When Your Website Steers a Buyer and Breaks Fair Housing

It is 9:40 on a Tuesday night. A buyer has three of your listings open in three tabs. She types into the chat widget: "We have two kids under five. Which of these neighborhoods is the better fit for a young family?"

The AI chatbot answers in about a second. "Maplewood, definitely. The schools rate higher, and it is the quieter, safer area of the three."

Helpful. Friendly. Instant. It is also a violation of the Fair Housing Act, and the complaint will name your brokerage, not the software.

Steering Does Not Require Bad Intent

The Fair Housing Act, codified at 42 U.S.C. § 3604, makes it unlawful to discriminate in the sale or rental of housing on the basis of race, color, national origin, religion, sex, familial status, or disability. Steering is one of the oldest violations under that statute. It means influencing where a buyer looks, toward or away from a community, in a way that tracks a protected class.

Here is the part that catches good agents. Steering does not require intent. Courts look at the effect on the buyer’s choices, not the motive behind the words. An agent who calls one area "safer," or "better for families," can create liability when that description lines up with the racial or family makeup of the neighborhood. The agent meant to help. The law does not grade on effort.

Real estate professionals are trained for exactly this. They redirect the neighborhood question. They point buyers to objective public sources and never rank communities on a buyer’s behalf. Your chatbot was trained on the open internet. It will answer the question the second it is asked, with confidence, because answering questions is the only thing it was built to do.

Three Things Your Chatbot Will Say That You Cannot Take Back

The neighborhood question

"Is this a good area?" "Is it safe?" "How are the schools?" A trained agent knows these are tripwires and points the buyer to public data. A chatbot summarizes, ranks, and reassures. Every one of those answers is a statement made on your brokerage’s website. It is written down and time-stamped, and a buyer can screenshot it in a second.

The lender recommendation

A buyer asks, "Who should we use for the mortgage?" The chatbot replies, "We recommend Riverstone Lending." Under RESPA, a referral is any action that affirmatively influences a buyer’s choice of a settlement service provider. If your brokerage has any arrangement with that lender, that one sentence becomes part of a referral. Section 8 of RESPA (12 U.S.C. § 2607) prohibits giving or accepting anything of value for referrals tied to a federally related mortgage loan. The exposure runs to fines and civil liability of up to three times the amount involved.

The price opinion

"Is this house overpriced?" "What should we offer?" In most states, opinions of value and advice on offer terms are licensed activities. An automated tool that hands a buyer a number, or a recommendation, steps into a space the license reserves for a person. And if the number is wrong, you have handed the other side a misrepresentation claim with your name on it.

Your Disclaimer Is Not a Shield

Most firms answer this worry with a line of fine print under the chat box. The thinking is that if the widget says "this is not professional advice," the firm is covered.

It is not. In May 2024, HUD issued guidance that addressed this head-on. The Fair Housing Act applies to housing decisions and housing-related communications whether or not artificial intelligence produced them. The housing provider stays responsible, and that includes the tools and vendors it puts to work. "The bot said it, not us" is not a defense HUD recognizes. The conduct is what the law measures. That conduct happened on your site, under your brand, and it sits in a transcript you cannot delete.

Why the Instruction Quietly Fails

Most vendors will tell you this is solved with a better prompt. Tell the chatbot never to comment on whether a neighborhood is safe or good for families, and the problem goes away.

It does not, and the reason is structural. A language model follows an instruction when the request matches the phrasing it was warned about. Change the phrasing, and the guard slips.

Tell the bot never to say whether an area is good for families, and the buyer simply never phrases it that way. She writes, "I have three kids, between these two areas which would you pick?" To the model, that is not the forbidden question. It reads a helpful request and gives a helpful answer, ranking the two neighborhoods for her. The instruction was sitting right there in the prompt. It just did not catch the sentence that actually mattered.

That is the whole problem in a single moment. An instruction asks the model to behave. It does not stop the model from speaking. Architecture is different. A boundary built into the system decides what the agent is allowed to do before it answers, so a neighborhood ranking for a buyer never leaves the building, no matter how the question is worded. "Will not" is a suggestion. "Cannot" is an architecture.

What a Fair Housing Complaint Costs You

The dollars are real. A HUD complaint can carry civil penalties that reach tens of thousands of dollars for a first violation, and they climb from there. That is before private litigation, where damages and attorney fees stack up fast. Add the cost of defense counsel and the hours pulled away from selling. A state license investigation can also put your ability to operate in question.

Then add the part that never shows up on an invoice. Fair housing complaints are public. For a brokerage that sells trust, a discrimination headline is the kind of damage that outlasts the settlement.

The Question to Ask Before a Bot Speaks for Your Brokerage

The chatbot on your listings is not a junior agent who can be coached. It is a system that will answer anything a visitor types, in writing, in your name, at 9:40 on a Tuesday night when no one is watching.

So the question is not whether your AI is smart. It is whether your AI can be made unable to cross the lines your license depends on. If the answer is that you told it not to, you do not have a boundary. You have a hope.

Fred is built the other way. It answers from your own listings and your own content. It captures the lead and scores it. Anything that calls for licensed judgment goes to a licensed human. Fred runs more than 50 industry guardrail packs, and the real estate pack is built around the lines the Fair Housing Act draws. Fred does not rank neighborhoods for a buyer. It cannot. It listens and answers what it should. The rest it hands to you.

Frequently asked questions