Most articles about AI for real estate agents are a list of fifty prompts that produce listing descriptions. They almost never mention the thing that actually matters: you are legally responsible for what those descriptions say, and the model does not know that.
AI is genuinely good at real estate copy. It is fast, it is tireless, and it does not get bored on the eleventh listing of the week. It is also fluent in exactly the kind of warm, aspirational language that the Fair Housing Act prohibits — because that language is all over the training data, written by agents who also did not think about it.
This is a solvable problem, but it is solved in the prompt, not in the edit.
What follows is practical guidance, not legal advice. Fair Housing rules and state AI-disclosure laws vary and change — your broker and your own counsel are the authority on what applies to you.
The phrases AI reaches for that you cannot use
The Fair Housing Act prohibits expressing a preference, limitation, or discrimination based on race, color, religion, sex, disability, familial status, or national origin. That sounds easy to comply with until you see what it actually rules out in ordinary listing language:
- "Perfect for families" or "great for a growing family" — familial status
- "Ideal for young professionals" — potentially age and familial status
- "Walking distance to St. Mary's" — religious preference, even as a genuine landmark
- "Safe neighborhood" or "quiet area" — long treated as coded language about neighborhood demographics
- Any description of who lives in the surrounding community
Ask an AI to write "a warm, inviting listing description for a three-bedroom colonial" and there is a good chance you get at least one of these back. Not because the model is careless — because that is what warm, inviting listing copy has historically sounded like.
Put the constraint in the prompt, not the edit
The reliable fix is telling the model the rule up front and giving it the substitution to make. Describe the property, never the buyer.
Write a listing description for the property details below.
Hard rules:
— Describe the PROPERTY only. Never describe or imply who would live there, who it suits, or who the neighborhood is for.
— Do not use: family, families, kids, children, couples, singles, professionals, retirees, safe, quiet, exclusive, walkable to any named religious institution, or any reference to the demographics of the area.
— Convert any buyer-focused framing into a feature statement. "Great for families" becomes "fenced rear yard, four bedrooms."
— Amenities and distances may be named factually (schools, transit, parks) without characterizing who uses them.
Property details: [PASTE YOUR DETAILS]
After the description, list any phrase you considered and rejected under these rules, and why.
That final line is the part worth keeping. It turns the model into something that surfaces its own near-misses instead of quietly smoothing them over — the same self-checking idea we covered in how to stop AI from being confidently wrong. Over a few listings you learn which phrases keep coming up, which tells you what to watch for in your own writing too.
Run a compliance pass on copy you already have
Worth doing on your existing listings, not just new ones:
Review this listing description for language that could be read as expressing a preference based on race, color, religion, sex, disability, familial status, or national origin under the Fair Housing Act. Flag each phrase, explain how it could be interpreted, and offer a property-focused replacement. If nothing is flagged, say so plainly rather than inventing issues.
[PASTE LISTING]
The last sentence matters. Without it, a model asked to find problems will find problems, whether or not they exist.
A model is not a compliance tool
Be clear about what the above does and does not buy you. A well-constrained prompt catches most of the obvious language and is far better than an unconstrained one. It is not a review, it is not a guarantee, and it does not shift responsibility anywhere.
Every listing you publish is yours, legally, regardless of what wrote the first draft. Read it before it goes live. If your brokerage has a compliance review process, AI-drafted copy goes through it exactly like anything else.
Images are a separate question, and the law is moving
Written copy and altered photography are governed differently, and photography is where disclosure rules are actually arriving. California AB 723, effective January 1, 2026, requires disclosure of AI-altered listing images along with a link to the original unaltered photo.
As of 2026 there is no general legal requirement to disclose that AI helped write marketing text — but this is a fast-moving area, requirements vary by state and by MLS, and your MLS may impose rules stricter than the law. Check yours rather than assuming.
Where AI actually saves a working agent time
Listing descriptions get the attention, but they are not where the hours go. The bigger wins are the repetitive written work around the transaction:
- Turning a CMA into a client-readable summary
- Drafting the follow-up you keep meaning to send to a past client
- Neighborhood guides built from factual amenities — again, describing the place, not the people
- Converting one open house into a week of social posts
- First drafts of the same explanation you have given fifty buyers about inspection contingencies
None of that is glamorous. All of it is time you are currently spending.
Where to start
Try the approach first — our free 10-prompt sampler covers marketing, sales and content, enough to see whether structured prompting fits your workflow.
For the full set built around an agent's actual week, The Real Estate Agent's Prompt Pack is 48 fill-in-the-blank prompts spanning listings, lead generation, client communication and social content. If listing and social copy is the specific bottleneck, 25 Viral Content Hooks for Creators is a $7 mini pack focused just on openers.
These are starting frameworks you adapt to your market, your brokerage's rules and your own voice. They remove the blank page and a great deal of retyping. They are not a compliance product, they do not review your listings, and they do not guarantee any result — the review step stays yours.