Skip to main content
Back to Resources

Part of the Espo.ai real-estate Resources library.

Video November 10, 2026 6 min read

How to Price a Home Using AI: A Free Prompt for Real Estate Agents

How to price a home using AI: a free prompt that researches recent sales, asks before it prices, and gives you an estimated range, a spreadsheet and a one-page CMA.

#ai#real-estate#prompts
In this post
  1. The short version
  2. Why asking AI what a house is worth goes wrong
  3. The prompt
  4. What it does
  5. The three catches

The short version

How to price a home using AI comes down to one rule: make it research first, and make it ask you questions before it names a number. Ask a chat window "what is this house worth?" and it will often answer from memory, sound sure, and never check a thing.

This is the written version of the 1:39 video, with the free prompt in a box you can copy. The house in the example (1418 Wrenfield Lane, Bozeman) is made up. The market, Bozeman, is a real place. The agent, Dana Brooks, is a sample agent.

What you get back is an estimate: a likely price range and a CMA. It is not an appraisal.


Why asking AI what a house is worth goes wrong

An AI chat can sound very accurate without ever looking anything up. Sometimes it doesn't search the web at all. And a quick answer skips the questions a good agent asks first:

  • Is there an ADU or a studio?
  • Is there a pool?
  • What was upgraded?
  • How old is the roof?

An online estimator won't ask you these either. So the prompt below does the asking.


The prompt

Copy this into Claude or ChatGPT. Any plan works.

Price [address]: [type], [beds] bed, [baths] bath, [sq ft] sq ft, built [year]. [What I know.]

1. Research everything first: the record, sales near it in 6 months (same type, beds), the market. Attached export first. Can you browse? Open listing sites. Trace sales to dated pages. No sale price? List price x sale-to-list, labeled. Never a home value site's estimate.
2. Rough first read, then 5-8 questions that move price (space over the garage, solar owned or leased, roof, issues). Wait.
3. Math in code twice, per sq ft, never counting a studio's size or rent. Adjust only with a basis. Rate comps better, similar, worse. Low: median of similar and worse. High: median of similar and better.
4. An estimate, not an appraisal. Seller CMA (one page, no agent names/remarks) needs 3+ traced comps; else a labeled working range.

How to fill in the brackets:

  • [address]: the home you're pricing. In the video: 1418 Wrenfield Lane, Bozeman.
  • [type], [beds], [baths], [sq ft], [year]: what the record says. In the video: three bed, two and a half bath, about 1,860 square feet.
  • [What I know.]: the rest, in plain words. In the video: solar and an updated kitchen.

If you have a listing export your brokerage allows you to use, attach it before you send. The prompt reads the attached export first.

In my test, ChatGPT (Instant mode) asked the same questions as Claude but stopped short of giving a range. If that happens, ask it for a labeled working range and check what it gives you.


What it does

  1. It researches first. It looks at the property record, sales nearby in the last six months, and the market. It opens listing sites the way you would. In the video that was Redfin and Homes.com, then each recent sale's own page.
  2. Every sale gets a source. Each one is traced to the page it came from and the date.
  3. It gives you a rough first read. In the demo that was $680,000 to $800,000, before it knew anything else.
  4. It asks before it prices. Five to eight questions that move the price. Then it waits for your answers.
  5. It does the math twice. Per square foot, with every adjustment tied to a reason.

After the demo's answers, the range for the main house came back at $695,000 to $789,000. That is a sample on a made-up home, not a number to use for yours.


The three catches

In the video, three details from the answers could have thrown off a normal estimate. Here is what it did with each.

1. The studio over the garage. The public record said three beds, two and a half baths and about 1,860 square feet. Nothing in it, or in the message, mentioned a studio. So it asked, and the answer was a permitted 440 square foot apartment rented for $1,450 a month. It did not add the 440 square feet to the house. It did not turn the rent into a price either.

2. The solar. Solar sounds like a plus. So it asked whether the panels are leased or owned. Leased means the buyer takes over the lease. That is not an upgrade. That is a conversation.

3. The 1999 roof. It did not make up a repair number. It told you buyers will ask about it, so have an answer ready.


What you get back

  • An estimated range. A likely price range for the home, with the reasoning shown.
  • A spreadsheet with real formulas, so you can check the math.
  • A one-page CMA for the seller, labeled "not an appraisal."

Read every word before anything goes to a client. It goes out with your name on it.


Montana keeps sale prices private

Montana keeps sale prices private, so the prompt often can't find a recorded sale price. When it can't, it uses the last list price instead and labels it clearly: "verify in your MLS." Check those in your MLS before you lean on them.


Before your real export goes in

  • Use only an account your brokerage has approved for this kind of work.
  • Check that your MLS allows its data to go into an AI tool.
  • If you're not sure on either, ask first. The made-up example works fine to practice on.

The full skill is in Skool

The prompt above is the free version. The full skill is available now in my community on Skool. Join Real Estate AI for Agents on Skool to get it.

The house and the agent are made up. The Bozeman market is real. Results depend on what the AI finds that day, so read what it gives you.

Matthew Esposito
Matthew Esposito

Founder of Espo.ai. I build the growth engine for real estate teams, and I post the AI tutorials behind it.

Follow on YouTube
New videos weekly

Want this kind of system running for your team?

We work with a small number of real estate teams at a time. Subscribe for the next build, or see how the engine would look for yours.

Subscribe on YouTubeBrowse Resources