An agent at a briefing asked a question that deserves a straight answer. “I heard HUD came out against AI. Does that mean I can’t use it to write my listings?”
No. And the guidance isn’t about listing copy at all. The longer answer is more useful than the short one, because the reason it still matters to you is not the reason most people assume.
What the guidance actually covers
In May 2024, HUD published two guidance documents on how the Fair Housing Act applies when housing decisions involve AI and other automated systems. One addresses tenant screening: the use of automated tools and third-party screening reports to evaluate rental applicants. The other addresses advertising: how housing ads get targeted and delivered on digital platforms.
Two things are worth holding onto. First, this is guidance on existing law, not a new rule. The Fair Housing Act did not change in 2024. Second, and this is the part that reaches past screening and ad targeting: using a vendor’s system does not move responsibility to the vendor. A housing provider or advertiser can be answerable for a discriminatory outcome produced by a tool it chose to use.
Why that reaches your listing text anyway
Listing copy sits under an older part of the same law: the Fair Housing Act’s long-standing prohibition on publishing statements that indicate a preference, limitation, or discrimination based on a protected characteristic. That has been enforced against newspaper ads, flyers, and MLS remarks for a very long time.
The connecting idea is the one the 2024 guidance states plainly about automated systems. The output is still your statement. If a chatbot writes “ideal for a young professional couple” and you publish it, you published it. And chatbots write exactly that, unprompted, because human-written real estate copy is full of it and that is what the model learned from.
A review workflow that takes 45 seconds
A small brokerage that had every agent drafting however they liked replaced that with 3 checks. Drafting time stayed at roughly 4 minutes. Review added about 45 seconds.
Scan for people words. Read the draft looking only for nouns and adjectives that describe a person or group rather than the property. Family, couple, retirees, professionals, singles, kids, bachelor, mature, active. Also religious landmarks used as directions, and characterizations of who the neighborhood is. Keep the list on one card and use the same list every time.
Check every fact against the source. The invented detail and the risky phrase are different problems and they need separate passes. Trying to catch both at once means catching neither.
One named person signs off. Not the tool, not the assistant, not the group chat. Write down who it is.
You can also push the first check upstream by telling the tool to describe only the physical property and its features, never who might live there. That reduces the problem. It does not remove the review step, because the instruction does not hold reliably across every draft.
This is the workflow we teach in the workshop, and it is worth saying plainly: it is a review habit, not legal advice. Your policy belongs to your broker of record and your own counsel, and they should see the list you use before you standardize on it.
The takeaway
Build the word list this week. Ten minutes, one card, the words your office actually types. Then read every AI-drafted description twice: once for the facts, once for the people. The tool didn’t create this risk. It just made it faster to produce.