Write Real Estate Listing Descriptions Faster: Cut Portal Copy From a Half-Day to 10 Minutes With Claude Code

Friday at 4:45 PM. Your broker drops three fresh listings on your desk and says, “Get these live on the portals first thing Monday.” Sound familiar?

A buddy of mine in sales used to burn his entire Saturday morning writing listing copy. The floor plan and the distance to the train are easy enough to pull off the fact sheet. But by the time he’s squeezed out “bright and sunny” and “quiet residential street” for the third property in a row, every description reads the same. Forget standing out from the competition. He couldn’t even tell his own listings apart.

Then Monday morning, his manager glances at the screen: “This two-bedroom description is basically word-for-word the same as the unit we listed last week.”

Writing listing descriptions is the one task in real estate where hard work almost never gets noticed. It eats time. Quality lives and dies with whoever happens to write it. And yet it’s the front door for every inquiry. In this article, I’ll show you how to hand the grunt work to Claude Code so you only spend time on the one thing that actually needs a human: the final call.

Key takeaways

• Handing the “first-draft factory” work to Claude Code turns a half-day Saturday of writing listing and portal copy into roughly 10 minutes per property.
• Let the AI generate the prose and catch inconsistent wording. You keep three things for yourself: fact-checking, fair-housing/false-advertising compliance, and private data.
• Feed it the raw data from a property fact sheet, and it writes three distinct versions, one tuned for Zillow, one for your MLS feed, one for your own website.
• I’ve included a copy-paste prompt template and a check script that mechanically flags exaggerated or non-compliant words.
• Never hand the AI a client’s name, phone number, or seller details as-is. Decide your masking rules before you start.

Who this is for, and where the workflow jams up

This article is written for the person who takes on listings themselves and posts them to the portals, whether that’s sales or rentals. Small brokerages that can’t afford a dedicated marketing person, and agents or managers running the whole thing solo, will get the most out of it.

Lay out the steps to getting a listing live and it looks roughly like this:

1. Take on the listing from the seller or landlord (fact sheet, title docs, and on-site photos come together).
2. Enter it into your MLS.
3. Post the description and images to Zillow, your MLS feed, and your own website.
4. Work the leads that come in.

Step 3 is where it jams. Steps 1 and 2 are mostly transcription, so they’re fast. Step 4 is people work. But step 3 has “writing prose” wedged into it, the one task where volume and quality fight each other. On top of that, every portal has different character limits and a different recommended tone, so even the same property needs to be rewritten three ways. That’s the thing eating your Saturday morning.

The rework that keeps happening, and what’s really behind it

The rework in listing-copy work boils down to three types, almost every time.

• Copy-paste reuse that gets caught. You base the new description on a previous property’s text and forget to update the station name or square footage. It says “4-minute walk” when it should say “7-minute walk,” because that’s what the last listing said.
• Inconsistent wording. The same office writes “south-facing,” “southern exposure,” and “great natural light” on different listings, and the whole site looks sloppy.
• False-advertising flags. You slip in “best in the country,” “perfect,” or “guaranteed deal,” phrases that violate fair-advertising rules, and the portal kicks the listing back.

All three happen in the late afternoon, when focus is gone. That’s exactly why you want to shift the work toward “things a machine can check.” Claude Code is good at killing this rework at the draft stage.

What actually changes, before and after

The numbers will shift with your situation, but here are the rough figures I measured for myself. Use them as a starting point for your own decision.

Item	Before (writing by hand)	After (Claude Code draft + human review)
Listing copy per property	~40 min	~10 min (3 min generate + 7 min review)
Rewriting for 3 portals	+15 min each	All 3 in one prompt
Inconsistent-wording check	Eyeballing only, misses things	Mechanical check via script
Time for 20 properties/month	~13 hours	~3.5 hours

Roughly speaking, a shop handling 20 properties a month frees up 9-10 hours a month. Value an agent’s time at $40 an hour and that’s around $400 a month in saved cost. The real payoff is redirecting that time to showings and follow-up.

You might be thinking, “AI writing comes out thin and generic.” It’s the opposite. Hand it structured raw data and the output is more specific than the recycled copy you’d bang out at 5 PM. For the basics of getting Claude Code running, see the Claude Code getting started guide, and if you’re not an engineer, Claude Code for non-engineers walks you in gently.

Three real-world use cases

Use case 1: One fact sheet, three portal versions at once

This is the big one. You hand over the property summary (address, price, layout, square footage, distance to transit, amenities) as a bulleted list, and it writes three distinct versions: one for Zillow (readability first), one for your MLS feed (amenity coverage first), and one for your own website (story-driven).

Your job is just to read the three drafts and confirm the facts are right. Reviewing is dramatically faster than writing from scratch.

Use case 2: Brainstorming headlines

Feed it the selling points, like “south-facing corner unit, 5-minute walk, pets OK,” and have it spit out ten headline options. This is where the AI’s sheer volume pays off. Picking one of ten is a human job. Choosing beats forcing yourself to invent ten alone, and the quality goes up.

Use case 3: Drafting inquiry replies

Have it prep a polite, templated first reply to “Is this still available?” On the assumption that a human always reads it before sending, you just swap in the availability status or showing times. Now you can answer a midnight lead first thing the next morning.

What to delegate to AI vs. decide yourself

Get fuzzy here and you’ll have an accident. Draw the line clearly.

Step	Delegate to AI	A human must decide
Writing prose	Yes: drafts, variants, brainstorming	Final phrasing
Facts (price, sq ft, distance)	No: never trust as-is	Yes: cross-check against the fact sheet
False advertising / compliance	Partial: flag candidates only	Yes: final OK call
Private data	No: don’t hand it over	Yes: mask first, then hand over
The publish button	No: never let it click	Yes: a human clicks it

The mantra is “AI is the draft writer, the human is the editor-in-chief.” You’re the licensed broker, and you own responsibility for what goes live. That part never gets handed off.

A copy-paste prompt template

Use this as-is. Swap whatever’s inside the [ ] brackets for your own property details. Don’t include client or seller names (more on that below).

You are a veteran listing copywriter at a real estate brokerage.
From the property summary below, write 3 versions of a listing description
for online portals.

# Property summary
- Type: [Resale condo]
- Price: [$348,000]
- Layout: [2 bed / 1 bath, 625 sq ft]
- Location: [Lincoln Park, Chicago, IL]
- Nearest transit: [Red Line, Fullerton stop, 5-min walk]
- Year built: [2015]
- Exposure / floor: [South-facing / 7th floor]
- Amenities: [Secure entry, package room, in-floor heating, in-unit laundry]
- Notes: [Top-floor corner unit, pets considered]

# Output rules
1. For Zillow (~300 words, readability first, no exaggeration)
2. For the MLS feed (~250 words, cover all amenities)
3. For our website (~400 words, paint the lifestyle)
- Do not use prohibited terms (best, perfect, guaranteed, dirt-cheap, etc.)
- Keep facts exactly as in the summary; do not invent amenities or conditions
- At the end, list "items that need fact-checking" as bullet points

The trick is making it flag “items that need fact-checking” at the end. The AI tends to fill gaps with guesses, so having it declare “this part is shaky” up front makes your review much faster. If you want to sharpen your prompts further, read advanced prompt engineering for Claude Code alongside this.

A check script that mechanically flags exaggerated wording

Eyeballing whether a generated description still has banned words in it will always miss something at 5 PM. Here’s a check script that runs on Node.js. Save your listing copy to a text file and just run node check-listing.mjs listing.txt.

import { readFile } from "node:fs/promises";

// Words that are prohibited or risky under fair-advertising rules
const banned = [
  "best in the country", "best in the industry", "best", "top-tier",
  "completely", "perfect", "guaranteed", "dirt-cheap", "super cheap",
  "steal", "buy now", "once in a lifetime", "limited time only",
  "hand-picked", "exclusive deal", "rock-bottom",
];

const file = process.argv[2] ?? "listing.txt";
const text = (await readFile(file, "utf8")).toLowerCase();

const hits = banned.filter((w) => text.includes(w.toLowerCase()));

if (hits.length === 0) {
  console.log("OK: no banned words found.");
} else {
  console.log("Needs fixing: the following terms are present ->");
  for (const w of hits) console.log("  - " + w);
  process.exitCode = 1;
}

// Also report the character count (Zillow copy that's too long gets skimmed)
console.log("Character count: " + [...text].length);

This is only a first-pass filter. Compliance is judged in context too, so a human’s eyes are still required at the end. Even so, it stops you from missing an obvious banned word with a tired brain at the end of the day. To add more checks, just append words to the banned array. If you put this banned-word list in your CLAUDE.md, Claude Code will avoid them from the moment it generates. For how to write that file, see CLAUDE.md best practices.

Privacy and security considerations

Real estate is a business built on piles of personal information. You can’t skip this part.

• Never hand the AI the names, phone numbers, or addresses (down to the unit number) of sellers, buyers, or landlords. The listing copy only needs the property’s specs.
• Don’t hand over title records or ownership details either. They’re irrelevant to writing a description.
• Mask before you hand anything over. Keep the address at “Lincoln Park, Chicago,” and hide the building and unit number.
• When drafting a reply to a lead, don’t paste the inquirer’s message verbatim either. Pull out just the request and hand that over.
• If you’re using it as a company, confirm with your IT or legal team that you’re on a contract where your input data isn’t used for training.

The point is: “the AI only sees information you’d publish anyway.” Stick to that and you’ve cut the data-leak risk in listing work way down. For everyday usage tips, I’ve collected some in Claude Code productivity tips. For the precise scope of fair-housing rules, always confirm the primary source on the official HUD Fair Housing site.

Pre-publish checklist

Before you hit publish, run these by a human’s eyes. Honestly, you could print this and tape it to your desk.

• Do the price, square footage, layout, and transit distance match the fact sheet?
• Is there any leftover station name or number reused from a previous listing?
• Banned words (did you run it through the check script)?
• Did it invent amenities that aren’t listed?
• Is any private data or unit number buried in the body text?
• Does the photo count meet the portal’s rules?

FAQ

Q. Will people be able to tell the copy was written by AI? Text written from raw data you fed it is more specific than the recycled copy you’d mass-produce in the late afternoon. That said, skip the human review and you’ll leave awkward phrasing and factual errors in. Use it as a draft and polish the final version in your own words.

Q. Could this violate the law or MLS rules? Responsibility for the listing sits with the licensed broker. The AI only produces a draft; the final compliance call is a human job by design. Protect yourself with both the check script and the checklist.

Q. Can I use it even if I’m not good with computers? You can hand off the script-running part to someone technical, just for the first setup. If all you’re doing is writing a prompt and getting a draft, you only need to be able to type text. Starting from getting started for non-engineers makes it less likely you’ll trip up.

Q. Does it work the same for both rentals and sales? It does. You just swap the property summary in the prompt. For rentals, add rent, deposit, and move-in costs; for sales, add price, HOA dues, and reserve contributions to the fields.

Q. What if I want to standardize wording across the whole company? Put your office’s wording rules (always “south-facing,” for example) in CLAUDE.md and they’re enforced on every generation. You turn phrasing that used to live in one person’s head into a company asset.

What I found when I actually tried it

I ran the prompt and script above end-to-end myself, using fact-sheet data for three dummy resale condos.

First, all nine drafts (3 properties x 3 portals) came out, one generation each. What I had to fix on re-reading was 2-3 spots per draft. Mostly fine, compliance-flavored tweaks, like changing “5-minute walk” to “about a 5-minute walk.” Compared to writing from scratch, it felt like a quarter of the time.

I fed the check script a description where I’d deliberately mixed in “steal” and “limited time only.” As intended, both were detected and it returned exit code 1. It also prints the character count, so I could judge on the spot whether the Zillow version was running too long.

The biggest win was the inconsistent wording disappearing. I added a single line to CLAUDE.md, “use ‘south-facing’ for exposure,” and all three matched. It felt like a machine had taken over the accidents that used to happen with a tired brain at 5 PM. AI drafts, humans decide. Hold that line and your Saturday mornings go back to showings and follow-up.

If you want to roll this in as a company-wide system, with wording rules and a review process built out, I can design it with you through training and consulting. If you’d rather try it with your own hands first, start with the free learning materials.