Describe the Property, Not the Person: Honest Limits for Realtors
AI will walk you into a Fair Housing violation without knowing it. It fabricates facts it states with confidence. It doesn't know your market. Four hard limits every agent must hold.
AI will write a listing description that sells the home and violates the Fair Housing Act in the same paragraph.
It will do this confidently, fluently, and without any indication that something is wrong. The AI is not being reckless. It is completing a pattern based on language it has seen millions of times. The problem is that the patterns in real estate copy — the phrases that sound warm, inviting, and helpful — include some that are legally prohibited. AI has no idea. You do.
This lesson is about the four limits you have to hold, because AI does not hold them for you.
The Fair Housing Trap
The trap is not that AI writes obviously discriminatory content. It rarely does. The trap is that AI writes content that sounds warm and helpful while describing who should live in a home rather than what the home offers.
"Perfect for a young family" — that is familial status language. It implies families with children belong here and others do not.
"Safe quiet neighborhood" — that is a phrase that has historically been used as a coded reference to racial composition. Fair Housing enforcement agencies recognize it.
"Great for professionals" — this implies the neighborhood is not for people who do not fit that category.
"Ideal for empty nesters" — familial status again, now pointing the other direction.
"Walking distance to houses of worship" — national origin and religion associations, depending on which houses of worship and the neighborhood context.
"Good school district" — this one is subtle. On its own it describes a feature. In context, it is sometimes used as a demographic proxy. Know your market and your legal context.
The AI will include phrases like these because they appear constantly in real estate copy. It cannot read the law. You can. Reread every AI-generated description before it reaches anyone with one question in your mind: does this sentence describe the property, or does it describe who belongs there?
Property: yes. Person: rewrite it.
AI Fabricates Facts
The second limit is less about law and more about liability.
AI will state square footage, school district, HOA fees, recent sale prices, lot size, year built, and days on market as if it knows them. Often it does not. It is completing a pattern. The numbers sound plausible, fit the context, and are wrong.
A listing description that contains an invented square footage goes out under your name. A client who relies on it and discovers the error is a client who no longer trusts you. A client who makes a purchase decision based on it is a problem of a different order.
The rule is simple: every fact in an AI-generated draft must be verified against your own data before it reaches anyone. Not because the fact is probably wrong. Because you cannot tell from the draft which facts were invented and which were accurate, and the only way to distinguish them is to check.
Use AI to write. Never use it to know.
It Does Not Know Your Market
AI has processed an enormous amount of text about real estate. It does not know your market.
It has not driven the neighborhoods at different times of day. It has not tracked which streets see bidding wars and which sit. It has not talked to the agents who consistently work that side of town. It does not know that the school district boundary changed two years ago, or that the HOA just assessed a major repair, or that the property three blocks over sold significantly under ask because of an issue that never made it into any database.
That knowledge is yours. It took time to build. It is what clients are actually paying for when they hire a local agent instead of attempting to navigate a market themselves.
AI produces text. You produce judgment. They are not the same thing, and they are not interchangeable.
When you use AI to draft a comps summary, the market read — what those comps actually mean, what a buyer will pay, whether the price is defensible — that is your analysis, not the AI's output. The summary helps. The judgment is yours.
Client Data Stays Out of Public Tools
The fourth limit is about what goes into the prompt.
When you ask AI to draft a lead follow-up message, a client update, or any communication that involves a specific person, you are describing someone's situation to a third-party service. Public AI chat tools — ChatGPT, Claude, Gemini, and others — may retain the content of conversations for product improvement.
Client names, property addresses, financial details, and deal terms are sensitive. A client's situation is private. It is not yours to share with a third-party service, regardless of how the end product is used.
The guard is straightforward: when you prompt AI for anything client-related, use generic, non-identifying details. Not "my client Sarah Chen at 14 Maple Street with a $450,000 budget" but "a buyer in the $400-500K range looking in an urban neighborhood." The draft will be just as useful. The client's information stays where it belongs.
The Practice
None of this should discourage you from using AI in your workflow. The practice is worth building. The time savings are real.
The limits are simply the checklist you run after the draft arrives:
Does any sentence describe a person instead of the property? Rewrite it.
Does any fact need verification against your own data? Check it.
Does the draft make any claim about market conditions, value, or trends that requires your local read? Make that read.
Does any prompt you sent to the AI contain client information that should have stayed private? Fix the prompt for next time.
Running those four checks takes five minutes. The alternative — not running them — is not a time savings. It is a liability waiting to materialize.
Continue building this practice at academy.jeremyknox.ai — or explore how AI is reshaping professional roles across industries at jeremyknox.ai.