AI Agents for Real Estate Agents: Listings, Leads and Follow-Up

AI agents for real estate are most useful on four jobs: drafting listing descriptions, drafting follow-ups to new leads, proposing showing times, and keeping the checklist for each transaction up to date. In every one, the agent drafts and a licensed person approves before anything reaches a client, a lead or another agent. The legal line that matters most is the Fair Housing Act: an ad may not indicate a preference or limitation based on a protected characteristic, and that applies to text an agent writes exactly as it applies to yours. The simplest rule to give an agent is to describe the property, never the people who should live in it.

This page is about the work of a real estate agent or a small brokerage. The same approach for any small business is in AI agents for small businesses, and the rules for sales email and calls are in AI sales agents. Neither is repeated here.

Four jobs, and who signs off

1. Listing description drafts

• The agent drafts: a description from the facts you give it (square footage, bedrooms, lot, updates, features, the MLS fields you already filled in) in two or three lengths, for the MLS, the website and social posts.
• You approve: every fact against the listing file, and every phrase against the Fair Housing section below. An agent will happily invent a “newly renovated” kitchen if the photos look new.

2. Lead follow-up drafts

• The agent drafts: a reply to each new inquiry that answers the question asked, offers two showing times and says who you are. After a showing, it drafts the recap and the next step.
• You approve: each message before it goes out, at least until the drafts go out unchanged for weeks. Anything about price, offers, other buyers or the seller’s situation stays with you.

3. Showing scheduling

• The agent drafts: proposed times that fit your calendar, the seller’s instructions and the listing agent’s rules, plus the confirmation and the reminder.
• You approve: anything that changes the seller’s instructions, a showing outside the stated hours, and any request to enter a property without an agent present.

4. Transaction checklists

• The agent drafts: a checklist for each deal from your standard list and the dates in the signed contract, such as inspection, appraisal, financing and closing, and a weekly note of what is open.
• You approve: the dates themselves, read from the contract by a person. A wrong contingency date is the kind of mistake that costs a client money.

Describe the property, not the people

HUD’s regulation on discriminatory advertisements, 24 CFR 100.75↗ (opens in a new tab), makes it unlawful to publish any notice, statement or ad for the sale or rental of a dwelling that indicates a preference, limitation or discrimination because of race, color, religion, sex, handicap, familial status or national origin. It covers written and oral statements, and it names “words, phrases, photographs, illustrations, symbols or forms” that convey a home is or is not available to a particular group.

Language models are trained on a great deal of old real estate copy, and some of it describes buyers rather than homes. Put the rule in the agent’s instructions and check every draft for it:

• Describe features: “three bedrooms, fenced yard, 0.4 miles to the elementary school” rather than “perfect for a young family”.
• Describe access, not ability: “single-level, no steps from the driveway” rather than a phrase about who can or cannot manage stairs.
• Describe the place, not the neighbors: distances, parks, transit and amenities, never the kind of people who live nearby or a religious or ethnic label for the area.
• No “ideal for”, “best suited to” or “great for” followed by a kind of person.
• Photos and alt text count too. If the agent writes image descriptions, the same rule applies.

The same regulation also covers where ads are shown: selecting media or locations that deny particular segments of the market information about housing because of a protected characteristic is itself a discriminatory practice. If an agent sets up ad audiences, a person reviews the targeting as carefully as the copy.

HUD’s 2024 AI guidance, and what happened to it

On May 2, 2024, HUD announced two guidance documents↗ (opens in a new tab) on artificial intelligence: one on tenant screening, saying screening companies that use AI must comply with the Fair Housing Act, and one on advertising through digital platforms, warning about ad targeting and delivery that deny people information about housing because of protected characteristics.

That guidance should no longer be relied on. HUD’s Office of Fair Housing and Equal Opportunity withdrew the advertising guidance, along with other documents, in a notice of withdrawal dated September 17, 2025↗ (opens in a new tab), and published that withdrawal in the Federal Register on April 6, 2026. The companion tenant-screening guidance is no longer on HUD’s website either. The notice explains that guidance documents are not binding and do not create obligations beyond the statutes and regulations themselves.

What did not change: the Fair Housing Act and HUD’s regulations, including the advertising rule above, are still in force. HUD’s Fair Housing Act overview↗ (opens in a new tab) lists the protected characteristics as race, color, national origin, religion, sex, familial status and disability. HUD has also proposed removing its discriminatory-effects regulations; in a January 2026 proposed rule↗ (opens in a new tab) it said it would leave questions of disparate-impact liability to the courts. That rule is a proposal, not final, and state and local fair housing laws may add protected characteristics of their own.

This is a plain-language summary, not legal advice. Your broker, your state’s real estate commission and counsel are the right people to ask about your market.

Tenant screening stays with a person

If you manage rentals, keep applicant screening out of the agent’s hands. It can check that an application is complete and list what is missing; it should not rank applicants, score them or recommend who gets the unit. Decisions about people are where an automated pattern does the most harm and is hardest to see, and more examples of where a person steps in are in human-in-the-loop AI examples.

Follow-ups: email, texts and calls

Lead follow-up is outreach, so the CAN-SPAM Act covers commercial email and the Telephone Consumer Protection Act covers calls, including calls with an AI-generated voice. The requirements, and what they mean for an agent, are set out in AI sales agents. The short version for real estate: the agent drafts from a template that already carries your opt-out and address, checks the opt-out list before a send, and a person places the calls.

What a human must approve

• Every listing description, ad and social post before it is published, checked for facts and for Fair Housing language.
• Ad audiences and targeting, if the agent sets them up.
• Every message to a lead or client, until you have months of drafts that went out unchanged.
• Anything about price, offers, negotiations, other buyers or a seller’s circumstances.
• Contract dates on the transaction checklist, read from the signed contract.
• Any decision about an applicant or a buyer: who gets a showing, a callback or a unit.
• Changes to a seller’s showing instructions or access to a property.

Patterns for building that sign-off step into the work are in AI agent approval workflows, and checking the drafts themselves is covered in verifying AI-generated work.

A board for the hand-offs

Once an agent drafts, something has to hold what is waiting for you. Keep dated items where they live now, in your transaction software and calendar, because fenbs has no due dates and no assignee you can set. Use a fenbs board for the decisions: “approve the listing copy for Maple Street”, “review 12 lead replies”, “confirm the inspection date from the contract”. An AI assistant connected over MCP files each one as a task with the details in the note and a priority from 1 to 10, and you move it through To Do, Next Up, In Progress and Completed. History records each change under the assistant’s name.

Put the standing lines, such as “describe the property, never the people” and “no message to a lead without a person”, on the Decisions and rules page. Every connected assistant reads the rules before it starts, and a rule is always decided by a person. Give the assistant the suggested Reporter role, which adds tasks and comments but cannot move tasks or change anyone else’s. Keep client names, financial details and contract documents out of task notes; a transaction ID or an address is enough for you to find the rest.