AI can draft a listing description, a Facebook post, or a lead-nurture email in seconds. It can also slip in a phrase that tells some buyers or renters they are not welcome. Here is why that happens, what the law says about advertising, and how to catch it before anything goes live.
A language model writes by predicting what text usually comes next, based on the enormous amount of writing it learned from. Decades of real estate marketing are part of that writing, including ads that described who a home was "perfect for" rather than what the home actually offers. So when you ask for "warm, inviting copy," the model may reach for the same patterns: a neighborhood that is "ideal for young families," a condo "perfect for singles," a community that is "quiet and exclusive."
The model is not trying to discriminate, and it does not know Fair Housing law unless you tell it to follow it. But intent is not the test for an ad. What matters is what the words communicate to a reader. You publish the copy, so you own it.
The federal Fair Housing Act makes it unlawful to make, print, or publish any notice, statement, or advertisement about the sale or rental of a dwelling that indicates a preference, limitation, or discrimination based on a protected class (42 U.S.C. § 3604(c)). The federal protected classes are race, color, religion, sex, disability (the statute says "handicap"), familial status, and national origin.
HUD's regulation at 24 CFR § 100.75 spells out what that covers: words, phrases, photographs, illustrations, or symbols that suggest a home is or is not available to a particular group; oral as well as written statements; and choosing media or locations for ads in a way that denies particular groups information about housing.
The examples below are illustrative, written for this guide. They are not drawn from any real listing, and they are not a complete list of risky phrases. When in doubt, ask your managing broker.
| AI draft (describes people) | Revised (describes the property) | |
|---|---|---|
| Familial status | "Perfect for a young couple or empty nesters, not a place for kids." | "Two-bedroom condo with an open floor plan and low-maintenance finishes." |
| Disability | "Ideal for active, able-bodied buyers who love stairs." | "Three-level townhome; primary suite on the top floor; no elevator." |
| Religion | "Wonderful Christian neighborhood with great values." | "Quarter-mile from a public park and the Main Street shops." |
| Race / national origin | "Exclusive enclave, a safe area for the right kind of buyer." | "Cul-de-sac lot with mature trees and a fenced backyard." |
| Sex | "A bachelor pad any single guy would love." | "One-bedroom loft with 12-foot ceilings and a wet bar." |
Watch the subtler patterns too: descriptions of who lives in a neighborhood, claims that an area is "safe" or has "the right demographics," and AI-generated images that show only one kind of family. Those can signal a preference just as clearly as an explicit phrase.
Discrimination risk is not only in the words. It is also in who sees the ad. In June 2022 the U.S. Department of Justice sued Meta Platforms (formerly Facebook), alleging that its housing advertising system, including a lookalike-audience tool called "Special Ad Audience," relied in part on characteristics protected by the Fair Housing Act to decide which users received housing ads. Meta settled the case the same month, agreed to stop using that tool for housing ads, built a new "Variance Reduction System" meant to reduce demographic disparities in ad delivery, and paid a civil penalty of $115,054. A third-party reviewer has since issued compliance reports.
The lesson for agents: platform ad tools, AI audience builders, and "optimize for leads" settings can narrow who sees a listing in ways you never chose. Avoid targeting by ZIP code, interests, or demographics that act as stand-ins for a protected class. Keep records of how each campaign was targeted, and ask your brokerage what platform settings it approves.
In April 2024, HUD issued guidance explaining how the Fair Housing Act applies to housing ads delivered through digital platforms, including targeting and delivery driven by algorithms and AI. That guidance has since been withdrawn. HUD published a notice in the Federal Register on April 6, 2026, withdrawing it and seven other fair housing guidance documents, effective September 17, 2025.
AI chatbots on your website, texting assistants, and automated lead-nurture sequences talk to consumers for you, often with no one reviewing each reply. HUD's regulation at 24 CFR § 100.70 treats steering as unlawful: restricting a person's housing choices, discouraging them from viewing a home, or suggesting they would not be comfortable in a neighborhood, because of a protected class. That includes the protected characteristics of people who already live in an area. A chatbot that answers "Is this a good neighborhood for people like me?" with commentary about who lives there creates exactly that risk.
If you are unsure whether a phrase or targeting choice crosses a line, confirm with your managing broker, your state real estate regulator, or a real estate attorney before you publish. For team rules on which tools are approved, see tool setup; for what information should never go into an AI tool, see data sensitivity.