Redditors typically seem skeptical about the whole idea, rebuffing people with lines like, "Is the most expensive thing you’ll ever buy the kind of thing you want to leave up to a computer that doesn’t know many letters are in a word?" Some have come through with their own tips and tricks, however, like using Claude to create a project checklist after an offer is accepted.

Megan Brenn-White, founder of Compass’s Upstate Curious team in New York, says she noticed an uptick in buyer/seller AI use around six months ago. It’s only in the last two or three months, though, that she says it’s become an overwhelmingly large problem in the industry, leading her to chastise homeseekers’ use of AI multiple times in her company’s newsletter that goes out to potential buyers and sellers. After a New York Times story followed a reporter that successfully listed and sold his upstate New York property using an AI chatbot to gather advice on everything from pricing to the best photos to use on the listing, Brenn-White posted an Instagram Reel with 10 arguments against the writer’s case, telling Dwell that she did it because "it’s important to educate the people that the agents on my team are going to have to work with, period."

"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."

Brenn-White says her team—which is composed of 23 agents who will do about 450 deals this year—has run up against clients relying too heavily on advice from AI chatbots about things like listing prices or future interest rates, only to miss out on properties they really loved. "It feels like it went very quickly from, ‘Huh, this happened with this deal, isn’t that strange,’ to, ‘Whoa, a lot of deals are falling apart—or not coming together—because of this,’" she says. The Upstate Curious team is trying to head off any faulty information AI might deliver, though the only thing they’ve found that really works is putting any and all queries they can think of about a property into chatbots themselves in advance of posting a listing, just so they can preemptively guess what a buyer or seller might find.

Unrepresented buyers habitually use ChatGPT or Gemini to compose their offers, Brenn-White says, noting that her company had a listing where two unrepresented buyers submitted essentially the same offer in the same wording thanks to their independent use of the same AI platforms. "They were strange enough in their actual terms too that the seller couldn’t really accept them," she says. "Buyers can be unrepresented and that’s fine, but this was giving them so much incorrect information and strategy that the offers just weren’t strong. Plus, when the offers come in and they’re strange, that is probably what’s going to happen again during the inspection process or negotiation. It makes sellers not want to take those kinds of offers."

Upstate Curious agent Hillary Kolos, who was handling the listing, says she sensed something was up. "In the past we had unrepresented buyers reach out but usually if they were interested in putting in an offer they would pair up with a buyer’s agent," she says, adding that when unrepresented buyers have sent her offers, "it’s usually a couple sentences in an email and not written in a very organized way." In this case, however, both offer emails were overly organized, divided into sections with headings, and used gently persuasive language. Both offers also came in at almost exactly the same amount—about five percent under asking—leading Kolos to strike up her own informal investigation. Posing as a potential buyer for the property, she asked ChatGPT to help her make an offer, and unsurprisingly, the chatbot’s response came back at about five percent under ask and with very similar language to that in the emails.

Kolos says the sellers did end up taking one of those offers, but some fairly routine issues came up in the inspection that could have easily been addressed if the buyer had an agent to reassure them that those things were minor or could be addressed with renegotiation. That buyer then backed out and lost the $1,000 they spent on the inspection, and the sellers were forced to relist the property.

Speaking of inspections, Brenn-White says she also often runs into buyers who put a home’s reports through AI for analysis. When buyers then bring those concerns to their realtors, those professionals are somewhat limited in how they can address them. "Inspections are a very regulated field and they can be pretty terrifying if an inspector doesn’t explain the results well," she says. "As realtors, we’re not allowed to get into what they mean but we can say, ‘If the inspector says the roof is falling apart, here are three contractors that can give you a quote on the roof.’"

The issue arises when buyers also rely on AI tools to give them contractor recommendations and quotes. According to every realtor we talked to for this piece, AI chatbots are notoriously bad at region-specific recommendations, meaning if you’re asking for a roofer in upstate New York, it might give you someone working in Albany rather than in the Hudson Valley. Those realtors also say AI chatbots are bad at estimating costs, sometimes giving what could be Michigan pricing even if a home is in the Bay Area, where labor and costs are patently higher. "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," Brenn-White says. "It undermines their trust in us because if we’re like, ‘We have contractors who almost certainly can do that for less because we’ve seen this a million times,’ they’re like, ‘Well, ChatGPT says it could be this amount so this is what we want to negotiate.’"

Kati Spaniak, a realtor in the Chicago suburbs, says that when buyers use AI to parse their inspection reports, they’re often losing out on a realtor’s local expertise. "They’ll ask ChatGPT, ‘Is this an issue?’ and they’ll get generic comments, like, ‘Yes, that’s an issue,’ You need to bring in all these people to fix it," she says. "A local agent on the ground is going to be able to tell you more about that specific situation. Like if somebody’s inspection finds that a home has foundational cracks, the agent could say that these are the same kinds of cracks you’re going to find in ninety-nine percent of the homes in this location."

Shantih Moriarty, a real estate agent working with Redfin in the Bay Area, says that she’s even seen situations when AI agents hallucinate issues with inspections: "I’ve seen AI talk about cracks in foundations that aren’t there. I’ve seen it talk about particular types of piping, even if those pipes aren’t present." Brenn-White says that buyers’ misplaced trust in AI often gives heavier weight to its baseless assertions. Suzanne Goldklang, a licensed real estate salesperson in New York, adds: "Just because a computer spits something out and boldfaces and puts bullet points on it doesn’t mean it’s the best or most accurate information. AI is very good at making things look authoritative that are actually just random guesses."

Some buyers are willing to risk that authority, though, for a sense of autonomy. In Texas and Florida, newly launched real estate site Homa offers AI agents that walk users through the homebuying process before connecting them with centralized human agents to seal the deal. Aiming to reduce the amount of total commission buyers pay ostensibly by cutting one or more humans out of the process, the company says its AI chatbot is specifically trained on real estate issues in its markets. CEO and cofounder Arman Javaherian says this is particularly of interest to buyers who might like the ease or isolation that comes with a digital-only life, like ordering groceries through Instacart versus going to the grocery store. "We did a survey of a thousand homebuyers that closed in the last three years and something like sixty one percent said that they held back a question they wanted to ask their realtor because they worried it would be too annoying or cause them too much work," Javaherian says. Over 100 homes have closed or come under contract on Homa’s site to date and Javaherian says the company plans to expand into California soon.

While most real estate agents we spoke to were generally skeptical if not downright dismissive about forgoing agent expertise when buying or selling a home, some did admit clients could find value turning to AI sparingly during the process, provided they take the results with a grain of salt. For instance, Moriarty thinks AI tools can help generate questions about a report or particular listing and suggests her clients occasionally run documents like condo CC&Rs through AI, since they can sometimes run upward of 300 pages. "Nobody’s reading that in a week and coming up with questions, and there is stuff buried in there," she says. "AI can come up with a list of questions about things that you might be worried about and then you can talk to someone about those. Maybe it’ll come up with a hundred questions, ten of which are very pertinent and three of those you may not have asked. That’s valuable." She recommends clients adhere to the philosophy of "trust but verify"—specifically, with an agent—when it comes to AI. After all, like that one Redditor said, is such an expensive purchase the kind of thing you’d really want to leave up to a computer?

Related Reading:

How Realtors Are Really Using AI in the Age of "Real Estate Slop"

Ask a Realtor: I’m a First-Time Homebuyer. Should I Even Bother?