How do AI receptionists work? A plain-language guide
An AI receptionist answers a business phone line with a synthetic voice, understands what the caller says, decides what to do from a set of instructions the business wrote, and acts: it answers a question, books an appointment, takes a message, or transfers the call to a person. This guide explains what happens in the two seconds after a caller speaks, what these systems do well and badly, the consent rules, and what they cost. We sell one, and we say so where it matters.
Updated September 7, 2026. Written by the LeadSixty team.
What is an AI receptionist?
An AI receptionist is software that answers phone calls in a business's name and holds a spoken conversation with the caller. It is not a phone tree ("press 1 for sales"), because the caller talks normally and the system responds to what they actually said. It is not a live answering service, because no human is on the line. And it is not a chatbot, because it works over the phone with voice, in real time, where a two-second pause feels like a dropped call.
Four things people confuse it with, and how they differ:
| Option | Who answers | Can it book, look things up, transfer? | Hours |
|---|---|---|---|
| Voicemail | Nobody | No | Always, but nobody calls back until someone listens |
| Phone tree (IVR) | A menu | Routes only; cannot answer a question | Always |
| Live answering service or virtual receptionist | A person at a call center | Yes, from a script; booking depends on the service | Whatever you pay for; 24/7 costs more |
| AI receptionist | Software with a voice | Yes, within the instructions it was given | Always, at the same price |
| Front desk staff | Your employee | Yes, with judgment | Business hours, one call at a time |
What happens during a call, step by step
Five things happen in a loop, many times per call, each taking a fraction of a second.
- Listen: speech becomes text. The caller's audio is streamed to a speech-recognition model that turns it into words as they are spoken, not after the caller stops. Good systems start processing before the sentence ends, which is most of the difference between a natural pause and an awkward one.
- Understand: a language model reads the transcript and the business's instructions. The instructions are a written brief: who the business is, hours, services, what to say and not say, when to book, when to transfer. The model works out what the caller wants and what the right next action is. This is the same kind of model behind text assistants, running with a short, strict brief instead of an open conversation.
- Act: the model uses tools. Booking an appointment means checking a real calendar and writing to it. Answering "are you open Saturday" means reading the hours from the brief. Transferring means dialing a person and, if they answer, bridging the call with a spoken summary. Taking a message means writing it down and sending it by text or email. Each of these is a function the model can call; a system without them can only talk.
- Speak: text becomes voice. The reply is turned into audio by a text-to-speech model and played to the caller. Latency, the gap between the caller finishing and the reply starting, is the number that decides whether the call feels normal. Vendors compete on it and publish targets under one second; test it on a real call rather than trusting the number on a pricing page.
- Remember: everything is logged. The transcript, the outcome, the booking and any message are stored on a contact record and usually summarized for the business owner. A question the AI could not answer is logged so the brief can be improved.
The business's written brief is the whole product. Two systems using the same underlying models will behave completely differently depending on how carefully the hours, services, boundaries and hand-off rules were written and tested. That is where the setup time goes, and where most do-it-yourself setups fall short.
How does an AI receptionist book appointments?
It reads live availability from the calendar the business already uses, offers the caller two or three real openings, confirms the one they pick, writes it to the calendar, and sends the caller a confirmation by text. A well-built one respects working hours, buffers between appointments and per-person calendars, so it will not book a consult with a provider who is off that day. If the caller needs to reschedule, the same loop runs again. If the business needs the caller's details first, name and phone are captured before the slot is offered.
What happens when it cannot answer a question?
A well-configured AI receptionist is told to say so and move on, not to guess. The usual rule is: if the answer is not in the brief, take a message or transfer. The question is logged, and the business adds the answer to the brief if it comes up again. The risk in this technology is a system that was allowed to improvise; the fix is a brief that limits what it may say and a hand-off path that is always available. Ask any vendor what their system does with a question it was not given the answer to. The right answer is "it transfers or takes a message"; the wrong answer is "it figures it out".
What can an AI receptionist do well, and what can it not?
Good at: answering at any hour and never being busy; handling several calls at once; answering the same twenty questions consistently; booking, confirming and rescheduling; taking a clean message with a call-back number; transferring when a person is available; texting the caller a confirmation or a link.
Not good at, and what a careful setup does about it:
- Heavy accents, background noise, two people talking. Speech recognition degrades. Mitigation: the system asks the caller to repeat, confirms details back ("that was 555-0142, correct?"), and transfers if confidence stays low.
- Emotional or complicated calls. An angry customer, a medical worry, a legal question. Mitigation: these are transfer triggers written into the brief, not conversations the AI is allowed to run.
- Anything outside the brief. Left unguarded, a language model will produce a plausible answer to almost anything, including wrong ones. Mitigation: the brief restricts it to approved answers and a "take a message" fallback; test calls before launch ask off-script questions on purpose.
- Judgment calls. Whether to squeeze in an extra appointment, whether a discount is warranted, whether a caller is a good fit. Mitigation: those stay with a person, and the AI's job is to get the caller to that person with the details already collected.
Will callers know they are talking to AI?
Often, yes, and the honest setups do not hide it. The receptionist introduces itself as the business's assistant, and callers who ask are told plainly. In practice most callers care far more about being answered on the second ring at 7 pm than about who answered. What callers dislike is being deceived, or being trapped: a system with no way to reach a human. Both are configuration choices, and both are worth asking a vendor about.
Is it legal to record calls with an AI receptionist?
Most AI receptionists record and transcribe calls, so recording law applies. Under federal law, 18 U.S.C. § 2511(2)(d), a call may be recorded with the consent of one party to the conversation, which the business itself can supply. A number of states go further and require every party to consent. The states usually listed as all-party consent include California, Florida, Illinois, Maryland, Massachusetts, Pennsylvania and Washington; a few others are treated as all-party in some readings and not in others, so check the Reporters Committee's state-by-state recording guide for the states you operate in and the states your callers are in.
The practical answer used by most businesses is a one-line disclosure at the start of the greeting ("this call may be recorded") which satisfies the all-party rule because a caller who continues has consented. A careful setup includes that line by default in every state, because a business in a one-party state still receives calls from all-party states. This is a description of how the rules work, not legal advice; a compliance question about your specific situation belongs with a lawyer.
How much does an AI receptionist cost?
Software-only products price by minutes or by callers and start well under $100 a month. Human answering services price by minutes or calls and cost several times more. Done-for-you setups, where a company writes the brief, connects the calendar and phone, tests and maintains it, sit in between, priced as setup plus a retainer. These figures were read on each vendor's own pricing page in September 2026 and will change; the links go to the current page.
| Product | Type | Published price | Source |
|---|---|---|---|
| Dialzara | AI, self-serve | $29/mo for 60 minutes up to $349/mo for 1,000 minutes; overage $0.35 to $0.48 per minute | dialzara.com |
| Rosie | AI, self-serve | $49/mo for 250 minutes, $149/mo for 1,000, $299/mo for 2,000 | heyrosie.com |
| Goodcall | AI, self-serve | $79, $129 or $249 per agent per month for 100, 250 or 500 unique callers, unlimited minutes | goodcall.com |
| My AI Front Desk | AI, self-serve | Free tier with 20 minutes; $99/mo for 200 voice minutes; overage $0.25 per minute | myaifrontdesk.com |
| Smith.ai | Human receptionists | $300/mo for 30 calls up to $2,100/mo for 300 calls | smith.ai |
| Ruby | Human receptionists | $250/mo for 50 minutes up to $1,725/mo for 500 minutes | ruby.com |
| LeadSixty AI receptionist | AI, done for you | One-time setup plus a month-to-month retainer, quoted on a call. We are the authors of this guide. | leadsixty.com |
The price is rarely the deciding factor. A $49 plan that was never given a proper brief loses the same calls voicemail loses, just more politely. What you are really buying is the hours someone spends writing and testing the brief, which is either your hours or a vendor's.
AI receptionist versus answering service versus hiring: which is right?
| Front desk employee | Live answering service | AI receptionist, self-serve | AI receptionist, done for you | |
|---|---|---|---|---|
| Answers 24/7 | No | If paid for | Yes | Yes |
| Several calls at once | No | Yes | Yes | Yes |
| Books onto your real calendar | Yes | Sometimes | Yes, if you connect it | Yes |
| Judgment on unusual calls | Yes | Limited, scripted | No; transfers | No; transfers |
| Who writes and maintains the brief | You train them | The service, from your notes | You | The vendor, with you |
| Cost pattern | Salary | Per minute or call, high | Per minute or callers, low | Setup plus retainer |
| Best for | Businesses where the front desk sells | Businesses that want a human voice at any cost | Owners willing to configure and test it themselves | Owners who want the result without becoming the operator |
How many calls are small businesses actually missing?
The most cited figure comes from a field test that 411 Locals published in 2016: over 30 days, across 85 businesses in 58 industries, 37.8% of calls were answered, 37.8% went to voicemail and 24.3% got no response at all (the source gives the same 37.8% for both of the first two, and the three figures sum to 100%). Vendors round this to "62% of calls go unanswered" and often date it 2024; the underlying test is from 2013 to 2015. It is small and old, and it is the only published field test of its kind we could find, so treat it as an indication, not a benchmark. Your own number is easy to get: most phone systems report answered versus missed calls by hour, and the after-hours share is usually the surprise.
What a missed call costs depends on what a caller is worth. For a dental office, a home services company or a med spa, a single new-patient or new-job call can be worth hundreds of dollars, which is why the after-hours line is the first place these systems pay for themselves. The same logic applies to text leads, which is the subject of our speed to lead guide.
Which businesses benefit most?
- Home services. HVAC, plumbing, roofing, electrical. Calls arrive while every tech is on a roof or under a sink, and a missed call is a job that goes to the next listing. More on home services.
- Dental and medical offices. New-patient calls during lunch and after 5 pm; appointment changes all day.
- Med spas and aesthetic clinics. Consult requests by phone and DM in the evening, after the person has seen an ad. More on med spas.
- Law firms. Intake calls where a missed first call means the caller hires whoever picked up.
- Insurance agencies and mortgage offices. Quote and pre-approval calls, routed to a licensed person with the details already taken. Insurance and loan officers.
- Car dealers and brokers. Inventory and availability calls at night, booked to a rep for the morning. More on dealers.
How do you evaluate an AI receptionist before buying?
Call it. Every vendor will give you a demo number. Do these six things on the call:
- Ask an ordinary question (hours, location) and time the pause before it answers.
- Ask something that is not in its brief and see whether it admits it does not know or makes something up.
- Ask to speak to a person and see how it handles the hand-off.
- Try to book an appointment and see whether it offers real times or just says someone will call.
- Speak with background noise, interrupt it mid-sentence, and mumble a phone number.
- Ask "are you a real person?" and listen to the answer.
Then ask the vendor who writes the brief, who tests it, who updates it when the hours change, and what happens to a question it cannot answer. The answers tell you whether you are buying software or a result.
AI receptionist questions, answered
How does an AI receptionist actually answer the phone?+
Is an AI receptionist the same as an answering service?+
Can an AI receptionist book appointments directly on my calendar?+
What happens when an AI receptionist cannot answer a question?+
Do AI receptionists sound robotic, and can callers tell?+
How much does an AI receptionist cost per month?+
Is an AI receptionist worth it for a small business?+
Can it handle a heavy accent or background noise?+
Is it legal to have an AI answer calls without telling the caller?+
What is the difference between an AI receptionist and a virtual receptionist?+
Can an AI receptionist transfer to a real person for emergencies?+
How long does it take to set up an AI receptionist?+
What happens if the AI receptionist makes something up?+
- 18 U.S.C. § 2511(2)(d), federal one-party consent provision. law.cornell.edu
- Reporters Committee for Freedom of the Press. "Reporter's Recording Guide," state-by-state summary of recording consent laws. rcfp.org
- 411 Locals. "Small business owners don't answer 62% of phone calls," January 2016. Field test of 85 businesses across 58 industries over 30 days: 37.8% answered, 37.8% voicemail, 24.3% no response. 411locals.us
- Vendor pricing pages, read September 2026: Dialzara, Rosie, Goodcall, My AI Front Desk, Smith.ai, Ruby.
Updated September 7, 2026. Prices are re-checked quarterly. LeadSixty sells an AI receptionist; this guide names competitors and their prices so you can compare.
Want the phone answered without becoming the operator?
Book a free 30-minute Where AI Fits call. We look at your missed-call report, tell you honestly whether an AI receptionist would pay for itself, and if it would, build and run it for you.
Book a Where AI Fits callor tell us in writing
Tell us what you want AI to handle and we come back with a straight answer, usually the same day. If it is not a fit, we say so.