Quick answer: No, not reliably. Automated tools like Zillow's Zestimate, Redfin's Estimate, and chatbots like ChatGPT can give you a number, but that number is built from public records, old listing data, and statistical pattern-matching, not from a walk-through of your home, knowledge of which side of the lake holds value, or a read on who's actually buying in Wolfeboro or Moultonborough right now. Zillow's own published data shows a median error rate of roughly 1.9% for homes currently listed and roughly 7% for homes that aren't, which on a $600,000 home is a swing of $40,000 or more. Recent tests by real estate professionals found AI chatbots inventing comparable sales that don't exist. And a 2026 Stanford study found AI models agree with users far more often than humans do, even when the user is wrong, a trait researchers call sycophancy. None of this means AI is useless. It means it isn't a substitute for someone who knows your market.

I'm Dan Batten, The NH Agent, and I work Belknap and Carroll Counties every day. I'm writing this because I'm seeing more buyers, sellers, and owners lean on AI for decisions that deserve better information, and I'd rather give you the straight version than let you find out the hard way.

Everyone Is Asking AI About Real Estate Right Now

If you've typed your address into Zillow to check your Zestimate, or asked ChatGPT what your house might be worth, you're not alone. It's become as normal as checking the weather. The problem isn't that people are curious. The problem is what happens next: people anchor to that first number. A seller who sees a high estimate gets disappointed by a real CMA. A buyer who sees a low estimate walks into a negotiation with a number in their head that has nothing to do with what the seller's home is actually worth in today's market. Either way, the AI number becomes the starting point for a decision, even though it was never built to carry that weight.

This article walks through where these tools come from, where they break down, a concept called AI sycophancy that almost nobody outside of tech research is talking about yet, and a few real, recent examples of what happens when people trust the machine over the market.

The Tools You're Probably Using (and a Couple You've Never Heard Of)

There isn't just one AI valuation tool. There's a whole ecosystem, and most consumers only see the front-facing layer.

Zillow's Zestimate is the most recognized. Zillow publishes its own accuracy numbers: a nationwide median error rate of about 1.9% for homes currently on the market and about 6.9% to 7.5% for homes that are not listed, according to Zillow's own Zestimate page. Zillow itself describes the Zestimate as a starting point, not an appraisal, and not a substitute for a comparative market analysis from a local agent.

Redfin's Estimate works similarly, pulling from MLS data and public records, and updates more frequently when a home is actively listed.

Realtor.com and RPR (Realtors Property Resource) offer their own automated estimates, generally built on similar public-record and MLS-derived data.

ChatGPT, Google Gemini, and other general-purpose AI chatbots are a different animal entirely. They aren't connected to live MLS data at all. When you ask one to value your home, it's drawing on whatever real estate content existed in its training data, which can be months or years old, and filling gaps with statistically plausible guesses. That's a very different process than pulling actual recent sales.

Then there's a layer most people never see: CoreLogic, HouseCanary, Black Knight's Home Value Explorer, and similar automated valuation models (AVMs) quietly power a lot of the mortgage industry. When you refinance or a lender considers waiving a full appraisal, there's a good chance one of these tools, not Zillow, is generating the number behind the scenes. Most homeowners have never heard these names, but they may already be shaping a financial decision in the background.

What These Tools Actually Get Wrong

They don't know your home's condition. An AVM can see square footage and bedroom count from public records. It generally cannot see that you redid the kitchen last year, that the basement has water issues, or that the primary bath was gutted to the studs. Unless that information was reported to an assessor or entered manually, it's invisible to the algorithm.

They don't know your neighborhood the way a buyer does. Two streets can sit half a mile apart and trade at very different price points because of school catchment, road noise, water access, or simply reputation built over decades. National models smooth over those differences because they're built to scale across millions of homes, not to understand one block.

They hallucinate when asked for specifics. This is the part that surprises people most. When a real estate broker in Austin, Texas asked ChatGPT to list the comparable sales it used to justify a home value, the tool returned three addresses. One had sold years earlier than claimed, one was in a different neighborhood entirely, and the third didn't appear to exist at all in MLS or county records. The chatbot didn't flag any uncertainty. It just presented all three with the same confident tone, a behavior researchers call "hallucination," where a language model generates plausible-sounding but false information because it's built to produce fluent text, not verified facts.

They're frozen in time. A chatbot's knowledge has a cutoff date. It has no idea what closed last week, what just went pending three doors down, or that a new buyer pool showed up the moment mortgage rates ticked down last month. A live MLS feed knows that within hours. A general AI chatbot may be working from data that's a year old or older.

They can't read a room, a negotiation, or a person. This one matters more than people expect. A broker quoted in a recent industry story put it well: AI cannot weigh emotional attachment as part of a negotiation, and it can't know which words will land with a widow who needs to sell a home she's not ready to let go of. Pricing and negotiating real estate isn't pure math. It's math plus people, and AI only has the math.

What "Sycophancy" Means, and Why It Should Worry You

Here's a term most homeowners haven't run into yet, but should understand before they make a six-figure decision based on a chatbot conversation: sycophancy.

In AI research, sycophancy describes a chatbot's tendency to tell you what you want to hear rather than what's true, especially when you've signaled what you already believe. It's not a bug exactly. It's closer to a side effect of how these systems are trained: they're optimized to keep users engaged and satisfied, and agreeable answers tend to score better with users than blunt or contradictory ones.

A Stanford-led study published in the journal Science in March 2026 put real numbers on this. Researchers tested eleven major AI models, including the systems behind ChatGPT, Gemini, and Claude, against thousands of real interpersonal scenarios pulled from an online community where people describe conflicts and ask for judgment. The AI models affirmed the user's side of the story about 49% more often than human respondents did, even in situations involving deception or clearly questionable behavior. In follow-up experiments with roughly 2,400 participants, even a single conversation with an agreeable chatbot made people more convinced they were right and less willing to take responsibility or compromise.

Now translate that to real estate. If you tell a chatbot "I think my house is worth $750,000," there's a real tendency for that chatbot to find a way to agree with you, or at least not push back hard, rather than tell you the comps don't support it. If you tell it "I think this offer is too low," it may validate that instinct instead of walking you through why a seller in today's market might see it differently. The tool isn't lying to you on purpose. It's built in a way that leans toward telling you what you already think, which is close to the opposite of what you need when six figures are on the line.

A good agent's job, frankly, is sometimes to be the person who disagrees with you, respectfully, with data, because that's how you avoid an expensive mistake. That's a service a system designed to keep you happy in the moment isn't built to provide.

Three Recent, Real Examples

These aren't hypotheticals. All three happened within the last few months.

A Florida homeowner sold his house with ChatGPT as his "agent," and the industry took notice. In March 2026, a Cooper City, Florida homeowner named Robert Levine told NBC South Florida he used ChatGPT for much of his home sale, from marketing materials to pricing strategy, and said it led to meaningful savings. The story made national real estate and mortgage news. What got less attention: Levine still hired a lawyer to review his legal documents, and he told reporters he didn't think AI would actually replace real estate agents. Industry professionals interviewed alongside the story were more pointed. One South Florida broker noted that AI cannot weigh emotional attachment in a negotiation or know which words will resonate with a grieving seller. A National Association of Realtors innovation director added that a Realtor advocating for your interests isn't an optional step you can skip just because the technology is impressive.

A practicing Realtor tested ChatGPT against his own live MLS data, on a house he knew cold. In May 2026, an Austin-area broker ran an experiment: he gave ChatGPT detailed information about a specific home he was deeply familiar with and asked it to estimate the value, deliberately withholding live comp data to see what the average homeowner would get. ChatGPT came back with a value range, and explained its reasoning by citing "recent comparable sales." When the broker asked it to list those comps, one had sold years earlier than represented, one was a different neighborhood, and the third couldn't be found in MLS or county records at all. His real CMA, pulled from live MLS data with proper comps, came in roughly 11% higher than ChatGPT's estimate, a gap of about $60,000 to $75,000 on a home in the mid-$600,000s. He was careful to note he isn't anti-AI. His conclusion: AI fed real, current local data can be a useful research partner, but AI guessing from old training data is a liability when there's real money on the line.

A new study reframed an old debate about Zillow's Zestimate. In May 2026, reporting on a federal antitrust lawsuit between Zillow and a Chicago-area MLS noted that Zillow itself had used Zestimate data, a metric the article pointed out some critics consider inaccurate, as the basis for a separate company study on private listings. The episode is a useful reminder that even the company behind the most well-known home value tool treats the Zestimate as directional, not definitive, when the stakes (or the argument) are big enough to matter.

What This Means for Buyers

If you're shopping in Belknap or Carroll County right now, an AI tool might tell you a property is overpriced based on a generic per-square-foot calculation. It has no idea that the property backs up to conservation land, that it's one of the few homes left with deeded lake access on that particular pond, or that three other buyers are already circling it because, over the last 30 days, a third of similar homes in Moultonborough went under agreement in less than a week, while more than half of the similar homes still sitting on the market have been there longer than the average days on market. That gap between the homes that move fast and the homes that linger is exactly the kind of signal a national algorithm can't see, but it tells a buyer almost everything they need to know about how to approach a given listing. Trusting an AI number over what's actually happening in the market can cost you the house, or worse, cost you a fair negotiating position because you walked in anchored to a number that was never grounded in reality.

What This Means for Sellers

If you're getting ready to list, an AI estimate is one of the worst ways to set expectations before you talk to an agent. Price too high because a chatbot or an off-market Zestimate told you to, and you risk sitting on the market while buyers assume something's wrong with the house. Price too low because you didn't account for your specific lake frontage or access, or a design element that's in demand with buyers right now, and you leave real money on the table that the market would have paid. A proper CMA accounts for what's actually closing right now, in your specific neighborhood, not a regional or national average wearing a local disguise.

What This Means for Owners Not Currently Buying or Selling

Even if you're not transacting today, checking your Zestimate or asking ChatGPT what your home is worth "just to see" isn't harmless. That number sticks in your head. It shapes how you think about refinancing, how you talk to your insurance company, how you plan for retirement income if you're thinking about downsizing or tapping equity down the road. If you want a number you can actually plan around, ask for a real market analysis. It costs nothing and it's grounded in what's happening in your town right now, not a five-year-old training snapshot.

The Lakes Region Doesn't Fit in an Algorithm

This is the part that matters most if you live here. Belknap and Carroll County real estate doesn't behave like a national model expects it to.

Waterfront and water-access properties trade on criteria no AVM captures well: lake frontage versus deeded access, septic and well capacity on older shoreline lots, dock and boathouse rights, and the difference between, say, a Lake Winnipesaukee address and a quieter pond a few miles inland. Buyer behavior up here has been shifting too. Buyers know they're paying premium prices, and even in our high-demand market, they've slowed the pace and raised their standards. Instead of making an offer on a home they feel is a little overpriced, they're choosing to wait for the next opportunity rather than stretch on something that doesn't quite fit. That's a real, current shift in buyer psychology, and it directly affects how a home should be priced and marketed today versus six months ago. None of that shows up in a Zestimate. None of it shows up in a ChatGPT answer either, because that information lives in conversations with buyers, sellers, and other agents working this market every day, not in a public dataset.

What a Professional Agent Brings to the Table That No AI Tool Can

I'm not going to tell you AI is useless, because it isn't. I use it myself for parts of my business. What I will tell you, plainly, because I think you deserve straight talk on something this important, is that pricing and negotiating real estate in this region is built on relationships, pattern recognition built over years of transactions, and judgment calls that data alone can't make. It's knowing which buyers are circling which towns before the data catches up. It's knowing that the house down the road sold fast because the buyer needed to close before a school deadline, not because the market is suddenly hotter everywhere. It's being the person who tells you the number you want to hear isn't the number the market will actually pay, and then backing that up with real comps, not vibes.

An algorithm can give you a starting point. It cannot replace someone who's walked the dock, talked to the neighbors, watched this specific stretch of shoreline for years, and will be in the room when the negotiation gets hard.

If you're thinking about buying, selling, or just want an honest read on what your property is actually worth in today's Lakes Region market, reach out. I'll give you real numbers, not a guess dressed up in confident language.

Dan Batten, The NH Agent Coldwell Banker Realty | Moultonborough, NH 603-259-4788 | Dan@TheNhAgent.com | www.TheNhAgent.com

Sources

  1. Zillow, "What is a Zestimate?" (Zestimate accuracy methodology and published error rates) — https://www.zillow.com/zestimate/
  2. National Mortgage News, Andrew Martinez, "What happens when you use ChatGPT to sell your home?" (March 18, 2026) — https://www.nationalmortgagenews.com/news/what-happens-when-you-use-chatgpt-to-sell-your-home
  3. NBC South Florida, "Man uses ChatGPT to sell his Cooper City home: 'It exceeded our expectations'" — https://www.nbcmiami.com/news/local/innovation-on-6/man-uses-chatgpt-to-sell-his-cooper-city-home-it-exceeded-our-expectations/3778919/
  4. Neuhaus Realty Group, Ed Neuhaus, "ChatGPT Told Me My House Was Worth $X. I'm a Realtor. Here's the Truth." (May 15, 2026) — https://neuhausre.com/chatgpt-home-value-test/
  5. Cheng, M. et al., "Sycophantic AI decreases prosocial intentions and promotes dependence," Science (March 2026) — https://www.science.org/doi/10.1126/science.aec8352
  6. TechCrunch, "Stanford study outlines dangers of asking AI chatbots for personal advice" (March 30, 2026) — https://techcrunch.com/2026/03/28/stanford-study-outlines-dangers-of-asking-ai-chatbots-for-personal-advice/
  7. The Real Deal, "Reactions to Zillow-MRED lawsuit expose fault lines in fight over listing data" (May 17, 2026) — https://therealdeal.com/chicago/2026/05/17/zillow-lawsuit-deepens-industry-rift-over-listing-data/