How to Write Accurate Listing Descriptions With ChatGPT

Every listing your brokerage takes needs public remarks, written by one of your agents and sent under your name to every site your MLS feeds.

A claim the house cannot back is a misrepresentation; a word about who should live there is a fair-housing problem. California's real estate regulator said this winter who answers when an AI tool writes either: the licensee and their responsible broker, not the technology provider.

By hand, the agent writes from memory or edits the last similar listing. In NAR's 2025 technology survey, 46% of Realtors said they use AI-generated content, listing descriptions among it, and ChatGPT writes the remarks in seconds.

We tested it on one house we made up for this page, 2318 N Wrenfield Ct in Boise (there is no Wrenfield Court in Boise), and sent ChatGPT the same listing three ways on November 1, 2026: typed from memory, with a fact sheet pasted in, and with the fact sheet plus five rules.

ChatGPT invented no number.

It filled every gap the prompt left with a plausible sentence about the home's condition, its street or its buyer.

That stops when it writes from a fact sheet and checks its draft against the same sheet. It is one house on one day: read the counts as what to look for, not as rates.

Run the five steps below on your next listing. After them comes the AI agent Qvery builds to do the same on every listing, on autopilot.

This writes and checks the public remarks. It does not measure the house, replace the seller's disclosure, or stand in for your broker's fair-housing review.

Step 1: Write the Facts Down Before You Open ChatGPT

List what you know about the home, one fact per line, with the source beside any fact a buyer could dispute. The sheet does two jobs: ChatGPT writes from it, and you check the draft against it.

Here is the sheet for the invented house, exactly as we pasted it. Three kinds of line carry most of the work:

1. The finished area, with its source.

   Fannie Mae has appraisers measure with the ANSI standard, which counts only area suitable for year-round use, so the unheated sunroom gets its own line.

2. The seller's word, marked as such.

   "Seller says roof about 2014, no receipt" keeps the roof's age out until someone shows a receipt.

3. What conveys and what does not.

   The washer and dryer stay with the seller, and the sheet says so.

Schools, distances and the neighborhood go on the sheet only with a source beside them. Whatever the sheet leaves out is a gap, and Step 2 shows what ChatGPT puts in one.

Step 2: Paste the Sheet Under Five Rules

Typed from memory, the prompt asked for about 150 words and named the address, the beds and baths, the year, an updated kitchen, a fenced backyard and a two-car garage. Across three drafts, ChatGPT filled the rest:

Asked to highlight the neighborhood and the schools too, it searched the web and named no school. It still wrote "convenient access to local shopping, dining, parks", placed the house in the Boise School District and began a sentence "Families will appreciate".

None of it is a number, so none of it trips an agent who checks only numbers.

Start a new chat, paste this prompt, and paste your fact sheet over the last line:

Write the public remarks for this MLS listing from the fact sheet below.
Rules:
1. Use only facts written in the fact sheet. Do not add any feature, material, measurement, school, distance, view or neighborhood detail that is not there.
2. Leave out anything the sheet marks "seller says".
3. Describe the home, never the kind of person who should live in it.
4. Do not use: stunning, nestled, boasts, oasis, dream home, must-see, charming.
5. About 150 words, plain sentences, no exclamation marks.
After the remarks, under the heading MISSING, list what a buyer would expect to read that the sheet does not say.

<paste your fact sheet here>

Change rule 4 to your brokerage's own list of tired words and rule 5 to your MLS's length limit. The first draft under these rules began:

Here is the same house under each prompt, three drafts apiece:

The fact sheet did most of the work; the rules matter most once the seller has been talking (Step 4).

Every draft under the rules also ended with a list headed MISSING: roof documentation, laundry location, utilities, schools.

Don't ask ChatGPT to fill it. Each line is a question for the seller.

Step 3: Check the Draft Against the Same Sheet

Open a new chat and paste the check prompt, then the sheet, then the draft:

Check these listing remarks against the fact sheet.
1. List every factual claim in the remarks, one per line.
2. Mark each FOUND (quote the sheet line) or NOT IN SHEET.
3. List any words about who should live in the home rather than the home itself.
Do not rewrite the remarks.

<paste your fact sheet here>

REMARKS:
<paste the draft here>

A good check reads like a ledger: every claim on its own line, FOUND with the sheet's words beside it, or NOT IN SHEET. Part of what came back for one typed-from-memory draft:

Read the third line twice.

In both checks where it came up, ChatGPT marked "fully fenced" FOUND because the sheet says fenced.

A word that strengthens the sheet's fact reads to it as the fact, so reopen every FOUND line worded more strongly than the sheet.

Asked to check a draft with nothing to compare against, ChatGPT listed 20 statements to verify, the bedroom count and the year built among them: without the sheet, the check becomes a list of everything.

Step 4: Strike Every Word About Who Should Live There

The Fair Housing Act bars any ad for a home that indicates a preference based on race, color, religion, sex, handicap, familial status or national origin. HUD's word-by-word reading is a 1995 memo: describe the property, not the residents.

HUD has since taken the memo off its website without replacing it, and Oregon REALTORS advises treating such guidance as valid until HUD says otherwise.

Sellers talk about who they picture buying, so we added the kind of note an agent hears: "perfect for a young family or empty nesters, quiet street, short walk to St. Mark's Church."

From the sheet alone, both drafts wrote in the church, the walk to it and the quiet street. One turned the young family into "growing households" and "those looking to simplify", the same preference in other words.

Under the five rules, neither draft used any of the note.

State and local laws add protected classes, one MLS's rules allow only the property and its vicinity in the remarks, and a landmark is a judgment call. Your broker reads the flagged words against your rules before anything goes to the MLS.

Step 5: Save Both Prompts in One Shared Project

In ChatGPT, create a project for listing remarks. From its more-options menu choose Project settings and paste the five rules as instructions. Add a blank fact sheet as a file, and keep the check prompt in the project's first chat.

Project instructions apply only inside the project and override each agent's own custom instructions, so personal instructions stay out of the remarks.

Share it with your agents with chat access, so only people with edit access can change the rules. OpenAI's help center, which these steps follow, lists sharing on every plan, Free included.

Our real estate playbook carries this routine as one of its use cases, listing drafts checked against the record, beside the rest of the inbox work a property makes from listing to move-out.

How Qvery Builds the AI Agent That Writes and Checks Every Listing

We build the AI agent that does this for one brokerage, as each listing is entered, so nobody on your team fills in a sheet or runs a prompt. It does the job end to end:

• It reads the listing input where your team already keeps it.
• It drafts the remarks in your brokerage's voice, under your rules.
• It checks every claim against that input and holds back every word about buyers.
• It hands the listing agent the draft, the lines it could not trace and the questions for the seller.

The listing agent still sends the remarks to the MLS; the AI agent runs on its own, on autopilot, beside the people who own the listing.

We start by learning how your listings get written today: who fills in the input, which fields your MLS requires, which phrases your broker has already struck. A first working version runs within two weeks, and we train the agents and coordinators who read its drafts.

Every week it runs, we count the minutes each listing's remarks take from input to MLS and the claims it stopped before the MLS saw them.

Say your agents write 40 listings a month at 30 minutes each, an assumption to swap for your own: that is 20 hours a month, brought down to the minutes it takes to read a draft and its flagged lines.

Ask the Seller the Missing List at the Listing Appointment

The MISSING list is the most useful thing ChatGPT wrote. Take it to the next listing appointment and ask the seller each line: the sheet starts complete, and the remarks have nothing left to guess.

The same habit, facts first and a person on every line the AI cannot trace, fits most of the paperwork a brokerage sends out under its name.