AI Receptionist for Real Estate Agents

What it actually handles on a real estate call

• A sign or listing call: the caller has a specific address in mind and wants price, beds and baths, and whether they can see it. Matched to a call-tracking number tied to that listing, the AI can answer with the actual property rather than a generic greeting.
• A portal-generated buyer lead calling in: needs timeline, financing status, and must-haves captured immediately, because the same lead is frequently routed to more than one agent.
• A seller inquiry: wants a valuation conversation started or a listing appointment on the calendar, without waiting for a callback window.
• An existing client mid-transaction: needs a person, quickly. This is a routing job, not a scripted one.
• Everything else: vendors, other agents, wrong numbers, and solicitors, which is a meaningful share of an agent's inbound volume and none of it needs the agent.

Why the calls matter more here than the miss rate suggests

Real estate is not the worst offender for unanswered calls. CallRail's January 2025 "From Conversations to Conversions" report, an analysis of 1.1 million leads, put real estate's missed-call rate at 9%, the lowest of the categories it measured and well under healthcare (32%), legal (28%), and home services (14%). The cost per miss is what's unusual: the same report found up to 85% of callers whose call goes unanswered never call back, and a real estate caller has an unusually easy substitute one tap away, the next agent, the portal's other contact button, or a colleague at the same brokerage.

That's the case for live answering rather than recovery, and it's covered in more depth from the recovery side in the missed-call text-back guide for agents. The rest of this guide is about the constraint that makes live answering different in real estate than in a trade like plumbing: what the thing answering the phone is permitted to say.

The limit vendors don't mention: an unlicensed answering layer has a narrow script

Every state regulates who may perform real estate brokerage activity, and most real estate commissions publish rules on what an unlicensed assistant may and may not do on a licensee's behalf. The National Association of REALTORS® maintains a state-by-state table of those statutes and regulations precisely because the line moves between jurisdictions. The common shape of the rule is narrow: an unlicensed assistant may confirm that a property is listed, identify the listing broker, and provide the kind of factual information that would appear in a simple classified advertisement, typically address, price, and basic specifications. North Carolina Real Estate Commission guidance states the restriction in close to those terms, and Connecticut's regulations go further, barring unlicensed assistants from answering consumer questions on listing, title, financing, or closing at all.

An AI receptionist is not a licensed person, so the conservative reading is that it operates inside that same envelope. In practice that means the script can state published listing facts and set an appointment, and must not: discuss or explain an offer, a contract, an agency agreement, or a disclosure; characterize value or advise on price; negotiate or solicit terms; or answer financing, title, or closing questions. This is not a limitation to engineer around, and a vendor advertising that its agent will "answer any question about your listings" is describing a compliance problem, not a feature. Confirm the specific rule with your own state commission or brokerage counsel before the script goes live; none of this is legal advice.

Fair housing: the section that changed in April 2026

The other constraint is the Fair Housing Act, which prohibits steering, meaning directing prospects toward or away from neighborhoods on the basis of a protected characteristic. For years the practical industry advice was to refuse questions like "is this a safe neighborhood?" outright, and NAR published material to that effect, including an article titled "The Safety Series: 'Is This a Safe Neighborhood?' Don't Answer That."

That guidance was directly challenged. On April 24, 2026, HUD issued a Dear Colleague letter, HUD No. 26-028, signed by Craig Trainor, Assistant Secretary for Fair Housing and Equal Opportunity, stating that real estate professionals do not violate the Fair Housing Act merely by discussing the prevalence of crime or the quality of schools with prospective buyers or renters, and that unlawful steering requires intentional discrimination based on a protected characteristic. The letter explicitly called on the industry to revisit its ethics training materials and named NAR's "don't answer that" guidance. Coverage from RISMedia, National Mortgage News, HousingWire, and The MortgagePoint in late April 2026 all describe it the same way.

What did not change is the risk underneath. HUD's letter addresses whether discussing the topic is itself a violation; it does not immunize the content of the answer. Courts have treated racially coded remarks about crime and schools as evidence of discriminatory intent, and the safe practice is unchanged: share objective, verifiable, sourced data, give the same information to every caller who asks, and avoid subjective characterization, because "it's a good area" is an opinion, not a data point.

Why an AI is well-suited to that and badly suited to it at the same time

Consistency is the whole ballgame in a steering analysis, and consistency is the one thing a scripted system is genuinely better at than a tired human on a Sunday evening. An AI receptionist that gives every caller the identical response to a neighborhood question, and logs that it did, is in a stronger evidentiary position than an agent relying on memory of what they said to whom.

The risk runs the other way too. A general-purpose language model that improvises will eventually produce a subjective characterization of a neighborhood, and that sentence, generated at scale across thousands of calls, is exactly the kind of evidence a fair-housing complaint is built from. So the practical build is a fixed, reviewed response to this category of question rather than a generated one: acknowledge the question, decline to characterize, point to published public data sources, and offer to connect the caller with the agent. Whatever policy the brokerage sets after the HUD letter, it should be written into the script deliberately and approved by the broker, not left to the model's discretion.

Does it have to tell callers it's AI?

Sometimes by law, and a well-built one does regardless. California's Bolstering Online Transparency Act, at Cal. Bus. & Prof. Code § 17941 (effective January 1, 2019), makes it unlawful to use a bot to mislead someone about its artificial identity in order to incentivize a sale, with a safe harbor for a clear and conspicuous disclosure that it's a bot. Utah's AI Policy Act (SB 149, 2024) requires disclosure that a consumer is interacting with generative AI when asked, and imposes a proactive disclosure duty on regulated occupations, those licensed through the state's Department of Commerce, which is where Utah's Division of Real Estate sits; SB 226, signed in March 2025 and effective May 7, 2025, narrowed that proactive duty to high-risk interactions. State AI disclosure law is moving quickly and unevenly, so a multi-state brokerage should assume the strictest applicable rule rather than tracking fifty of them.

The reputational answer is simpler than the legal one. Callers who discover mid-conversation that the confident voice discussing their home purchase was software react badly, and an upfront identification costs nothing.

Recording and transcription, since that's how the CRM gets fed

The reason to run an AI receptionist rather than an answering service is usually that every call lands in the CRM as structured data, which means calls get recorded or transcribed. Federal law under the Electronic Communications Privacy Act sets a one-party consent baseline, but roughly eleven states, commonly listed as California, Delaware, Florida, Illinois, Maryland, Massachusetts, Montana, Nevada, New Hampshire, Pennsylvania, and Washington, require all-party consent, with several more unsettled. When a call crosses state lines the stricter law generally controls.

A real estate agent takes calls from out-of-state buyers routinely, so the workable default is an all-party notice on every call: an automated disclosure before the conversation begins, with continuing on the line treated as consent. That's one line at the top of the script and it removes the question entirely.

What it should hand to a person immediately

• Anything touching an offer, a counter, a contingency, or a contract term.
• Any question about value, what a property is worth, or what someone should offer.
• Financing, title, inspection, and closing questions, which belong to the lender, title company, or agent.
• An existing client with an active transaction, who should not be triaged by a script.
• Any caller who asks for a person, and any caller who's audibly frustrated, without making them repeat themselves to get there.

Where it fits with text-back and follow-up automation

Live answering, missed-call recovery, and follow-up are three layers of the same problem, not competing products. The AI receptionist answers the call. Missed-call text-back catches the ones that still slip through, two calls at once, a caller who hangs up at the fourth ring. Follow-up automation handles the async side, the portal leads and form fills that were never a phone call to begin with, and keeps a qualified lead warm past the first conversation. Teams that only buy the first layer tend to find the leak has just moved.

For the general mechanics of how these systems work on a call, independent of real estate, see how AI voice agents work and what an AI receptionist is.

Sources

• CallRail, "From Conversations to Conversions" (analysis of 1.1 million leads, published January 2025) — supports the 9% real estate missed-call rate, the healthcare/legal/home-services comparison, and the 85% no-callback figure.
• National Association of REALTORS®, "State Statutes and Regulations for Unlicensed Assistants" table; North Carolina Real Estate Commission guidance on unlicensed assistants; Connecticut regulations on unlicensed assistants — support the scope-of-permitted-activity description. State rules vary; not legal advice.
• U.S. Department of Housing and Urban Development, Dear Colleague letter HUD No. 26-028, "HUD Empowers Real Estate Agents to Better Support American Homebuyers," issued April 24, 2026 by Assistant Secretary for Fair Housing and Equal Opportunity Craig Trainor, as reported by RISMedia, National Mortgage News, HousingWire, and The MortgagePoint (April 2026) — supports the steering-guidance change and the intent standard.
• Cal. Bus. & Prof. Code § 17941 (Bolstering Online Transparency Act, effective January 1, 2019) — supports the California bot-disclosure requirement and safe harbor.
• Utah Artificial Intelligence Policy Act, SB 149 (2024), as amended by SB 226 (2025, effective May 7, 2025), per contemporaneous analyses from Greenberg Traurig, Davis Polk, and Davis Wright Tremaine — supports the generative-AI disclosure duty and the regulated-occupation / high-risk-interaction narrowing.
• Electronic Communications Privacy Act (federal one-party consent baseline) and state all-party consent statutes, per 2026 state-by-state compilations — support the recording-consent description. State lists differ at the margins; confirm for your states.