AI Wrote Your Listing Description. Fair Housing Still Holds You Responsible.
- Jerad Larkin
- 1 day ago
- 8 min read
An agent showed me a listing description last week that took about nine seconds to generate. It was clean. It was warm. It read better than anything she would have written herself at ten at night after a full day of showings.
It also said the home was perfect for a young family looking for a quiet street near great schools. Three phrases, three different protected classes, and the tool that wrote it carries none of the liability. She does.
Can an AI-written listing description violate Fair Housing law?
Yes. AI listing descriptions routinely generate steering language that violates the Fair Housing Act, and the Denver Metro agent who publishes it carries the liability, not the software.
As a Sales Executive with Chicago Title Colorado, I teach AI classes to Denver Metro real estate agents almost every week, and listing descriptions are the first thing agents ask for help with every single time. I am completely in favor of using AI for this. The time savings are real, and I have built a listing description workflow where you type an address and get everything back because I think agents should stop spending an hour on copy.
What almost nobody is doing is the review that has to happen after the copy comes out. General-purpose AI models were trained on decades of real estate marketing, including a mountain of copy written before anyone was paying attention to this. They will hand you polished, friendly, legally risky language and never flag a word of it.
What Does Fair Housing Law Actually Say About the Words in Your Listing?
The 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. That is the whole thing. NAR's fair housing guidance boils it down to one sentence agents should tape to their monitor: describe the property, not the people.
Two things about that rule catch agents off guard. First, there is no intent requirement. You do not have to mean anything by it. The question is what the advertisement indicates to a reasonable reader. Second, the rule attaches to publication. A model is a tool, the same as a keyboard. The Fair Housing Act applies to copy a model wrote exactly the way it applies to copy a person wrote, and the agent and the broker who put it in the MLS are the ones who published it.
Why Does AI Produce This Language in the First Place?
Three reasons, and understanding them makes the copy easier to fix.
The training data is old real estate copy. Millions of listing remarks scraped from an era when "great starter home for a growing family" was standard. The model learned that phrasing as what good real estate writing sounds like.
The model is optimizing for warmth. When you ask AI to make copy feel inviting, the fastest path it knows is to put a person in the house. Picture your family in this kitchen. Ideal for the buyer who wants to slow down. That instinct to name an occupant is exactly the instinct Fair Housing law prohibits.
The model has no idea where you are. It does not know that Colorado protects four categories federal law does not. HousingWire published a checklist for exactly this problem, and the core point is that the confident, finished tone of AI output is the trap. Copy that reads like it came from a compliance department did not come from one.
Which AI Phrases Actually Get Denver Agents in Trouble?
These are the categories I see come out of AI tools most often, with the fix next to each one.
Familial Status Language
"Perfect for families with children." "Great starter home for a growing family." "Walk the kids to top-rated schools." Cut the occupant and keep the feature. Four bedrooms upstairs. Fenced yard. Two blocks to the elementary school on Xavier. The buyer can do their own math about who fits.
Age Language
"Ideal for young professionals." "Perfect for empty nesters." "Great retirement home." Same fix. Say what the house has. Main floor primary suite. Ten-minute light rail commute to downtown Denver. Low-maintenance xeriscape.
Religion and National Origin Signals
"Walking distance to churches." "In the heart of the community." Naming a house of worship as a selling point signals a preference even when you did not mean one. If proximity matters, give distance to a geographic landmark instead. Half a mile to Sloan's Lake. Four blocks to the Highlands.
Disability Language
"Not suitable for wheelchairs." "Buyers must be able to handle stairs." Describing who cannot live somewhere is the clearest version of the violation. Describe the structure instead. Three-level floor plan. No bedroom on the main level. Step-down entry from the garage. Same information, no exclusion.
Neighborhood Character Words
"Quiet neighborhood." "Safe area." "Good part of town." "Exclusive enclave." These are the ones agents fight me on, and they are the most common AI output of the whole list. The problem is that they describe the people who live there rather than the property, and they have a long history of functioning as proxies. Replace them with facts you can point at. Cul-de-sac. No through traffic. HOA maintains the greenbelt.
What Protected Classes Does Colorado Add That Federal Law Does Not?
This is the part an AI model trained on national data will never catch for you. The Colorado Anti-Discrimination Act covers everything federal law covers and adds marital status, creed, ancestry, and source of income. The Colorado Civil Rights Division is where complaints get filed, and the protections reach sales, rentals, lending, insurance, and advertising.
Source of income is the one that trips people up most. Colorado banned housing discrimination based on source of income in 2020, which means vouchers, disability benefits, and other lawful income cannot be screened out. If you handle rentals alongside your sales business, "no vouchers" and "verifiable W-2 income only" are Colorado problems whether an agent typed them or a chatbot did.
Marital status is the quiet one. "Perfect for a married couple starting out" is a phrase AI produces constantly and it is a protected class in Colorado.
How Do You Run a Two-Minute Fair Housing Check on AI Copy?
This is the process I walk Denver Metro agents through in class. It takes less time than reading this section.
1. Hunt for people nouns. Read the copy once looking only for words that name a human being. Family, kids, professionals, retirees, couple, singles, seniors, students. Every one gets cut or converted into a feature of the house.
2. Delete character adjectives. Quiet, safe, exclusive, established, up-and-coming, family-friendly. If the word describes the vibe of the people rather than the geography of the property, it goes.
3. Convert proximity to distance. Anywhere the copy names an institution as a benefit, swap it for a measurable distance to a place. Miles and blocks are facts. Institutions are signals.
4. Check accessibility wording. Make sure every sentence about stairs, ramps, doorways, or layout describes the building and never the occupant.
5. Send it back through the model as an auditor. Paste your cleaned copy in and ask it to review the text against the federal Fair Housing Act and the Colorado Anti-Discrimination Act, including marital status, creed, ancestry, and source of income, and to flag anything that describes a person rather than the property. The second pass catches things you skimmed past. It is a second reader, not the final one. You are the final one.
6. Save your approved version. Once you have copy you are confident in, keep it. Building a saved AI project with your own approved examples and rules means the next listing starts from clean copy instead of starting from whatever the internet trained the model on.
Should Your Brokerage Have a Written AI Policy?
It should, and most do not yet. Florida Realtors made the case well in a piece on why every brokerage needs an AI use policy, and NAR publishes a policy template brokers can adapt. A workable policy covers four things: which tools are approved, who reviews AI-generated marketing before it publishes, how AI-generated images and video get disclosed, and what client information may never be pasted into a public model.
That third one matters more every month. If you are using AI video tools to market listings, generated imagery of a property carries its own disclosure duty, and consistently generating people of one age or background in your marketing creates the same advertising exposure the words do. It is also worth knowing who actually owns your listing content once it hits the MLS.
Compliance keeps showing up where agents least expect it. The same thing happened with database texting, where the TCPA rules most Denver agents get wrong quietly became a real exposure while everyone was focused on getting more leads.
Part of what I do as a Sales Executive at Chicago Title Colorado is help Denver Metro agents keep the marketing side and the transaction side from tripping over each other. A listing that draws a complaint is a listing that stops moving toward closing, and everything downstream of that, the title work, the escrow timeline, the seller's next purchase, sits still while it gets sorted out. I am not an attorney and this is not legal advice, so run specifics past your managing broker or counsel. But the two minutes of review is not a legal question. That is just doing the work.
Frequently Asked Questions
Can I get in trouble for a listing description ChatGPT wrote?
Yes. The Fair Housing Act attaches to whoever publishes the advertisement, and that is you and your broker. "The AI wrote it" is not a defense any more than "my assistant wrote it" would be. Review every piece of AI copy before it goes into the MLS or onto social.
Is "walking distance to top-rated schools" a Fair Housing violation in Colorado?
It is a familial status risk, and interpretation varies by jurisdiction and by how the rest of the copy reads. The safer version states a fact without an implied audience: "Three blocks from Steck Elementary." School ratings also change year to year, which makes the claim a truthfulness problem on top of a Fair Housing one.
What are the penalties for a Fair Housing violation on a listing?
Federal civil penalties start in the tens of thousands of dollars for a first offense and escalate for repeat violations, separate from any damages awarded in a private lawsuit. In Colorado you can also face a Colorado Civil Rights Division complaint and license discipline. The bigger practical cost for most agents is the time, the legal fees, and the reputation hit inside a referral-driven business.
Do I have to disclose that AI wrote my listing description?
Text copy generally does not carry a disclosure requirement the way altered or generated imagery does. AI-generated or materially altered photos and video are the ones to watch, since MLS rules and truthful advertising duties both apply. Check your MLS rules and your brokerage policy, because this is moving fast.
What is the safest way for a Denver real estate agent to use AI for listing copy?
Give the model facts, not adjectives. Feed it the square footage, the bed and bath count, the finishes, the lot, the mechanicals, and the actual distances to real places. Tell it up front to describe only the property and never the buyer. Then run the two-minute check anyway. The tool gets you a great first draft, and you stay the editor.
I teach this in my AI classes across the Denver Metro, and the pre-publish check is the part agents tell me they wish someone had shown them a year earlier. If you want the tools, the prompts, and the class schedule, everything lives at milehightitleguy.com. Reach out anytime and I will send you the listing copy workflow and the Fair Housing check as a one-pager you can keep next to your desk.
Jerad Larkin
Sales Executive | Chicago Title Colorado
milehightitleguy.com
Comments