The Listing Description Problem: Why AI Copy Sounds Like AI Copy, and How to Fix It
You can spot it in about four seconds. Nestled. Boasts. Stunning. A chef’s kitchen in a house with a standard range. Every room described as spacious, and nothing in the whole listing that could only be said about this property.
The short answer
AI listing copy sounds generic because it was given generic input. A model with no listing data and no local knowledge falls back on the average of every listing description ever written, which is exactly the copy everyone recognizes as AI. The fix is not a better prompt or a better model. It is feeding the tool the specifics only you and the MLS have.
Three reasons the copy comes out flat
1. It does not have the property
Ask a general AI tool to write a listing description and it will politely produce one from whatever you typed. Four bedrooms, three baths, updated kitchen. From that, there is nothing to work with but adjectives. The result reads like a template because functionally it is one.
The listing itself contains dozens of details that would make the copy specific: lot orientation, year of the roof, school assignment, HOA terms, the sunroom addition, the fact that it backs to green space rather than to another house. None of that reaches the model unless something puts it there.
2. It does not know the market
Good listing copy answers a local question. Whether a finished basement matters, whether a two-car garage is a differentiator or table stakes, whether buyers in this price band care more about the kitchen or the yard. A general model has no idea. It writes for a national median that exists nowhere.
3. It was trained on the median listing
Language models produce likely text. The most likely listing description is the most common one, and the most common one is the tired one. Without something pulling it toward the specific, the copy drifts to the middle by default. That is not a flaw you prompt your way out of. It is what the tool does when it has nothing better to go on.
The fair housing problem nobody mentions
This is the part worth slowing down for, because it is the one that carries actual risk.
AI writes toward warmth. Warmth in listing copy tends to arrive as language about who a home is for, and language about who a home is for is exactly what the Fair Housing Act restricts. Descriptions must speak to the property, not to the buyer, and must not indicate a preference or limitation based on race, color, national origin, religion, sex, familial status, or disability.
Phrases that AI produces readily and that agents should catch on every draft:
- Perfect for families, family-friendly, great for young couples, ideal for empty nesters
- Safe neighborhood, quiet area, good part of town
- Walking distance to any house of worship named specifically
- Exclusive, private, prestigious community, when used to describe who lives there rather than an amenity
- Master bedroom, which many brokerages and MLSs have moved away from in favor of primary bedroom
- Any reference to school quality rather than school assignment
An AI tool with real estate context should flag these. A general tool will write them cheerfully, because in ordinary English they read as friendly. Whatever tool you use, the draft is yours the moment you publish it, so read every one against your brokerage and MLS guidance. This is general guidance rather than legal advice, and your broker is the authority on your market.
What good input actually looks like
| Generic input | What the copy becomes | Specific input | What the copy becomes |
|---|---|---|---|
| Updated kitchen | A stunning chef’s kitchen perfect for entertaining | 2023 renovation, quartz counters, gas range, original 1940s built-in pantry kept | A 2023 kitchen renovation that kept the original built-in pantry, paired with a gas range and quartz counters |
| Large yard | A spacious backyard oasis | 0.34 acres, backs to a protected greenbelt, mature oaks, no rear neighbor | A third of an acre backing to protected greenbelt, so the tree line behind the house stays a tree line |
| Great location | Nestled in a highly desirable neighborhood | Four blocks from the town square, on the quiet side of the through street | Four blocks from the square, set on the quieter side of the street |
| Well maintained | This home has been lovingly cared for | Roof 2021, HVAC 2022, water heater 2024, one owner since 2009 | Roof, HVAC, and water heater all replaced since 2021, under one owner since 2009 |
Notice what changed. Not the writing quality. The information density. Every line in the right column contains something a buyer could not have guessed, and none of it required a better model.
A five-step process that produces copy worth publishing
- Start from the listing record, not from a blank prompt. If your AI tool connects to the MLS, the property details are already there. If it does not, paste the full listing sheet rather than a summary.
- Add the three things the data does not carry: why the sellers bought it, what they will miss, and the one detail you noticed walking through that is not in any field.
- Ask for two versions in different structures. One that leads with the property, one that leads with the location. You will know which is right for this listing when you see them side by side.
- Cut every adjective that could apply to any house. Stunning, gorgeous, spacious, charming, and immaculate carry no information. If removing a word changes nothing, it was doing nothing.
- Run the fair housing pass as a separate read, not as part of the edit. Looking for one thing at a time is how you actually catch it.
Steps one and three are where AI saves the time. Steps two, four, and five are where you make it yours, and together they take a few minutes rather than the half hour a description costs from scratch.
Where REX fits
REX is built around real estate work rather than pointed at it, so listing copy starts with real estate context already in place. With MLS Connectivity enabled, it works from the listing record itself, which removes the single biggest cause of generic output: a model writing from four facts you happened to type.
It is the same pattern that separates useful AI from impressive-sounding AI everywhere in this job. We wrote about it in the context of pricing work in Can ChatGPT Do a CMA, and the principle transfers exactly: a tool connected to your data does the work, a tool without one writes about it.
Frequently Asked Questions
Why does AI listing copy sound generic?
Because it is working from generic input. A language model produces likely text, and the most likely listing description is the most common one. Given only a few details typed into a prompt, it has nothing to write with but adjectives, so it defaults to the average of every listing description in its training. Feeding it the full listing record and local specifics is what changes the output, not a better prompt.
Can AI write listing descriptions that pass fair housing review?
A draft still needs a human read. AI writes toward warmth, and warmth in listing copy often arrives as language about who a home suits, which is the category the Fair Housing Act restricts. Watch for phrases like perfect for families, safe neighborhood, and proximity to a specific house of worship. Describe the property, never the buyer, and follow your brokerage and MLS guidance.
Should listing descriptions be written by AI at all?
AI is well suited to the drafting and structuring work, which is most of the time cost. It is not suited to deciding what matters about a specific house, which is the part that requires having stood in it. Agents who use AI for the first draft and their own judgment for the specifics get the time savings without the generic result.
What details should I give an AI tool before it writes a listing description?
The full listing record rather than a summary, plus three things the data does not carry: why the sellers bought the house, what they will miss about it, and the detail you noticed in person that is not in any field. Those three are usually what makes a description sound like it was written by someone who has been inside.
Does an AI tool need MLS access to write listing copy?
It is not required, but it removes the biggest cause of generic output. Without a data connection, the copy is limited to whatever you paste in, so quality depends on how thorough you were that day. With a connection, the listing details are already there on every property, every time. REX offers MLS Connectivity as an optional add-on.
The takeaway
Nobody can tell AI copy from human copy when the copy contains information. They can always tell when it does not. The generic listing description is not evidence that AI cannot write, it is evidence that it was asked to write about a house it knew four things about.
Start a 30-day free trial of REX, or talk with our team about listing marketing for your office.
