Guide
How to write listing descriptions with ChatGPT or Claude, then check the wording
Draft listing remarks from facts in ChatGPT or Claude, then flag fair-housing wording before you post. Researched Oct 16, 2026. Not legal advice.
Listing remarks get posted where buyers, sellers, and fair-housing testers can read them. ChatGPT and Claude can draft from facts you supply. They can also invent a buyer, a neighbor, or a school claim you did not give them. This page is a researched way to keep the draft on the property, then run a wording pass before you paste it into the MLS. It is not legal advice, and it is not a hands-on test of a listing.
ChatGPT and Claude are on the hands-on list. This guide did not run the prompts below on a live listing, and it does not score either model. Subscription prices are on the AI tool pricing page, checked October 15, 2026. This guide does not re-check those prices. A separate price sort is Free and cheap AI tools for real estate agents.
The rule this check opened
24 CFR § 100.75, read on Cornell’s copy of the eCFR on October 16, 2026, makes it unlawful to make, print, or publish a notice, statement, or advertisement for the sale or rental of a dwelling that indicates a preference, limitation, or discrimination because of race, color, religion, sex, handicap, familial status, or national origin. The section covers written ads, including flyers, brochures, signs, and other documents used for that sale or rental. It also covers words, phrases, photographs, or symbols that convey the dwelling is available, or not available, to a group because of those characteristics.
That is a preference test, not a magic word list. HUD removed the old advertising guidelines in 24 CFR part 109 effective May 1, 1996. The Federal Register notice says part 109 was nonbinding guidance and was deleted from the Code of Federal Regulations. Section 100.75(d) still points at part 109. This page does not revive the removed word list as the current rule.
Two limits belong in the same paragraph. State and local law, and your MLS rules, can restrict more than the federal list. This page does not catalog those rules. The Fair Housing Act also has a housing-for-older-persons exception to the familial-status provisions. This page does not restate the current test for that exception. If the property is in a qualifying community, use that community’s approved language and your broker’s instructions, not a model sentence.
Write from the property, not the buyer
Collect only facts you can point to: beds, baths, square feet, lot, year built, verified updates and their years, parking, and distances you have measured or that the listing file already states. Leave out who you think should buy it, who lives nearby, school quality, crime, and “safe.” If a fact is not in your notes, the model does not get to add it.
Paste this into ChatGPT or Claude. Replace the brackets. Delete any line you cannot verify.
Write public MLS remarks for this listing. Use only the facts below. Do not add rooms, finishes, school names, crime, safety, or buyer types.
Beds: [BEDS]. Baths: [BATHS]. Living area: [SQ FT] sq ft. Lot: [LOT]. Year built: [YEAR]. City: [CITY]. Verified features: [FEATURE 1]. [FEATURE 2]. [FEATURE 3]. Updates, with year: [UPDATES]. Location, objective only: [MINUTES] to [TRAIL, FREEWAY, OR STORE]. No neighborhood demographics. Character limit: [LIMIT].
Lead with the strongest verified feature. Then layout, then outdoor space, then the objective location line. Describe the home, not an ideal buyer and not the neighbors. No “stunning,” “must-see,” “perfect for,” or “exclusive.” End on a fact. Then give me: a shorter version under [SHORTER LIMIT] characters that adds no facts, eight bullets with one fact each, and a “Facts used” list. If a sentence is not in my notes, omit it.
Read the “Facts used” list against your notes before you read the prose. If the model added a view, a renovation, or a school, delete that sentence. The character limit is the MLS’s, not a number this page invented.
Then run a wording pass
The second prompt does not clear the listing. It shows you sentences to reread. Paste the draft, not the whole chat.
Scan this listing copy for fair-housing problems. The standard I am using is 24 CFR 100.75: do not indicate a preference, limitation, or discrimination because of race, color, religion, sex, handicap, familial status, or national origin. Also flag wording that describes who should live there, the neighbors, schools as a selling point, crime, or safety. For each flag give the quote, why it may indicate a preference, and a rewrite that stays on the property and adds no facts. If my state or city protects more classes, I will check that myself: [LIST ANY YOU ALREADY KNOW, OR WRITE “not supplied”]. If nothing in the draft describes people or those claims, say PASS and quote the sentences you reviewed. Do not tell me the listing is approved.
You post only after you agree with the rewrite. If the draft describes a family, a religion, a disability, or a national origin, either as a welcome or as a limit, take it out. “No kids,” “perfect for a young couple,” and “exclusive community” are the kind of preference the regulation describes. They are examples of what to reread, not a holding from a court and not a complete list.
Housing ads on paid platforms have their own targeting rules. This page did not open those platform policies. If the same sentences are going into a paid ad, check that platform’s housing-ad category before you boost the post.
What this page does not claim
It does not say ChatGPT writes better remarks than Claude, or the reverse. It does not say a PASS from the model is a legal clearance. It does not say this site posted a listing with these prompts. Hands-on reviews stay unpublished until the notes are written. The method is on How we test.
Questions
Is this legal advice?
No. It is a writing workflow plus the federal advertising rule this check actually opened. Your broker, your MLS, and your state or city can be stricter. A model draft is not a compliance approval.
Did this site run the prompts on a real listing?
No. The prompts below were not submitted to ChatGPT or Claude for this page, and no listing was posted from them. ChatGPT and Claude are on the hands-on list for other work. This page is researched, not a scored test.
Does ChatGPT or Claude have a fair-housing checker?
Not one this page found as a product feature. The second prompt asks the model to flag people-focused wording. You still read the draft against the rule and your MLS before you post.
Is the old HUD advertising word list still the regulation?
No. HUD removed 24 CFR part 109 effective May 1, 1996. The rule still on the books, 24 CFR 100.75, still mentions part 109. This page does not treat a copied word list from the removed part as current law.
Related on this site
Free
Get the free AI prompt pack
You get the 35-prompt pack now, plus a weekly tools roundup. Free. Unsubscribe anytime.