AI receptionist: what an AI receptionist service and an automated receptionist cost, and what they handle on a call

An AI receptionist is a voice agent that answers your inbound business calls, understands what the caller is asking for in normal conversation, and completes the task rather than passing it along. In practice that means greeting the caller in your business name, answering the questions you get asked every day, capturing the details of a new enquiry, transferring the call when it needs a person, and writing an appointment into your calendar while the caller is still on the line.

The important distinction is between an AI receptionist and the two things it gets confused with. An auto attendant is a menu: press one for sales, press two for service. It routes, it does not converse, and it cannot answer a question that was not anticipated by whoever built the tree. A traditional answering service is a human bureau that takes a message and sends it to you, which means the work of calling the person back is still yours. An AI receptionist sits between them, holding an open conversation like the human service and completing the transaction like software. The fuller comparison against a phone tree is on AI receptionist vs auto attendant.

The category is also sold under names that mean much the same thing. An automated receptionist, an auto receptionist, a virtual AI receptionist and an AI phone receptionist are the same product with different marketing, though vendors selling a phone system tend to use automated receptionist for what is really a voice-enabled menu. When you are comparing two products with different names, ignore the label and ask one question: does it book, or does it only route and take messages?

Published August 2026 prices run from $49 a month for a metered plan with 100 included minutes to $399 a month for a flat plan covering roughly 1,500 calls. The headline numbers are not comparable to each other, because they are not measuring the same thing, and that is where most budgeting goes wrong.

Metered plans meter minutes. RingCentral publishes $49 a month for its standalone AI Receptionist with 100 minutes included, or $39 a month as an add-on alongside a $30 RingEX Core plan, which is $69 in total. Overage is $0.50 a minute and call time is rounded up and billed in 30 second increments. At three minutes a call, 100 minutes is about 33 calls, which is roughly one call per working day. A dental practice taking 200 calls a month on that plan is buying 500 minutes of overage on top, which at $0.50 a minute is another $250.

Flat plans meter nothing. PhoneAgent.ai is $89 a month for about 150 calls, $199 for about 500 and $399 for about 1,500, and there is no per minute charge behind those numbers, so a long call and a short call cost the same and a busy month does not generate a surprise. At the $89 tier a three minute call works out at $0.59.

Human services are the reason the AI category exists at all. Ruby is $250 for 50 minutes, which is $15.00 for a single three minute call. Moneypenny and Abby Connect both start at $165 for 50 minutes, or $9.90 for that call. AnswerConnect has the cheapest published entry rate at $1.75 a minute, still $5.25 a call. The spread between an AI-native flat plan and a human bureau at typical volumes is roughly ten to one.

The two charges that catch buyers out are structural rather than hidden, and they apply to metered AI as much as to humans. The meter usually starts at the greeting, so the caller who dialed the wrong number is billable. And rounding is real: 30 second increments turn a 65 second call into 90 seconds. Neither is dishonest, but they mean your effective rate is always somewhat worse than the rate card. The full category worked through vendor by vendor is on AI receptionist pricing, and the legacy side of the market on answering service cost.

More than a phone tree and less than an experienced employee who has worked your front desk for three years. Being specific about the middle ground is the only way to judge whether it fits your business.

It answers immediately, at any hour, including the calls that arrive at 6:40pm on a Friday when the office has emptied. It greets the caller in your business name and, on any vendor operating honestly, discloses that it is an AI assistant. It answers the questions you get constantly, drawn from information you provide or from your own website: your hours, whether you are taking new clients, where to park, which insurance you accept, what a first visit costs, whether you cover a particular town.

It captures a new enquiry in a consistent shape. This is the part businesses underrate. A human receptionist on the fortieth call of the day forgets to ask the second phone number; software asks the same questions in the same order every time, which is what makes the resulting records usable. It books into the calendar during the call, which is the single biggest difference from a message-taking service, because the caller hangs up with a confirmed time rather than a promise that someone will ring back. It routes and escalates by rules you write, and it texts a confirmation afterwards.

It also handles callers in parallel. This is a capability humans structurally cannot match, because a person holds one line. When something happens that makes everyone call at once, a burst pipe in a building, a snowstorm, an ad that ran, software takes the fortieth caller as promptly as the first. For a business that loses money to busy signals, that alone can be the argument.

What it does not do is work inside your line of business software. We book into Google Calendar, Outlook, Calendly and Acuity, and that is the honest list. We do not write into a practice management system, an EHR, a property management platform or a field service dispatch board, and any vendor answering that question with an unqualified yes deserves a follow up about which specific fields they write.

On most of the volume, yes, and the cost difference is large. On a minority of calls, no, and those are usually the calls that matter most.

A person is better whenever the call requires reading someone rather than gathering facts from them. A caller who is frightened, grieving, furious or in the middle of an emergency needs a human who can hear that and change approach mid sentence. A caller with a heavy accent on a bad line is handled better by a person. So is the caller who wants something nobody anticipated, and the negotiation where the next question genuinely depends on judgment about the last answer.

Software is better at the routine majority. Answering on the first ring at 3am. Asking the same eight questions in the same order on the fortieth call as on the first. Booking during the call. Never taking a lunch break, never resigning, never needing to be covered on a Tuesday. And costing a small fraction of a salary: a full time front desk hire runs well past $40,000 a year loaded, and covering 24/7 with people takes 4.2 full time seats rather than one, because a 40 hour seat covers 40 of the 168 hours in a week.

The honest recommendation for most small businesses is not one or the other. It is AI as the default with a written escalation rule that sends the calls above to a person, whether that person is you, your on call staff member, or a small human plan kept for exactly that purpose. Businesses that skip the escalation rule get one of two failure modes: everything escalates and the human becomes the expensive default again, or nothing escalates and a call that needed a person got processed like a form. The longer version of this comparison is on AI receptionist vs human receptionist.

The good ones no longer do, and the tell has moved. Voice quality stopped being the giveaway around 2024. What gives an AI away now is behavior under pressure: an unnatural pause before it answers, missing an interruption, repeating a question the caller already answered, or failing to notice that the caller has changed the subject.

When you evaluate vendors, test that rather than the voice. Call the demo line and interrupt it mid sentence. Change your mind halfway through a booking. Give a date in an awkward format, say next Thursday rather than the 14th. Mumble a street name. Ask something slightly outside the script and see whether it improvises, admits it does not know, or invents an answer, because inventing one is the failure that costs you a customer.

On disclosure, our position is that the AI should say it is an AI. Some vendors treat this as a weakness to hide. In practice callers respond fine to a clear disclosure and badly to discovering it later, and several US states have moved toward requiring disclosure in various contexts. We would rather be plainly on the right side of that. There is more on how callers actually react in do AI receptionists sound human.

Four things, and a vendor that will not name them is a vendor to be careful with.

It cannot exercise judgment about a person. Emotionally loaded intake, family law, criminal defense, bereavement, mental health, an angry client threatening to leave: send those to a human. Not because the software will crash, but because handling them adequately is not the same as handling them well, and the first thirty seconds decide whether the person stays on the line.

It cannot write into systems it is not connected to. Our integration list is Google Calendar, Outlook, Calendly and Acuity. Not Dentrix, Eaglesoft, Open Dental, athenahealth or Epic. Not ServiceTitan, Housecall Pro or Jobber. Not Cloudbeds or Mews. It can capture everything needed and hand it off, which for most practices is enough, but it is not the same claim.

It cannot take on your regulatory exposure. PhoneAgent.ai does not sign a Business Associate Agreement, so if you need a BAA for protected health information, buy from a vendor that will sign one. There is also no such thing as a government HIPAA certification, so treat a certified badge on any vendor site as marketing rather than proof. What that actually means for a medical practice is set out on HIPAA compliant answering service.

And it cannot fix a business problem that is not really a phone problem. If callers are giving up because your prices are unclear or your first available appointment is five weeks out, answering the phone faster surfaces that sooner rather than solving it.

Six questions separate the options faster than any feature list, and asking every vendor in the same words is what makes the answers comparable.

Does it book into my calendar during the call, or does it take a message? This single question eliminates a large share of the market. What is the allowance, what is the overage rate, and what is the billing increment? A $49 plan with 100 minutes and 30 second rounding is a different product from a flat plan, and at 200 calls a month it is a different bill. Which systems does it actually write into, field by field? What happens on a call it cannot handle, and who does an urgent call reach at 2am when the first person does not answer? Is the AI disclosed to callers? And what does it cost to leave, including whether I keep my number, which under 47 CFR 52.35(a) has to complete within one business day once an accurate port request is submitted.

Then match the shape of your business. Volume decides pricing model: under about 30 calls a month a metered plan is genuinely cheaper, and above roughly 100 a flat plan almost always wins. Call type decides how much escalation you need. And your industry decides the integration question, which is where most disappointment originates.

If you are a small business weighing this for the first time, AI receptionist for small business works through it at that scale. For a side by side of the named vendors see best AI receptionist, and for the software category itself AI receptionist software. Practices and firms with specific requirements are better served by AI receptionist for medical practices, AI receptionist for dental offices or AI receptionist for law firms. If your phone system already offers one as an add-on, the arithmetic on that route is in RingCentral AI receptionist pricing.