An AI receptionist answers your business phone, handles routine requests such as hours, directions, simple questions and booking, and takes a clean message for the rest. It works well for predictable calls when it runs on information you have approved. Emergencies, complaints, money disputes and anything sensitive should go to a person, so write that handoff list before you choose a product.
An AI receptionist is worth considering when your phone rings for the same few reasons and nobody is free to answer. It is a poor fit when most of your calls need judgment, or when one bad answer could cost you a customer or cause real harm.
Picture a three person plumbing company on a Tuesday evening. One caller wants a quote appointment next week, which the AI can book. The next caller has water coming through a ceiling, and that call needs to reach the person on duty right away. The setup work is mostly about telling those two calls apart.
What does an AI receptionist actually do?
It holds a spoken conversation with the caller, answers from the information you give it, and then takes an action such as booking, messaging or transferring. Older phone menus make the caller press buttons. A voice AI lets them say what they need in their own words.
These are the jobs it usually handles well:
- Basic questions. Hours, location, service area, parking and what you do or do not offer.
- Booking. Finding an open slot and placing it on your calendar, if the two systems are connected and tested.
- Message taking. Name, number, reason for the call and urgency, sent to the right person by text or email.
- Routing. Sending sales, billing and existing customers to different people or inboxes.
- After hours coverage. Answering when the shop is closed, so callers get more than a voicemail greeting.
Its main weakness is confident mistakes. Voice AI can misunderstand a name or an accent, and it can state something plausible that is simply wrong. That is why it should answer from a short approved source, rather than improvising. Our explainer on what an AI agent is covers why these systems need limits.
Which calls should always go to a person?
Send a call to a person whenever being wrong would be expensive, unsafe or hard to undo. Your own list depends on your trade, but most businesses can start with this one.
- Emergencies and safety issues, such as a gas smell, flooding or a medical concern. The AI should give the instruction you approved and connect the caller straight away.
- Complaints, or any caller who sounds upset.
- Billing disputes, refunds and any request to change a payment.
- Custom quotes and promises about price or timing that you have not written down.
- Legal, medical or insurance questions.
- Anyone who asks for a person. This one should work on the first request, with no argument.
Also decide what happens when nobody picks up. A transfer that rings out is worse than an honest message. Set a time limit, then have the AI say when someone will call back and send the details to a phone that a person actually watches.
How much does an AI receptionist cost?
Published paid plans start at about $79 to $150 a month, but vendors count usage differently, so compare them using your own call volume. These are published prices as of November 2026, and they change, so check before you decide.
- Per customer. Goodcall charges $79 a month for 100 unique customers on its Starter plan, which it estimates at about 150 calls, and it does not meter minutes.
- Per call. Smith.ai lists a free plan with 25 calls, then Pro from $150 a month for 75 calls. It says complex calls can transfer to a live receptionist, so ask how those transfers are billed.
- Per minute. Developer platforms such as Retell advertise $0.07 to $0.31 a minute, with phone service and add ons charged on top. Someone also has to build and maintain the agent.
Here is the comparison with made up numbers, as a planning assumption. Say you get 150 calls a month, averaging three minutes each. Smith.ai's matching Pro tier lists $1.80 a call, or $270 a month. Goodcall's Starter plan would be $79 if those calls came from 100 or fewer customers. On a per minute platform, 450 minutes works out to about $32 to $140 before phone charges and build time.
The sticker price is not the only comparison. A plan with live human backup may be worth more to a business whose calls often need a person. Long calls favor plans that do not meter minutes, while a few short calls favor pay as you go.
How do you set up an AI receptionist?
Start with a small approved knowledge base and a narrow job, test it hard, and widen it only after real calls prove it works. A careful first version often covers after hours calls only.
- Review your last 50 to 100 calls, or ask whoever answers the phone to keep a tally for a week. Group them by reason.
- Pick the two or three reasons that are most common and least risky. Those are the first job.
- Write the approved answers in one document: hours, services, service area, policies and what it must never quote.
- Write the handoff list, with a name, a number and a backup for each case.
- Connect the calendar and the place messages should land, such as your CRM or shared inbox. Test that a booking really appears and a message really arrives.
- Call it yourself with awkward requests: background noise, a heavy accent, a changed mind, a request for a person, and a question it should refuse.
- Start with one time window, such as nights and weekends, with a person reviewing every transcript.
- Each week, correct wrong answers, add missing ones, and only then expand the hours.
Track a few simple numbers: calls answered, calls transferred, bookings created, and how many transcripts contained a mistake. Also read some transcripts by hand, because a call can count as completed even though the caller gave up.
What rules should you check before going live?
Check three things with a qualified professional: recording, disclosure and outbound calls. We can describe the considerations, but a lawyer should confirm what applies to your business and your states.
- Recording and transcripts. Consent rules differ by state, and one law firm's overview notes that the stricter standard generally applies when a call crosses state lines. Many businesses address this with a clear notice at the start of the call.
- Telling callers it is AI. Saying so up front is a sensible habit that also sets expectations. Ask your lawyer whether your state or industry has specific requirements.
- Outbound calls and follow ups. In 2024 the FCC ruled that AI generated voices count as artificial voices under the Telephone Consumer Protection Act. Wiley's summary covers the consent and disclosure points. Answering a call and placing one are different situations, so review any automated callback before you turn it on.
- Sensitive information. If callers share health, financial or similar details, find out where recordings are stored, who can see them and how long they are kept.
When is an AI receptionist the wrong choice?
It is the wrong choice when your calls are mostly complex or emotional, or when your phone is not actually the problem. There are simpler options, and one of them may fit better.
- A good voicemail and a fast callback. This can be enough if call volume is low and customers will wait a little.
- A human answering service. It can cost more per call, but it may suit businesses where callers are often stressed or need judgment.
- Online booking. If most calls are just to book, a booking link might remove many of them. Calls that need a conversation would still need an answer.
- A person at the desk. If the phone is your main sales channel and customers value that relationship, a person is often worth the cost.
If you want help working out which calls could safely be automated, our AI development page explains how we approach this kind of build, and our page for local businesses shows where it fits. You can also talk to us. If a better voicemail would solve it, we will say so.
