Artificial intelligence has found its way into nearly every part of the home buying and selling process. Buyers use it to evaluate neighborhoods. Sellers use it to second-guess their agent. And sometimes, it tells them exactly what they want to hear, because they asked it to.
This is the story of one of those situations. It is not hypothetical. It cost our clients real money, and it is worth sharing.
The Setup
We were engaged to list a property in the northwest suburban Chicago market. Our team conducted a thorough analysis covering the specific address and its characteristics, the condition of the home, comparable sales within the prior 12 months, active competition, current inventory levels, days-on-market trends, and prevailing list-to-sale price ratios. We also developed an integrated marketing strategy combining professional video, targeted digital advertising, print, and a reverse-prospecting campaign to surface buyers already in the pipeline at cooperating brokerages.
Our recommended list price was $480,000 (slightly below comparables to drive the price up with a bidding war).
Our client disagreed. They believed the home was worth $550,000 to $580,000, based on their own review of comparables. That is a reasonable instinct to have. Sellers know their homes. But rather than have a direct conversation about the data behind our recommendation, the client turned to AI.
The Question They Asked
The client submitted this prompt to ChatGPT, Claude, and Gemini:
"In the northwest suburban Chicago real estate market, how important is it for a new residential listing to sell quickly as opposed to being on the market for some time? I'm asking because we're contemplating pricing strategy. The agent wants us to price lower. I think we need to be higher and I'm not overly concerned if it takes a while to sell. I did a marketing analysis based on recently sold comparables and also on-market properties. Two different analyses. As best I can tell our property is valued between $550,000 and $580,000. The realtor wanted us to list at $480,000. That seems absurd."
Each model responded thoughtfully. ChatGPT validated that a $70,000 to $100,000 gap was "large enough that I would also question it," discussed buyer psychology around search filters, and suggested a list price around $539,000 to $559,000. Gemini called the gap "immense" and noted that deliberately cutting 15 to 20 percent below baseline value "crosses the line from strategic to highly risky." Claude affirmed the seller's instincts and concluded that pricing at $565,000 "is not reckless," while "leaving $70,000 to $85,000 on the table because your agent wants a quick transaction absolutely is."
These are not bad responses. They are, in fact, quite good responses given what the AI models were asked. That is the entire problem.
What the AI Did Not Know
Not one of the three models asked for the property address. None asked about condition. None asked for the average days on market in that specific price band. None asked for the list-to-sale price ratio our team had pulled from actual closed transactions. The models asked about comparable data, but the client never provided it, because the client disagreed with the direction the conversation was heading.
The AI answered the question it was given. The question was loaded. And so the answers, however well-reasoned, were built on a foundation of incomplete and framed information.
What Happened Next
Armed with AI validation, the client directed us to list at $580,000. We listed. We marketed. We executed.
No offers.
We reduced to $565,000 per our clients request. During this phase, we hosted a brokers open. Fourteen agents attended and we administered a survey. Of the 11 who responded, 8 rated the $565,000 list price as "High" relative to perceived value. The remaining 3 called it "Slightly High." Not one respondent said the price was fair or low. The market was telling us something. It had been telling us something from day one.
No offers.
We reduced to $549,000. Then to $525,000.
No offers at either price.
We took the property off the market for 30 days to let the days-on-market clock reset, then re-listed at $480,000. We hosted an open house. We received multiple offers. The home sold for $540,000.
Let that sink in. The property sat through four price reductions over roughly four months, generated zero offers, and ultimately sold for $540,000 after resetting to the price our data supported from the beginning. The client's decision to override that recommendation, reinforced by AI responses to a leading question, cost them months of carrying costs, market momentum, and the psychological weight of a prolonged sale.
The Lesson
AI models are exceptionally capable tools. Our team uses them too. But they are only as good as the information they receive, and they have no ability to push back on a framing that leads them toward a predetermined conclusion.
When you ask "my agent wants me to price low and I think that's absurd, am I right?" you are not asking an objective question. You are asking a leading one. And AI, like any capable communicator, will engage with the question as presented.
Your agent's pricing recommendation, when it is built on actual address-level data, real closed sales, condition-adjusted comparables, current absorption rates, and days-on-market analysis for your specific price band, cannot be replicated by a chatbot in 30 seconds. That recommendation is the product of market access, licensed expertise, and local pattern recognition that no public AI model currently has.
If you are going to use AI in your real estate process, we encourage it. But ask objective questions. Ask it to help you prepare for a conversation with your agent, not to replace one. And if you find yourself only submitting the data that supports the conclusion you already want, consider that the gap between what you believe and what the market reflects may be wider than you think.
Chances are, if you are using AI, your agent is too. The difference is that your agent's AI is working with the actual data. Let the data tell the story.
