If you write listings, you have already pasted a feature sheet into ChatGPT and watched it spit out a polished description in ten seconds. The speed is real. The problem is that the same tool that saves you ten minutes will, without blinking, write a sentence that violates the Fair Housing Act or states a fact you cannot back up. The line between the useful part and the part that gets you sued is the whole game. This is a plain look at where AI genuinely helps with listing copy, where it does not, and a set of prompts you can paste in and test on your own real listings.
The honest picture
AI is genuinely useful for the writing, the part that is slow and repetitive and that you do the same way every time:
- What AI does well today: turn your feature list into a clean MLS description, rewrite that description into the short forms you need (a headline, a portal blurb, a social caption, a buyer-email teaser), translate dry specs into plain benefit language, and vary the tone. It compresses the part of listing work that is just typing.
- What AI does not do: decide what is fair and true to say. It does not know which phrases steer, and it does not know your facts. It will cheerfully write "perfect for a young family," a textbook Fair Housing problem, and it will state a square footage, a school rating, or a "quiet, safe street" with total confidence whether or not it is true. You sign the listing agreement. You carry the legal exposure. AI drafts the words; the licensed agent owns every public line and the truth of every figure.
The right way to think about it: AI is a fast first-draft writer that you edit and approve, not an autopilot you trust. The speed is in the draft. The judgment stays with you.
The two lines AI will cross on its own
Two specific failures show up in almost every AI-written listing, and both are on you, not the model:
- Fair Housing steering. The federal Fair Housing Act bars advertising that signals a preference based on race, color, religion, national origin, sex, familial status, or disability. AI does not know this and writes to it constantly: "family-friendly," "perfect for empty nesters," "walk to church," "exclusive community," "great for professionals." These describe a buyer, not a property, and that is the trap. The fix is simple to state and easy to forget: describe the house, never the kind of person who should live in it.
- Facts it cannot verify. A language model has no idea whether the home is 2,400 square feet, whether the schools are A-rated, whether it is in a flood zone, or whether the HOA allows fences. The same goes for soft claims like "safe," "quiet," or "move-in ready," which sound factual and are not. It will assert all of it anyway. In a listing, a confident wrong number is misrepresentation, and it is your name on the agreement.
The setup that keeps you out of trouble
You can make AI much safer with two habits, and the prompts below build them in:
- Feed it only verified facts. Give it your spec sheet and tell it to use nothing else. A model that is told "use only these facts, do not invent" fabricates far less than one asked to "write a great listing for a 3-bed in Ponte Vedra."
- Make it check against a source, not its memory. When you want a Fair Housing review, paste in your brokerage's advertising policy or the relevant fair-housing guidance and tell the model to reason against that text and quote it. Treat the output as "here is what I flagged, you verify," never as a ruling. You are still the one who signs.
How to test it on your own work
Do not trust a polished demo, including this one. Pull two or three of your own recent listings, a timer, and the prompts below. For each, rate the output 1 to 5 on usefulness and accuracy, and compare the time against how you do it today. Keep the prompts that win and drop the ones that do not.
Paste-ready prompts
Copy these as written. Bracketed text is what you swap per listing.
Test 1: Draft the description from your facts (text model)
I am giving you the verified facts for one property. Write an MLS listing
description of about 150 words from ONLY these facts. Rules:
- Use only the facts I provide. Do not add, infer, or estimate any feature,
measurement, school, distance, or detail I did not give you. If something is
missing, leave it out. Do not guess.
- Describe the property, not the buyer. Do not use any language that could
signal a preference about who lives here: nothing about families, children,
age, religion, race, national origin, sex, or disability, and no
"safe / good / exclusive / friendly" neighborhood characterizations.
- Write the features as specific benefits in plain language. No hype.
Facts: [paste your feature sheet / spec sheet here]
Watch for: did it stick to your facts, and did it still slip in a steering phrase or an invented detail? Note every one. That is the work the model cannot do for you.
Test 2: Spin up the short-form variants (text model)
Here is my approved, fact-checked listing description: [paste it]. Produce four
short variants from it and invent no new facts:
1. An MLS headline under 80 characters.
2. A portal summary of about 40 words.
3. A social caption of 3 to 5 short lines with 2 to 3 relevant hashtags.
4. A buyer-email teaser of 2 sentences.
Keep every claim consistent with the description above. Add no features.
Watch for: consistency across the four, and whether the casual formats (the caption especially) sneak steering language back in.
Test 3: Fair Housing and accuracy audit (text model, the strongest use)
Review this draft real estate listing for two kinds of risk. For each issue,
quote the exact phrase, name the problem, and explain it in one line.
1. Steering or preference language: any phrase that could signal a preference
for or against a protected class (race, color, religion, national origin,
sex, familial status, disability), including coded terms like
"family-friendly," "walk to church," "exclusive," or "perfect for [group]."
2. Unverifiable or absolute claims: any stated fact (square footage, schools,
flood history, HOA rules, "safe," "quiet," "move-in ready," "never floods")
that would need a source to stand behind.
Reason only against the rules I paste below. Do not rule from memory. If a
phrase is fine, do not flag it.
Rules: [paste your brokerage Fair Housing advertising policy or the relevant
fair-housing guidance]
Draft: [paste any listing, the AI's or your own]
Watch for: does it catch the coded terms, and does it flag the claims you cannot actually prove? Run it on a listing you already published and see what it finds.
Test 4: Rewrite to fix (text model)
Here is a listing draft and a list of flagged issues: [paste the draft and the
flags from Test 3]. Rewrite the description so every flagged phrase is fixed:
remove the steering language and either cut the unverifiable claim or rephrase
it as something I can stand behind. Change nothing else, add no new facts, and
keep it about the same length.
Watch for: did it fix only what was flagged without introducing a new problem? Read the result as a draft to approve, not a final you publish unread.
What success looks like, and where it could go
If your own testing shows real time savings, the next step is a small pilot: run your next week of listings through the prompts and measure it. If that holds up, the natural next step is a simple agent, running on your brokerage's own cloud, that you use in plain language. The most useful version reads your spec sheet and photos, drafts the description and every portal variant, and checks each line against the Fair Housing and disclosure rules your brokerage loads once, flagging steering language and unverifiable claims and pointing to the rule it relied on, before anything reaches an agent to sign off. It flags and cites; the agent still owns the listing.
The principle holds the whole way through: AI gives you a faster first draft and a second set of eyes. It does not give you a signature, and it does not carry your license. Keep that line clear and the rest is upside.
Want a straight answer for your brokerage?
I build practical AI and custom software for businesses, on Google Cloud. If you want a second set of eyes on how AI could fit your listing and marketing workflow, or on a tool you are considering buying, tell me what you are working with. No pitch, just a straight answer.