AI in real estate, signal vs noise

When a client brings you a ChatGPT valuation: the questions that expose its blind spots

6 min read

A client arrives with a confident number from a chatbot. The right questions do the work.

The short version

A general AI chatbot can describe a market but cannot price a specific home, because it has no live access to the comparable sales, the condition, or the block that set the price. When a client brings a number from one, arguing it is wrong invites a fight. Asking a few plain questions, where the number came from, which recent sales it used, what condition it assumed, lets the gaps show themselves. The goal is not to win the point. It is to move the client from a confident number to the right one.

Why a chatbot number sounds right and often is not

A large language model is a system that produces text by predicting likely words, not a valuation tool with live access to sales records. It writes in fluent, confident prose, and that fluency reads as authority, which is not the same as accuracy. The model has no live feed of the local multiple listing service, no address-level record of what sold last month two streets over, and no way to see the specific home. It produces a plausible-sounding number by pattern, not by pricing this house against real recent sales. For scale, even a data-driven tool struggles on homes like these: the Zestimate carries a published median error near 7.5% on off-market homes, about $45,000 on a $600,000 home, and a general chatbot works without even that much local data. The output can be articulate and still be built on nothing local.

The questions that expose the gaps

You do not have to challenge the number head-on. A short series of plain questions lets the client watch the estimate come apart on its own.

Question to ask	What a weak answer reveals
Where did the number come from?	No source, or a vague reference to averages, which means it was not built from this home
Which recent nearby sales did it use?	It cannot name live comparable sales, because it has no address-level access to them
What condition did it assume?	It assumed an average home, so a renovated or dated house is mispriced from the start
How current was its data?	The underlying data may be months or years old, which misses a moving market
What range did it give, and why?	A single point with no range and no reason, which hides how little it really knows

What AI can and cannot do in pricing

The honest line runs between summarizing and pricing. A capable model can explain how the local market has moved, summarize trends, and draft the words around a valuation. It cannot see the street, cannot pull the specific comparable sales, and cannot judge condition. Much of what is sold as a new AI valuation is the old automated-model method with a conversational voice on top, which is the distinction drawn in what AI changes in pricing, and what it relabels.

Turning the number into a listing conversation

Handled well, the client walking in with an AI number is an opening, not a threat. The sequence is simple. Ask how they arrived at the number. Show, gently, what it could not have seen about their home. Verify a few of the key comparable sales together, on the actual recent sales. Then let the facts move the number, rather than telling the client they are wrong. The seller ends up at the right price and feels they got there themselves, which is the whole point of the appointment.

What this means for an agent

Treat the AI valuation as a listing-appointment advantage. The seller has already done the thing that opens the door, they have gone looking for a number, and they have arrived with one they are not sure they can trust. Your edge is not a more confident number of your own. It is a reasoned range with the recent comparable sales behind it and an honest statement of what is and is not known about their home. That is the standard the whole category should be held to, set out in how to judge an AI valuation tool. The agent-facing version of this play, showing a client what an automated estimate missed on their specific home, is the basis of our home-value accuracy page, and the reasoned consumer estimate that models the honest-range approach is the home-value page.

Frequently asked questions

Can ChatGPT accurately value a house?

Not for a specific home. A general chatbot has no live access to local comparable sales and cannot see condition or layout, so it produces a plausible number rather than a priced one. It can describe a market but cannot reliably value a particular property.

What should I do when a client brings an AI home valuation?

Do not attack the number. Ask where it came from, which recent sales it used, and what condition it assumed. Those questions expose the gaps, and then you can price the home on real comparable sales and let the facts move the client to the right number.

Why does an AI give a confident home value with no real data?

Because a language model is built to produce fluent, confident text, and that confidence is a feature of the writing, not evidence of accuracy. It can generate a precise-sounding figure with no live comparable sales behind it, which is why the tone should not be mistaken for reliability.

Is an AI valuation better than a Zestimate?

Not inherently. A Zestimate at least runs on records and comparable sales, while a general chatbot often has neither for a specific address. Both miss condition and layout, and both should be treated as a starting point, not the answer.

How do I show a client the AI number is off without arguing?

Ask questions instead of making claims. When the client sees that the estimate cannot name the recent sales it used or the condition it assumed, the gaps speak for themselves, and you can move to real comparable sales without a confrontation.

Sources: Public descriptions of large language model behavior and their lack of live real estate data access, Zillow Zestimate methodology for comparison, and standard real estate valuation references on comparable sales and condition. Capabilities change quickly; treat these as general references.

This article is general information and analysis, not financial, lending, or appraisal advice. Verify any home value with a licensed professional before acting.