Claude AI MLS Data: 6 Jobs It Does for Buyers and Sellers

I connected Claude to my MLS about a month ago, and the thing that surprised me wasn’t the searching. It was how much of my week turned out to be formatting.

Pulling comps into a document. Typing showing instructions into a tour sheet. Writing the same weekly email to five sellers about what moved in their neighborhood. None of that is the job I got licensed for, and all of it was eating afternoons.

I’m William Zhang, a real estate agent here in Austin and the founder of Real Estate AI Society. Once Claude AI has MLS data behind it, there are six jobs I now hand it every week, three on the buyer side and three on the seller side. I want to walk through all six with the real documents it produced, including the one place it got a read wrong.

A warning before any of it. How you connect matters more than whether you connect. I ran a headless browser against my MLS, it moved through listings faster than a person could click, and I got a warning letter for it. I wrote up the three ways to connect Claude to your MLS and what each one actually costs you in compliance. Read that first if you haven’t.

What Having MLS Data Actually Changes

My feed carries 2.1 million property records, 1.45 million closed sales going back to February 1996, and 523 licensed fields. There were 73,510 closings in the last twelve months alone.

That depth is the whole point. An assistant with no MLS access will still answer your pricing question, confidently, off asking prices it found on a public site. Texas is a non-disclosure state, so sold prices aren’t published on Zillow or Redfin at all. You end up pricing someone’s house off data that doesn’t exist.

With the feed, the six jobs below stop being research projects and start being prompts.

Buyer Job 1: Scheduling the Tour and the Showing Brief

A buyer sent me a list of homes he wanted to see. I gave Claude the list and asked it to schedule the tour.

What came back was a tour sheet with all three homes, in route order, plus the emails to send to get each one on the calendar. That part is convenience. The part I didn’t expect to lean on is the second document it built, an internal showing brief that I never hand to a client.

Here’s what was on it for the first house, 17333 Graces Path in Pflugerville:

• Do not contact the listing agent. Limited listing service, call the builder’s sales counselor directly at the number on the brief.
• 140 days on market.
• Original list $644,251. Currently $525,000. The builder has cut $119,251, which is 18.5%.

That last line is a negotiation talking point I would have had to go dig for. It was sitting in the price history the whole time and the brief surfaced it without me asking.

The second house on the tour, 21013 Abigail Way, went the other direction. Twelve days on market and the seller had raised the price $49,000, from an original $450,000 to $499,000. The same house closed at $302,585 back in 2017, so I know the owner’s basis walking in. Interesting strategy in a market where 3,780 homes took a price cut across the Austin MLS in the previous seven days.

The brief also carries schools, tax rate, HOA and carrying costs. If I were working investors I’d have it add return on investment and an estimated payment at today’s rate instead. It’s a template, not a fixed form.

Buyer Job 2: The Property Sheet Clients Carry Through the House

This one is small and it lands every time.

For each home we’re touring, Claude builds a one-page sheet from the MLS record: square footage, year built, lot size, price per square foot, schools, tax rate, HOA, days on market, price history. I print them, put them in a folder, and hand it over at the first house.

Buyers read along with paper in hand instead of squinting at Zillow on a phone. It shows things Zillow doesn’t show, and it feels like the agent did something. Once the template exists it costs nothing per house.

Buyer Job 3: The Price Check Before We Write the Offer

Before we write an offer, I want my buyer to understand the number, not just approve it. So Claude runs a price check on the subject house and puts it in a document I can walk through on a Zoom call or print and go over in person.

For Graces Path at $525,000, here’s what it pulled for that submarket, 2,600 to 3,150 square feet in Blackhawk:

• 67 closed sales in the last twelve months. Median $197 per square foot, median price $560,000, median 61 days on market.
• At the closed median, 2,864 square feet indicates $564,705.
• 52 active competitors. Median $200 per square foot.
• The subject is asking $183 per square foot, which is $14 under the closed median.

Then the comps that actually make the argument. 19800 Zook Road is exactly 2,864 square feet, same build year, and it closed at $644,000 in July at $225 per square foot. A house on the same street, 179 square feet smaller, is active at $614,990.

Now here’s the part I want to be straight about, because I said it a little too fast on camera. In the video I look at that $564,705 against the $525,000 ask and say we’re getting a really good deal.

That’s further than the data lets me go. The feed can’t see the lot, the orientation, the finish level, or what’s behind the builder’s decision to come down $119,251 after 140 days of no takers. A $39,705 gap to the closed median is either a real repricing or the price is telling you something specific about that house, and nothing in the MLS distinguishes between the two. The honest move is to go see it and ask the sales counselor directly what changed between $644,251 and $525,000.

That’s the general rule with all of this. The pull is fast and reliable. The read is still yours.

Seller Job 4: How I Run a CMA With AI

A seller lead came in on a tenant-occupied rental in Austins Colony, 1,528 square feet, three bedrooms plus a study, built in 2001.

Claude pulled 17 closed sales in that subdivision between 1,350 and 1,750 square feet over the last twelve months, with the financing type on each one. Median $173 per square foot, median close $267,000, median 43 days on market, median sale closing 2.5% under its original list price.

Two of those 17 were distressed cash sales at $102 and $119 per square foot. I had it run the set again without them: median $179 per square foot, which indicates $273,056.

The comp that carried the whole report was an identical 1,528-square-foot house four doors down that closed at $285,000 at $187 per square foot.

Indicated range: $265,000 to $285,000, with $275,000 as the point estimate.

Then the part that made this a real recommendation instead of a number. The feed also has the ownership history. This owner bought it for $288,388, all cash, in August 2024, and it currently leases at $2,000 a month. So selling today at $275,000, less roughly 6% cost of sale, nets about $258,500 against $288,388 in. That’s a shortfall of around $29,900. Meanwhile the rent is $24,000 a year gross on that basis.

The recommendation I gave was not to list. Hold the tenant, revisit when the active count in the subdivision thins out or the median clears $190 a square foot. That’s the answer that wins the listing in eighteen months, and it’s also just the true one.

One adjustment I make every time: Claude defaults to a twelve-month comp window and I set mine to six. Austin is moving quickly enough that twelve months blends two different markets. Your market might want three, or eight. That’s a setting you should own, not inherit.

Seller Job 5: Turning the CMA Into a Pricing Strategy

This is the one I think is most underrated, because it’s the difference between having comps and winning the listing.

Same subdivision, all closed residential sales in the last twelve months, 37 of them. Claude pulled the behavior, not just the prices:

• 54.1% of sellers cut their price at least once. That’s 20 of 37.
• The median sale closed 4.7% under its original list price.
• 13 of the 37 sold in 30 days or less, and those got a median of 100% of their original asking price.
• 9 of the 37 took more than 90 days, and those closed a median 13% under original list.
• 13.5% were cash buyers, and 41% of the closings were FHA or VA, which means the house has to appraise.

Read those last two bullets together and you have the entire overpricing conversation in one line. Price it right and you get your full ask inside a month. Price it wrong and the market charges you 13%, which on the $267,000 median is $34,710.

It’s hard to argue with that. It’s very easy to argue with an agent’s opinion. When a seller says the neighbor got their price, I can say that 20 of the 37 homes that sold here cut at least once, and I didn’t have to spend an evening counting.

I still write the recommendations myself, and they’re short: price at today’s number, not last year’s. Don’t test high and correct down. I’ve tried that here and the home sits longer and nets less.

If listing presentations are where you want to start, I broke down the rest of that workflow in my post on the AI seller assistant.

Seller Job 6: The Weekly Seller Update, Written for You

Every agent knows they should send sellers a weekly market update. Almost nobody does it consistently, because it’s an hour of MLS work and writing per listing per week.

Claude now builds mine. Last 30 days in that same subdivision, it found 26 distinct listings with a change event and 37 separate events:

• 12 price decreases
• 6 went active under contract
• 5 new listings
• 4 back on market
• 3 on hold, 3 withdrawn
• 2 pending, 1 expired, 1 closed

The biggest cut in the window was $26,500, and that house was withdrawn 16 days later. Another went down $25,500 and then $24,500 again, $50,000 in two moves.

That’s not a newsletter. That’s a reason for a seller who won’t budge to have a different conversation with you. When your listing is the only one in the neighborhood that hasn’t moved on price, the email makes the point for you, and it arrives every week whether or not you had time to write it.

Where You Still Have to Be the Agent

None of these six replace judgment, and the price check above is the proof. The assistant handed me an accurate set of numbers and I put a slightly too optimistic read on them out loud.

The comp window is your call. The distressed sales you exclude are your call. Whether a hold recommendation or a list recommendation serves the client is very much your call. What changes is that you walk into the appointment with the data already assembled, instead of spending the afternoon before assembling it and the appointment itself improvising.

If you want the prompts I started from, the Claude for Real Estate Agents Starter Guide is free.

Start With the One That Eats Your Week

Don’t try to build all six. Pick the job that costs you the most hours right now, which for most listing agents is the weekly seller update and for most buyer’s agents is the tour prep, and teach the assistant that one first. If you’re not sure client service is even your bottleneck, my roadmap for what to build first covers how to check.

Then get the connection right before you scale it. I’m building an MCP connector so getting an assistant onto your MLS stops being a procurement project, and you can get on the waitlist here. If you want to see how these six fit into everything else I run, the full system from lead gen to closing is the other half of this.