AI on the Record · evidence register

Real estate · CMA and comps · Record set RE-CM

AI CMA in Real Estate: How Agents Use Claude for Comps

Updated 5 recordsevery claim cites a recordresults self-reported

As of October 2026, our 5 CMA cases from 3 agents show Claude collecting the comps and building the CMA report, while the agent still checks the comps and sets the price. All five name Claude; none names ChatGPT. No case gives a time saved or a result number for the CMA itself, and one of the three agents sells her CMA skill as a kit.

Contents
  1. 1How agents use AI for a CMA
  2. 2What AI CMA results they report
  3. 3Where AI CMAs go wrong
  4. 4What to take from these cases
  5. 5Questions agents ask
  6. 6All cases for this task
  7. 7Records at a glance
  8. 8Related tasks

Keep two facts in mind as you read. The 5 records come from only 3 people, and 4 of the 5 records have a business tie. One agent sells the CMA skill she shows (@coritessier, 3 cases). Only one record has no business tie (@eduardodurantx).

How agents use AI for a CMA

In most records, the AI collects the data and the agent judges it. All three agents use AI to find comps and to put them into a report. Two of them say they check the comps or set the price themselves (@eduardodurantx; @coritessier).

Comps from the browser

An Austin agent installed Claude as a Chrome extension. It clicks through web pages on its own to find comps, and it builds comps reports for his clients while he does other work (@eduardodurantx).

  1. Give it the client’s property. The extension then searches the web pages for comps.
  2. Let it build the report. He films a video while it works.
  3. Review every comp. He says he does not trust anything the AI does 100%.

Comps with his own rule

An agent in El Paso County, Colorado, gets seller leads from Facebook ads. The ad form asks few questions, so he had Claude search the county website for the leads’ home addresses. Then he had it write a CMA for each lead with his own comp rule: comps within 10% of the square footage and within 10 years of the build year (Jonathan Dickey).

He says the result is close to what he gets when he comps a hard house by hand. He did not run a full CMA for every lead.

A CMA skill, sold as a kit

A team lead built five report “skills” in Claude Code (@coritessier):

  1. Market snapshot. 12 months of comps plus active and under-contract homes near an address, for a listing appointment before she has seen the house.
  2. Seller CMA. It reads her MLS through the Claude Chrome extension, or uses a CSV of comps she uploads.
  3. Buyer CMA. The same, for an offer.
  4. Farming report. One page for a farming area.
  5. Equity report. For past clients.

Every report uses the same brand fonts, colours and logo from Claude Design. She sells the five reports as “The CMA Kit”.

In a follow-up she shows the steps for a seller CMA (@coritessier). She types “I’d like to do a CMA for” an address. The first time, the skill asks for her name, brokerage, logo and headshot. Then it asks about upgrades to the home, collects comps from the MLS (or from a CSV), and returns a PDF with links back to the MLS listings. She can swap a comp she does not like.

A CMA from the phone

In a later post she says a past client texted to ask what their house was worth while she was driving. She sent one sentence to Claude on her phone. It ran her CMA skill, pulled the comps and texted her when the report was ready (@coritessier). This record comes from the post caption only; we had no transcript of the video.

What AI CMA results they report

No record gives a time saved or a result number for the CMA itself. 1 of 5 records has numbers. They are mostly about Facebook ads and AI calls. For the CMA part, that agent says Claude found addresses for 14 leads, and he pays for a $200 Claude plan for the whole setup (he says the $20 plan would do the same work) (Jonathan Dickey). Every number is self-reported.

What the agents say instead:

  • “Much more efficiency than before.” The agent who uses the Chrome extension gives no number (@eduardodurantx).
  • Close to his own comping. The agent with the comp rule compares the AI CMA with his own work by hand (Jonathan Dickey).
  • “So much time.” The kit seller says the setup will save time, with no number (@coritessier).

Where AI CMAs go wrong

  • A wrong comp costs money. The Austin agent says a wrong comp “could make or break a deal”. He says it could cost him over $10,000 in commission, and about $30,000 on this deal. So he reviews every comp (@eduardodurantx).
  • The AI does not set the value. The kit seller says AI is for collecting the data, and “you go in and make your own CMA” (@coritessier). She checks the comps and the numbers and replaces any comp she disagrees with (@coritessier).
  • MLS rules. She tells agents to check their own MLS rules before they let the Chrome extension read it. If the MLS does not allow it, she downloads a CSV (@coritessier).
  • No check before the client sees it. The phone caption mentions no review of the report before it reaches the client. The phone tool is a research preview (@coritessier).
  • A home-made rule. The El Paso County CMA uses the agent’s own comp rule, and he did not run a full CMA for every lead (Jonathan Dickey).

What to take from these cases

These five records support a small set of points. They do not show that an AI CMA is accurate.

  1. Let AI collect, not decide. All three agents use AI to find comps and build the report (@eduardodurantx; Jonathan Dickey; @coritessier). Two say clearly that they review the comps or set the value themselves (@eduardodurantx; @coritessier).
  2. Check every comp before a client sees it. A wrong comp can cost commission (@eduardodurantx).
  3. Check your MLS rules first. Some MLSs may not allow a browser extension to read them; a CSV export is the other route (@coritessier).
  4. Write down your own comp rule. One agent gives the AI his rule for square footage and build year (Jonathan Dickey).
  5. Weigh the source. Three of the five records come from one person who sells the CMA skill (@coritessier, 3 cases).

Conflict of interest, 5 records (cma and comps)

  1. Sells a course, kit or service4 of 5
  2. none found in the video1 of 5
Sells something: Jonathan Dickey, @coritessier. "None found" means we found nothing in that video, not that the person has no business interest. How we flag →

Questions agents ask

Can AI do a CMA for a real estate agent?

It can collect the comps and build the report, but in these cases the agent still checks the comps and sets the price. One agent reviews every comp Claude finds (@eduardodurantx). Another says his AI CMA is close to what he does by hand (Jonathan Dickey). A kit seller says AI collects the data and the agent makes the CMA (@coritessier).

Can ChatGPT do a CMA for real estate?

Our cases cannot say. All 5 CMA records name Claude, and none names ChatGPT. The agents use Claude through the Chrome extension, Claude Code skills or a phone (@eduardodurantx; @coritessier, 2 cases). For what agents do with ChatGPT, see how to use ChatGPT for realtors.

What does an AI CMA example look like?

In one record, the agent types an address and gets a branded PDF with links back to the MLS listings (@coritessier). In another, the agent gets a comps report for a client (@eduardodurantx). The same kit seller also shows a market snapshot for a listing appointment before she has seen the house (@coritessier).

Can I let AI read my MLS for comps?

Check your MLS rules first. One agent says so, and uses a CSV download when her MLS does not allow the Chrome extension to read it (@coritessier). She sells the CMA skill.

See every record in the case register, the full analysis in how real estate agents use AI, and the method in How we collect cases.

6. All cases for this task

Each card opens the full record: video, workflow, result, where it goes wrong, conflict of interest and source. Filter across tasks in the case register.

Current workflows (2025–26)

  • Shares numbersBuilt with a groupYouTube2026-04-06

    Jonathan Dickey

    Jonathan starts with no database and runs Facebook ads for downsizers and for new construction.

    Claude + HiIQ (AI calls) · 11 AI calls, 0 appointments; 19 leads at $7.12 each · self-reported

  • Sells the CMA KitTikTok2026-10-01

    @coritessier

    In this follow-up she shows the steps.

    Claude CMA skill (sells the kit) · No number

  • Sells the CMA KitTikTok2026-09-30

    @coritessier

    She built five report "skills" in Claude Code for her team.

    Claude Code skills (sells the kit) · No number

  • TikTok2026-08-14

    @eduardodurantx

    The agent installed Claude as a Chrome extension.

    Claude Chrome extension · No number; "much more efficiency"

  • Sells kit + waitlistTikTok2026-10-23

    @coritessier

    From the caption only: a past client texted to ask what their house was worth while she was driving.

    Claude on mobile (sells the kit) · No number

7. The records at a glance

Shares numbers the person states a result number (self-reported)Sells … the person sells AI training, a kit or a serviceHow we order cases →

Cases that share numbers and don't sell anything come first; then cases with numbers; then the rest, newest first in each group. Negative, caution and earlier-approach records follow the current workflows.
Who · toolResult (self-reported)LabelsPublished
Jonathan DickeyReal estate agentClaude + HiIQ (AI calls) · YouTube11 AI calls, 0 appointments; 19 leads at $7.12 eachShares numbersBuilt with a group2026-04-06
@coritessierReal estate team leadClaude CMA skill (sells the kit) · TikTokNo numberSells the CMA Kit2026-10-01
@coritessierReal estate team leadClaude Code skills (sells the kit) · TikTokNo numberSells the CMA Kit2026-09-30
@eduardodurantxReal estate agentClaude Chrome extension · TikTokNo number; "much more efficiency"2026-08-14
@coritessierReal estate team leadClaude on mobile (sells the kit) · TikTokNo numberSells kit + waitlist2026-10-23

Self-reported numbers, the order of cases and the "Sells …" label are explained once in How we collect cases.