AI Insights Real Estate & Property
Fair Housing & RESPA Online: A 2026 Playbook for Brokerages
Talk to Fred
Ask Fred about Real Estate & Property
This is the same Fred you would put on your own site. Ask about Real Estate & Property, compliance, or how the guardrails work. Fred listens.
A real estate assistant that works is easy to build. One that works and stays inside fair housing law is a different project, and 2026 is the year the gap between the two stopped being theoretical. HUD has said plainly that the Fair Housing Act applies to housing communications whether or not a machine produced them. That single position turns every automated answer on your listings into something the law can read back to you.
This is the standard side of the story. The threat companion covers the sentence that gets a brokerage named in a complaint. This guide covers the opposite: what a compliant assistant has to refuse, what it is free to do, and how to deploy one without taking on the liability the threat piece describes.
The Law Measures Conduct, Not the Author
Steering under the Fair Housing Act, 42 U.S.C. § 3604, has never required intent. It is measured by the effect on a buyer’s choices, toward or away from a community, in a way that tracks a protected class. That doctrine predates artificial intelligence by decades, and it does not soften because the words came from software. HUD’s 2024 guidance closed the obvious escape hatch: the housing provider stays responsible for the tools and vendors it deploys, and "the bot said it" is not a defense the agency recognizes.
So the design question for 2026 is not whether your assistant is clever. It is whether your assistant can be prevented from doing the things a trained agent is trained never to do. A licensed agent learns to redirect the neighborhood question and hand buyers objective public sources. The compliance standard simply holds the machine to the same discipline, and holds it there structurally rather than hopefully.
The Three Acts a Compliant Assistant Will Not Perform
Three categories carry almost all the risk in real estate, and a compliant assistant treats all three as off limits.
The first is characterizing a community. Ranking neighborhoods, calling an area "safer" or "better for families" or "the nicer part of town," or steering toward and away from places is the fair housing tripwire, because those descriptions so often track the racial, ethnic, or family makeup of an area. The compliant move is the agent’s move: decline to rank, and point the buyer to objective public sources like school district data, crime statistics, and census records.
The second is the settlement-service referral. When a buyer asks who to use for the mortgage, title, or inspection, an answer that names a provider can become a referral under RESPA. Section 8, at 12 U.S.C. § 2607, prohibits giving or accepting anything of value for referrals tied to a federally related mortgage loan, with liability that can reach three times the amount involved. A compliant assistant offers neutral options or routes the question to a person, and never steers toward an affiliated provider.
The third is opinion of value and offer advice. In most states, opinions of value and advice on offer terms are licensed activities. An assistant that hands a buyer a number or a recommended offer is operating in the space the license reserves for a person, and a wrong number invites a misrepresentation claim with the brokerage’s name on it.
Disclosure and Recordkeeping Are Part of the Standard
Compliance in 2026 is not only about silence in the right places. Every conversation an assistant has is a written, time-stamped housing communication. That is a liability when the content is wrong, and an asset when the content is clean and the record is kept. A compliant deployment logs each exchange, keeps it reviewable, and treats the transcript as part of the brokerage’s recordkeeping rather than as something that disappears when the chat window closes. If a complaint ever comes, the difference between a defensible position and an indefensible one is often whether you can show what the system actually said and that it held its boundaries.
Why an Instruction Fails the Standard
The usual plan is to write the rules into the assistant’s prompt. Tell it never to comment on whether a neighborhood is safe or good for families, and treat the matter as handled.
It is not handled, and the reason is structural. A language model obeys an instruction when the request matches the phrasing it was warned about. Reword the request and the guard slips. You tell it never to say whether an area is good for families. The buyer does not ask it that. She writes, "we both work downtown and want the easiest commute with good schools, which of these two should we focus on?" The model hears a logistics question and answers it, and in answering it ranks the neighborhoods. The instruction was loaded the entire time. It just did not catch the sentence that mattered.
That is the gap between an instruction and a standard. An instruction asks the model to behave. It does not stop the model from speaking. A real boundary is built into the system and decides what the assistant is allowed to do before it answers, so a neighborhood ranking or a value opinion never reaches the buyer no matter how the question is worded. "Will not" is a suggestion. "Cannot" is an architecture.
What a Compliant Deployment Looks Like
Meeting the standard does not mean pulling the assistant off your listings. It means running one that was built around the lines the Fair Housing Act, RESPA, and your license draw, and that keeps the record those rules expect.
Fred is built that way. It answers from your own listings and your own content, captures and scores the lead, and routes anything that calls for licensed judgment to a licensed agent. It runs more than 50 industry guardrail packs, and the real estate pack is built around steering, settlement-service referrals, and opinions of value. Fred does not rank neighborhoods for a buyer, does not steer to an affiliated lender, and does not hand out a price opinion. It cannot. It answers what it should, points buyers to objective public sources, logs the exchange, and hands you the rest.
The point is not a smarter assistant. It is one that can be shown to be incapable of the answers that get a brokerage named.
Frequently asked questions
What is the single most important thing a real estate AI assistant should refuse to do?
Rank or characterize neighborhoods for a buyer. "Which area is better for a family like ours?" is the question buyers most want answered and the one most likely to be steering under the Fair Housing Act. A compliant assistant declines to rank and points the buyer to objective public sources instead.
Does the Fair Housing Act really apply to an automated answer?
Yes. HUD’s 2024 guidance states that the Fair Housing Act applies to housing-related communications whether or not artificial intelligence produced them, and that the housing provider remains responsible for the tools it deploys. The conduct is what the law measures, and the conduct happens on the brokerage’s site under its brand.
Can the assistant recommend a lender or title company?
Be careful. Naming a settlement-service provider can become a referral under RESPA Section 8, and if any thing of value flows for that referral on a federally related mortgage loan, it is prohibited. A compliant assistant offers neutral options or routes the question to a person and does not steer toward an affiliated provider.
