Two different risks sit inside a property listing, and generic listing prompts handle neither. One is fair housing risk, where the copy describes who the home suits rather than what the home is. The other is misdescription risk, where the copy states something inaccurate about the property and the listing later becomes evidence. A real estate listing compliance AI prompt has to constrain both at once, and the reader still has to check the output against local rules.
Two risks that are not the same
Fair housing rules are jurisdiction-specific. In common regimes, the general principle is that advertising may not express a preference, limitation, or discrimination about who may occupy a dwelling on the basis of protected characteristics. That covers wording, and it also covers the softer version: describing the ideal buyer rather than the property. Regulations, protected classes, and permitted exceptions differ by country, state, and in some cases city, so the operative rule comes from your regulator and your supervising broker, not from this article.
Misdescription is a separate exposure and a more common one in practice. A listing that claims a floor area, a parking space, a utility connection, a permitted extension, or a basement that does not exist can create liability for the seller and the agent. The problem is rarely deliberate. It is the model adding plausible detail because a listing with fewer than twenty features reads as thin.
The prompts below attack the first risk in the writing step and the second in the verification step, in that order.
Generate a features-only description
Write a property listing description from the fact sheet below. Describe only the property. Do not describe the neighbourhood in terms of its residents, and do not describe the kind of person or household the home would suit. Fact sheet: [PASTE VERIFIED FACTS]. Format: [PORTAL FIELD LENGTHS]. Tone: [PROFESSIONAL, PLAIN, OR A HOUSE STYLE]. Forbidden: any phrase that expresses or implies a preference about who should live there, including "ideal for", "perfect for", "suited to", "would suit", or "great for families, couples, professionals, professionals with children, singles, or those with pets". Forbidden: claims about schools and their quality, neighbourhood safety, demographics, commute convenience to a named employer, or prevailing quiet, unless the fact sheet contains a specific verified source for that statement. Forbidden: any amenity, appliance, outlet, storage space, parking arrangement, floor area, or utility connection not present in the fact sheet. If a field needs a claim I have not supplied, write [CONFIRM WITH SELLER] rather than inventing one. Do not add a closing paragraph inviting enquiries.
Fill in these details
- Fact sheet: the single source of truth. Anything absent from it cannot appear in the copy.
- Field lengths: portal limits change, so check the current spec rather than assuming.
- House style: tone is a brokerage matter, and you set it here rather than accepting a default.
- The forbidden list: keep it verbatim. It is the part that survives copy edits.
The "perfect for" family of phrases is the one to watch. It is not that describing a property as having three bedrooms is wrong. It is that the phrase makes a statement about an occupant, and that statement is the type of preference the rule targets.
Adapt by adding to the forbidden list rather than trimming it. Every portal and brokerage has its own wording sensitivities and yours are worth writing down.
Screen the draft for preference language
Below is a property listing draft. Do not rewrite it yet. Screen it line by line and return a table: the exact phrase, the category ([OCCUPANT PREFERENCE], [NEIGHBOURHOOD CLAIM], [UNVERIFIED FACT], [FORMATTING]), why it is in that category, and a neutral rewrite that keeps the selling point and removes the problem. Draft: [PASTE] Flag any sentence whose value comes from describing the area or the people in it rather than the property. Flag every factual claim you cannot trace to the fact sheet you have been given. Do not decide whether any of this is legal. Do not remove a feature to make the copy shorter. If a flagged phrase carries most of the appeal of the listing, say so, so I can replace it with a property feature instead.
Run the screen before you argue with the copy. Listing reviewers catch this inconsistently, and a screen makes the pattern visible in one pass rather than across three rounds of edits.
Verify every factual claim against the fact sheet
Take this listing draft: [PASTE] and this fact sheet: [PASTE]. Produce a verification table with one row per factual claim in the listing: the claim as written, the source in the fact sheet, the status ([VERIFIED], [PARTIALLY VERIFIED], [NOT IN FACT SHEET]), and who should confirm it ([SELLER], [SURVEYOR], [SOLICITOR], [MANAGEMENT], [UTILITY]). Do not accept a claim because it sounds right. Flag every number, every dimension, every year, every appliance, every space, and every inclusion. Then list the claims that must be corrected or removed before publication, ordered by how much exposure they create. Do not supply the correct values. If the fact sheet conflicts with the listing, show both and mark the listing as wrong rather than resolving it yourself.
Verification is the step that catches the misdescription risk, and it has to be explicit. Ask a model to "check the accuracy" and you will get a general impression. Ask for a row per claim with a status and an owner, and you get something a supervising broker can actually sign off.
Numbers are the priority. Floor areas, plot dimensions, parking, storage, appliance inclusion, tenure, and permissions are the claims that generate disputes.
Worked example: a draft that had to be rewritten twice
An agent asked for a three-bedroom terrace listing. The first draft from a prompt is shown below with the screen's findings.
| Draft phrase | Category | Neutral rewrite |
|---|---|---|
| Perfect for a growing family | Occupant preference | Three bedrooms over two floors, with a rear garden |
| In a very safe area | Neighbourhood claim | Removed, no verified source in the fact sheet |
| Close to the primary schools | Neighbourhood claim | Walking distance to [named road] with [named route] |
| Large open-plan kitchen diner | Unverified claim | Kitchen with [units] units, opening to a [size] dining space |
| Garage and off-road parking | Unverified fact | Confirmed with the seller; written as [CONFIRM WITH SELLER] until then |
The rewrite kept every selling point that belonged to the property and dropped two sentences that belonged to a viewer's imagination. The verification table then caught a floor area that did not match the fact sheet, which was corrected before publication.
Review the listing before it goes live
1. Check the fact sheet against the property file
Compare the sheet you gave the model with your own records. If the sheet has a gap, the listing has a gap, and the verification table cannot see what it was never told.
2. Read every adjective for an implied occupant
Words like ideal, perfect, cosy, quiet, safe and vibrant carry a second meaning about who should be there. Replace them with a measurable feature wherever the measurement exists.
3. Route every number to a named person
Floor areas, years built, parking, storage, appliance inclusion, and any reference to permissions go to the surveyor, the solicitor, or the seller. Nobody signs off a listing on the strength of a generated paragraph.
4. Apply your own compliance policy last
Whatever your regulator, your brokerage and your supervising broker require goes on top of anything here. Confirm current rules directly rather than assuming this page reflects them, and keep the professional judgement where it belongs.
Common listing prompt mistakes
- Generating copy without a fact sheet. The model fills the gaps, which is exactly how an inaccurate amenity reaches the portal.
- Keeping "ideal for" because it sells. It is the shortest route to a complaint, and the selling point usually survives as a feature.
- Treating neighbourhood description as neutral. School quality, safety, and quiet are claims about an area, and they carry obligations of their own.
- Asking the model to decide legality. It cannot. It can flag and rewrite, and your compliance function decides.
- Publishing without a verification table. The screen catches wording, not facts, and the facts are the expensive mistakes.
Frequently asked questions
Can an AI property description prompt guarantee a compliant listing?
No. A prompt can produce copy restricted to verified features and flag language for review, which removes a large share of avoidable risk. Compliance depends on the facts, the jurisdiction, and the review your supervising broker requires, none of which a prompt can verify.
Is describing a house as suitable for families fair housing risk?
It is the phrasing people argue about most. Rules vary by jurisdiction, and some regimes permit or restrict family-related references in specific ways. Write feature-only copy, which is safe under most regimes and is good practice regardless, and confirm the detail with your compliance function.
What should I do if a listing is already live with an inaccurate claim?
Correct it promptly through the same channel you published on, and keep a record of what changed and when. Whether the correction is enough or a withdrawal is required depends on the facts and your local rules, so raise it with your supervising broker rather than deciding alone.
Should I disclose every defect in the description?
Your description should be accurate, and legal obligations around disclosure vary by jurisdiction and by what the buyer is entitled to know. Ask your solicitor or compliance function what applies, and use the listing description to describe the property rather than to make legal statements about it.
Run the screen on your next draft before your reviewer does
Build the fact sheet first, generate from it, then run the screen and the verification table on the result. For more prompts and reference material on what your market requires, browse the real estate law prompt collection.