Key takeaways
- Score vendors on six areas: listing knowledge, showing requests, CRM write-back, fair housing guardrails, consent and disclosure, and human handoff.
- Ask for proof, not claims. The FTC has said there is no AI exemption from the laws on the books[2], and a feature that is coming soon is not a feature.
- Only 46% of REALTORS use a CRM[1], so check where call records land whether or not you run one.
- Run a two-week overflow trial on real calls, with a written pass rule for every checklist row, before you forward the main line.
Score vendors on your calls, not their demo
Every AI receptionist sounds good on a demo call the vendor scripted. The questions that separate products only come up on your own calls: the withdrawn listing a caller still asks about, the buyer who wants a showing in an hour, the seller fishing for a price opinion, the caller in a state that requires everyone's consent to record. This checklist covers six areas, each with the question to ask, the proof to request and the red flag that should end the conversation.
Pull two weeks of your own call log before you speak to any vendor. Sort the calls into listing questions, showing requests, seller inquiries, existing clients, vendors and junk. That mix tells you which checklist areas carry the most weight. A listing-heavy solo agent cares most about listing knowledge and showing requests; a 40-agent brokerage cares most about routing and handoff.
Give setup effort a weight too. In NAR's 2026 technology survey, 63% of REALTORS named the learning curve as their biggest challenge with new technology and 59% pointed to cost[1]. A product nobody on the team will maintain is a product giving stale answers by spring. It also has to fit the tools you already run, and those vary more than vendors assume.
1. Listing knowledge
Callers ask about price, bedrooms, HOA dues, parking, pets, school assignment and whether the home is still available. Ask where every one of those answers comes from and who on your side keeps it current.
- Source: an approved listing sheet you control, a feed, or a search of public websites. The last is a red flag, because portal data lags and you can't sign it off.
- Gaps: what the caller hears when a fact isn't on the sheet. The right answer is that it isn't confirmed and an agent will follow up, never a guess.
- Changes: how long a price drop or status change takes to reach the phones, and which person on your team makes it.
- Withdrawn and pending listings: what a caller hears about a home that is no longer active.
- Caller context: a forwarded call doesn't reveal which sign or ad the caller saw, so check that it asks.
For reference, the SI Employee on this site answers only from your approved listing sheet and marks anything else as not confirmed for your agent; sync from an MLS or IDX feed is planned, not live. Preparing listing data for an SI Employee covers how to build the sheet so it stays current.
2. Showing requests
There are two honest models. Some products take a showing request, with the property, preferred times and whether the buyer already has an agent, and hand it to an agent to confirm. Others book straight into a showing tool against access rules. Either can work. What matters is that the product says clearly which one it does, and that a request is never presented to the caller as a confirmed appointment.
Booking has a paperwork edge as well. Buyers working with an agent are asked to sign a written agreement before touring a home with that agent, in person or virtually, while visiting an open house alone doesn't require one[3]. A system that books private tours before your agent has spoken to the buyer can put a showing on the calendar ahead of that agreement. Asking about agent status and letting your agent confirm keeps the step with a licensee.
NAR found 68% of REALTORS use showing scheduling tools[1]. If you are one of them, ask whether the receptionist writes into that tool or only passes on a request. Direct booking into third-party showing systems is often described as on the roadmap, so get the answer in writing. How AI showing scheduling works explains the mechanics.
3. CRM write-back
A call that never reaches the CRM didn't happen as far as your follow-up plan is concerned. Fewer than half of REALTORS use a CRM, and among those who do, the usual failure is a lead that arrives as an email and never gets typed in.
- Native or connected: ask whether the link to your CRM is a built and tested integration or runs through signed webhooks or tools such as Zapier, Make or n8n. Both work; the second needs someone to map the fields once.
- Fields: name, phone, email, property, buyer or seller, timeline, agent status, summary and a link to the recording.
- Duplicates: what happens when a known contact calls again. A new lead record for every call clutters the database fast.
- Ownership: whether the record is assigned to the right agent under your routing rule.
- Proof: insist on a test call you watch arrive in your CRM before you sign anything.
The connection options for the product on this site, all via webhook, Zapier, Make or n8n rather than native connectors, are listed on the AssistCall integrations page. If you are still choosing a CRM, Follow Up Boss vs Lofty vs kvCORE compares the common ones.
4. Fair housing guardrails
The Fair Housing Act protects people from discrimination because of race, color, national origin, religion, sex, familial status and disability[4]. A phone system can break it without anyone intending to: by answering “is it a good area for kids?” with an opinion, by asking some callers more screening questions than others, or by nudging a caller toward or away from a neighborhood.
The stakes are concrete. Private fair housing groups, HUD, state and local agencies and the Justice Department received 32,321 fair housing complaints in 2024[5]. NAR requires members to complete fair housing training in every three-year cycle[6]. Your phone script deserves the same standard as your agents.
- Ask for the default answers to questions about schools, safety, crime, demographics and what kind of neighborhood it is. They should point to public sources or an agent and offer no opinion.
- Check that intake questions are identical for every caller on the same type of call.
- Check that you can read and edit every question and answer it uses.
- Ask how you would investigate a complaint about a call: are the recording and transcript kept and easy to find?
AI and fair housing: bias risks in lead handling goes deeper on how automated intake can drift.
5. Consent and disclosure
Recording and AI rules differ by state. California punishes recording a confidential call without the consent of all parties[7]. Washington also requires every participant's consent, and treats a recorded announcement to everyone on the call as consent[8]. Utah requires a business using generative AI in a consumer transaction to disclose it when a person clearly asks, and protects businesses that disclose clearly from the outset[9].
- The greeting can carry a recording notice and an automated-assistant disclosure, and you approve the exact words.
- It answers truthfully when a caller asks whether they are speaking to a person.
- Recordings and transcripts are stored where you can find, export or delete them on request.
- Outbound calling, if offered, is a separate question with separate consent rules.
State AI disclosure laws for calls and chat tracks the state rules in more detail.
6. Human handoff
Handoff is where most of the risk sits, because it is the moment the software admits it has reached its limit. Ask the vendor to walk you through four situations: an agent is free, nobody answers, the caller insists on a person, and the call is urgent.
- Transfers go only to numbers you approve, by listing or team rule. A call can't be transferred back to the line that forwarded it, so list agents' mobiles, not the main office number.
- If nobody picks up a transfer, the caller is told plainly and an urgent message is saved, rather than being left on hold.
- Urgent messages are marked urgent. Ask exactly how the agent hears about them and test it on a Sunday.
- A manager can listen to a live call, whisper guidance the caller doesn't hear, take over or end it.
The brokerage front desk routing example shows a written routing sheet for 40 agents built around one duty line and a named owner for every message.
The scoring sheet
| Area | Ask the vendor | Proof to request | Red flag |
|---|---|---|---|
| Listing knowledge | Where do listing answers come from? | A test call about a withdrawn listing | It guesses or quotes a portal |
| Showing requests | Do you book, or take a request? | A request arriving with property, times and agent status | Callers told a showing is confirmed when it isn't |
| CRM write-back | How does a call reach our CRM? | A test lead in your CRM with the right owner | A native integration nobody can demonstrate |
| Fair housing | What does it say about schools and safety? | The full list of default answers | Opinions about neighborhoods |
| Consent | What does the greeting say? | Greeting text you can edit and approve | No way to add a recording notice |
| Handoff | What happens when nobody answers? | A failed-transfer test call | Caller left on hold or dropped |
| Claims | Which features are live today? | A written live versus planned list | Roadmap items sold as current |
Weight by your call log. Example: if listing questions and showing requests make up 60% of your calls, double the scores on those two rows. A vendor that scores 2 everywhere except handoff still fails, because a failed handoff is the call that turns into a complaint.
Treat marketing claims the way you would treat a buyer's pre-approval letter from a lender you've never heard of. Announcing five enforcement actions against deceptive AI claims and schemes in 2024, the FTC said there is no AI exemption from the laws on the books[2]. Ask for a written list of what is live and what is planned, and score only what is live.
Size the stakes before you sign
How much evaluation effort the decision deserves depends on what unanswered calls are worth to you. Put your own numbers into the calculator.
| Input | Example |
|---|---|
| Inbound calls per week | 35 |
| Share that ring out or hit voicemail | 30% |
| Share from new buyers or sellers | 35% |
| Share of those that become a closed side | 4% |
| Gross commission income per side | $6,578 |
| Unanswered calls a month | 45 |
| From new customers | 16 |
| Bookings that would have followed | 0.6 |
| Monthly value at risk | $4,187 |
How it is worked out: Calls a week × 4.33 weeks × unanswered share × new-customer share × booking rate × average value.
Assumptions
- Value per side: the median REALTOR's 2025 gross income of $59,200 spread across the typical nine transaction sides, about $6,578[10].[10]
- 67% of agent-using sellers contacted only one agent[11], so a missed first call is often the only chance.[11]
- Call count, missed share and closing rate are example figures. Use two weeks of your own call log.
Example only. It shows the value at risk, not a forecast of what any product will recover.
Run a two-week trial on real calls
- Your team
- System
- SI Employee
01Your team
Pull two weeks of calls
Sort them into listing, showing, seller, client, vendor and junk calls.
02Your team
Write the sheet and rules
Listing sheet, routing, greeting, disclosure and boundaries, all signed off.
03Your team
Scripted test calls
Withdrawn listing, price question, fair housing question, failed transfer.
04System
Forward overflow only
Real calls reach it only when agents are busy or the office is closed.
05SI Employee
Answer, route, record
Takes requests, routes by rule, saves recording, transcript and summary.
06Your team
Review every record
Check summaries against recordings and CRM entries each morning.
Example sequence. Move to full forwarding only when every row of the scoring sheet has a pass on a real or test call.
Starting with overflow means a bad answer reaches a handful of callers instead of all of them. Running an AI pilot at a brokerage covers roles and how to gather honest agent feedback, and AI receptionist vs human answering service helps if your shortlist still includes a staffed service.
Frequently asked questions
What is the best AI receptionist for real estate agents?
The one that passes your own test calls in the six areas above, weighted by your call mix. Rankings and reviews describe other people's listings and scripts. Score only features you have seen work live.
Should an AI receptionist book showings directly?
Only if it handles access rules and the buyer agreement step. Buyers are asked to sign a written agreement before touring with an agent[3], so a request your agent confirms is the safer default.
Do I need a CRM before I buy one?
No, but you need a place where records land and someone who reads them. Fewer than half of REALTORS use a CRM[1]; an emailed summary with the recording and transcript is a reasonable start.
What should the greeting say?
Your business name, a short automated-assistant disclosure and a recording notice, in words you approve. All-party consent states such as California make the recording notice essential[7].
How long should the trial run?
At least two weeks on overflow, covering two weekends, so you see Saturday sign calls and Sunday evening seller calls. Extend it if any checklist row is still unproven.
References
National Association of REALTORS® (2026). REALTORS® Adopt Technology to Save Time and Improve the Client Experience, NAR Report Finds (2026 REALTORS® Technology Report) (opens in a new tab) ↑
Federal Trade Commission (2024). FTC Announces Crackdown on Deceptive AI Claims and Schemes (opens in a new tab) ↑
National Association of REALTORS® (2024). Consumer Guide: Written Buyer Agreements (opens in a new tab) ↑
U.S. Department of Housing and Urban Development (2026). Housing Discrimination Under the Fair Housing Act (opens in a new tab) ↑
National Fair Housing Alliance (2025). 2025 Fair Housing Trends Report (opens in a new tab) ↑
National Association of REALTORS® (2025). Fair Housing Training Requirement (opens in a new tab) ↑
FindLaw (California Penal Code text) (2026). California Penal Code § 632 - Eavesdropping on or recording confidential communications (opens in a new tab) ↑
Washington State Legislature (2026). RCW 9.73.030 - Intercepting, recording, or divulging private communication (opens in a new tab) ↑
Utah State Legislature (2025). S.B. 226 Artificial Intelligence Consumer Protection Amendments (2025 General Session, enrolled) (opens in a new tab) ↑
National Association of REALTORS® (2026). Experienced REALTORS® Anchor the Industry as Housing Affordability Remains Top Hurdle (2026 Member Profile) (opens in a new tab) ↑
Zillow Research (2026). Sellers: Results from the Zillow Consumer Housing Trends Report 2026 (opens in a new tab) ↑