Answer · Real estate
Can AI write a listing description without a fair housing problem?
Yes, with a review that catches the language nobody intends: neighbourhood proxies, family framing, and who the copy pictures living there.
Yes. Advertising rules apply to the words, not to the author, so the copy needs the same review it always did. Generated text fails through proxies — school and neighbourhood framing, family fit — rather than through explicit preference.
The rule is about the statement, not the author. It is unlawful to make, print or publish an advertisement for a dwelling that indicates a preference, limitation or discrimination based on a protected characteristic, and the statute reaches the person who publishes it. Nothing in that turns on how the sentence was produced, which means an agent who publishes generated copy has made the statement personally and the tool is not a party to it.
The federal regulator confirmed in 2024 that these rules apply where algorithms and machine learning are used in housing advertising and in tenant screening, which removed whatever ambiguity there was about automated systems occupying a different category. The practical reading for a brokerage is that outsourcing the copy to a system is legally the same as outsourcing it to a marketing assistant, and is supervised the same way.
The failure mode is specific and it is not the one training courses cover. Explicit preference does not appear in generated copy; models are heavily conditioned against it. What appears is proxy language, and it appears because the model is optimising for appeal: a description of the neighbourhood's character, an emphasis on school ratings, a picture of who the home suits, phrases about a quiet street or a safe area, references to nearby places of worship, and adjectives about the community rather than the property. Each is individually defensible and collectively they describe a resident rather than a dwelling.
Which suggests a review test that is quicker than a prohibited-words list and catches more. Read the description and ask whether it describes the property or the buyer. A listing that says four bedrooms, a south-facing garden, a rebuilt roof in a stated year and a parking space describes a property. A listing that says perfect for a growing family in a friendly community close to excellent schools describes a household, and it is the second kind that generated copy produces by default because that is what persuasive property writing looks like in the material it learned from.
Accessibility descriptions are the case where the same instinct produces the opposite error. A ground-floor unit with a step-free entrance and a wide doorway is a factual description of a dwelling and belongs in the copy. Saying it is ideal for someone with limited mobility describes the occupant. The first is useful to buyers who need it and is the version the rules encourage; the second is the same information framed as a preference about who should live there.
The supervision point is where brokerages get caught rather than individual agents. The advertising obligations attach to the brokerage as well as to the licensee, state licensing rules impose their own advertising requirements on top of the federal ones, and a tool deployed at office level generating hundreds of descriptions is a single point that can produce a consistent pattern across every listing. A consistent pattern is considerably worse than an isolated bad description, because it is evidence of a practice.
Nobody writes an unlawful listing on purpose any more; they write a description of who would be happy living there, which is the same statement with better manners.
Siddharth Sharma, Context Theory
Related questions
Is a prohibited-words filter enough?
It catches the copy nobody was going to write anyway. Word lists were built for a period when the problem was explicit language, and generated text almost never contains it. The proxies that do appear are ordinary marketing words in ordinary combinations, which means no list can catch them without also flagging legitimate descriptions. The property-or-buyer question is what a filter cannot do and a reader can, in about ten seconds.
What about the photographs and the targeting?
Both are advertising and both are covered. Who an advertisement is shown to has been the subject of enforcement independently of what it says, and a generated description on a narrowly targeted campaign combines the two problems. If the brokerage is choosing audiences by anything correlated with a protected characteristic, the copy review has been addressing the smaller half of the exposure.
METHOD
Every figure below carries its source and the date it was verified. Nothing on this page is asserted.
The numbers on this page.
| What | Value | Specific to |
|---|---|---|
| Real estate — largest YoY CPC increase of any tracked industry | +27.27% | Category-wide |
| Average agent inbound response time | 15+ hours | Category-wide |
LocaliQ / WordStream Search Advertising Benchmarks 2026 · Google + Microsoft Ads, 20 industries · Apr 2025–Mar 2026 · verified
2026 real estate lead-response benchmark · hours · verified
What is specific to this page.
| Kind | Claim | Check it against |
|---|---|---|
| Regulation | It is unlawful to make, print or publish an advertisement for a dwelling indicating a preference, limitation or discrimination based on a protected characteristic, and liability attaches to the person publishing the statement rather than to whatever produced the words. | The advertising prohibition in the Fair Housing Act, and the brokerage's own advertising policy for licensee-published copy. |
| Constraint | The federal housing regulator issued guidance in 2024 confirming that the advertising and tenant screening provisions apply where algorithms and machine learning perform those functions, removing any argument that automated systems sit in a separate category. | The Department of Housing and Urban Development guidance on the Fair Housing Act's application to advertising and tenant screening, issued in 2024. |
| Workflow | Generated listing copy fails through proxy language rather than explicit preference — neighbourhood character, school emphasis, family fit, community adjectives — because persuasive property writing in the training material describes the resident rather than the dwelling. | Reading a sample of generated descriptions and marking each sentence as describing the property or describing the occupant. |
| Licensing | Advertising obligations attach to the brokerage as well as the individual licensee, so an office-level tool producing a consistent pattern across every listing creates evidence of a practice rather than an isolated defect. | The advertising and supervision provisions of the state real estate licensing rules the brokerage operates under. |
Each row would be wrong on another industry's page. Where a sourced figure exists it is in the table above instead; these are the constraints that shape the work and do not happen to be numbers.
Start with the measurement.
Reading about a benchmark is not the same as knowing your own number. The audit produces yours, measured rather than estimated.
$497 · delivered in 5 business days · credited against month one