No, AI can’t replace your real estate agent

A few weeks ago, Stuart Thompson, a technology reporter for The New York Times, published a piece detailing how he’d sold his house in upstate New York without using an agent. Instead, he used AI chatbots to walk him through the process of pricing the home, writing the listing, managing agent communications and negotiating offers. He eventually sold the house for $605,000 — about $85,000 more than the roughly $520,000 he’d paid for it in 2022. In the story, Thompson claims that this figure was significantly more than the agents who'd toured the home told him he could expect.

Throughout the piece, you’re made to believe real estate agents may well be going the way of the travel agent, made obsolete by AI. Thompson himself invokes the comparison near the end of the story.

But little moments throughout the piece tell us that’s not the case. When the chatbots’ reassurances feel hollow after a string of rejections, Thompson picks up the phone and calls a friend. When he needs a quick gut check, he turns to his wife, noting that her blunt “No, that’s dumb” is far more useful than a carefully balanced AI analysis. When it’s time to close, he hires a lawyer. And in his final paragraphs, Thompson quietly concedes that replicating his results would be much harder for someone who doesn’t share his deep background in technology.

The actual takeaway, hiding in plain sight throughout the piece, is simpler than the headline implies: AI can be enormously helpful for research, planning and communication during a real estate transaction, but it still can’t replace an agent, someone who is there to help handle some of the pitfalls that Thompson himself experienced.

What AI did well

Thompson’s experiment clearly demonstrates where AI genuinely excels: providing information. Pricing research, listing language, photo sequencing, scheduling logistics and parsing document jargon are tasks that require knowledge and pattern matching, and AI has both in abundance. The chatbot understood that pricing slightly below market could trigger a bidding war. It knew that saying “I’m not playing games,” signals weakness in a negotiation. It could synthesize comparable sales into an ideal asking offer faster than any agent sitting across a kitchen counter.

The industry data backs this up. As of early 2026, nearly half of surveyed agents say they use tools like ChatGPT, Gemini or Claude at least daily. 

Zillow’s own products are built with those metrics in mind. Zillow AI mode is designed to help shoppers ask questions in plain language, understand homes and neighborhoods in context, and explore affordability before taking the next real step: scheduling a tour or connecting with an agent. And on the listing side, products such as Zillow Showcase^SM and 3D Home^® improve the information buyers get from the listing itself, with richer media, interactive floor plans and a clearer sense of the home before they ever walk through the door. 

For real estate agents, Follow Up Boss Smart Messages use AI to suggest context-aware email and text responses based on a lead’s history, which includes prior texts, calls, emails, notes and activity, helping agents respond faster.

What the story quietly glosses over

What Thompson’s account doesn’t fully reckon with is how cleanly his transaction went, and how much of that had nothing to do with AI. All three of his final bidders waived inspection and appraisal contingencies, and his eventual buyers had solid financing. Thompson treated this as normal. It isn’t. National Association of Realtors^® (NAR) data shows that only about 18% of buyers waive the inspection contingency, and 19% waive the appraisal contingency. And when deals do wobble, those problems are usually the ones AI can’t solve. In each of those scenarios, an experienced agent — one who knows which inspection findings are deal-killers and which are negotiating chips, and has contractor relationships to produce quick repair estimates — can earn their entire commission in a single phone call.

There’s also a telling moment that Thompson buries near the middle of the piece: The chatbot advised him to list the buyer’s agent commission as 0%, an action that would have subjected him to fines, since a 2024 legal settlement bars the publication of buyer-broker compensation on the MLS. He caught it only because the flat-fee listing service he was using had flagged it as a common mistake. A human real estate agent would have known that without being told.

There’s also the question of what “negotiation” actually means when a deal gets complicated. Thompson used AI to craft confident emails. That’s not the same as reading a buyer’s real motivation, knowing when a counter offer makes sense or managing the emotional temperature of a transaction when someone develops cold feet. Real estate agents know how to handle these things, and that often makes the difference. The NAR’s 2025 Profile of Home Buyers and Sellers found that agent-assisted sales fetched a median price of $425,000, compared to $360,000 for homes sold without an agent, a $65,000 gap.

The takeaway

The lesson of Thompson’s experiment is that AI can make selling a home a lot easier, but it can’t do the whole job for you. AI, as Thompson discovered, is great for pulling comps, checking your price, writing the listing, sequencing photos, scheduling, drafting replies and explaining the paperwork (all tasks that Zillow products do for sellers). But humans still tend to be better agents when the deal stops being about information and starts requiring smart judgment. At that moment, the most intelligent sellers thank AI for its good work and let an agent do what they’re best at: closing the deal.