What an AI Receptionist Really Does, From First Ring to Booked Slot

An AI receptionist is software that picks up your business phone, talks with the caller in a natural voice, finds out who they are and what they need, answers routine questions from information you gave it, books a slot in your calendar when the caller is ready, and passes everything else to a person with a written summary. It does not replace your front desk's judgment. It replaces the phone that rings while nobody has a free hand.

The clearest way to see what that means is to follow one call from the first ring to the booked appointment, stage by stage, including the parts the AI should hand to your team. After that: medical offices, pricing, free plans and a setup checklist.

What does an AI receptionist actually do?

Strip away the marketing and an AI receptionist does six jobs on every call:

1. Answers within a ring or two, at any hour, even when three people call at once.
2. Identifies the reason for the call: a new enquiry, an existing customer, a supplier, a sales pitch or an emergency.
3. Answers routine questions from a knowledge base you control: opening hours, service area, what you do and do not do, how booking works.
4. Qualifies new enquiries with the questions you chose, so the job lands with the right person and the right information.
5. Books a call, consultation or visit into a real calendar slot.
6. Hands over to a human when the call needs one, and leaves a transcript and a summary either way.

What it should not do is invent answers. If a price, a diagnosis or a promise is not in the information you gave it, a well configured AI front desk says someone from the team will confirm, and logs the question.

One inbound call, from first ring to booked slot

To make this concrete, take an illustrative example (not a real customer): a small physiotherapy clinic. It is 6:40 on a weekday evening, the front desk has gone home, and someone with a sore shoulder calls the number they found online.

Stage 1: The ring

First, the call has to reach the AI. There are three common setups:

• AI first. The AI answers every call and transfers the ones that need a person.
• Overflow. Your own phones ring first. If nobody picks up, the call moves to the AI.
• After hours only. A schedule decides when the AI takes over.

With overflow, timing matters. In Twilio's call routing instructions, for example, the time spent ringing a phone before giving up defaults to 30 seconds (it can be set anywhere from 5 to 600 seconds), and Twilio adds a five second buffer on top. If nobody answers, the attempt is marked "no-answer" and the call can move on to the next step, such as the AI.

Set that number on purpose: thirty seconds of ringing feels long, and some callers hang up before the AI gets its chance. In our example, the clinic sends calls straight to the AI after closing, so this caller is answered on the first ring.

Stage 2: The greeting and the AI disclosure

The first sentence matters for trust and for the law. A good greeting names the business, says plainly that this is an AI assistant, and offers a route to a person:

"Thanks for calling [clinic name]. You're speaking with the clinic's AI assistant. I can answer questions, book an appointment or take a message for the team. How can I help?"

In the EU this is not optional. Article 50 of the AI Act, which applies from 2 August 2026, requires AI systems that interact directly with people to be built so those people are told they are dealing with an AI, unless that is obvious. The information has to be clear and has to come at the latest at the first interaction. If you record calls, say so here too, following the rules where you and your callers are.

Stage 3: Listening without interrupting

The caller says: "Hi, yeah, I hurt my shoulder lifting a box last week and it's still... I can't really lift my arm above my head."

Behind the scenes, the system turns speech into text and decides when the caller has finished talking. OpenAI's documentation for its real-time voice API describes two approaches. One waits for a set stretch of silence, where shorter settings respond faster but are more likely to cut in. The other, called semantic detection, estimates from the words themselves whether the person is done, so it waits longer after "and it's still..." than after a complete sentence.

Callers notice this most. An AI that jumps in mid sentence feels rude; one that pauses too long feels broken. Test a vendor by calling it yourself with pauses, background noise and half finished sentences.

Stage 4: Working out why they are calling

From that first sentence, the AI classifies the call as a new patient with a new problem, and that decision drives everything that follows. An existing patient, a supplier, a sales call and a person in acute distress each need a different path. The categories, and what happens in each, come from you.

Stage 5: Qualifying with your questions

Next, the AI asks the handful of questions the clinic chose, one at a time:

• Is this your first visit with us?
• When did it start, and has anything made it worse?
• Is there a referral or insurance detail we should note?
• Which location suits you, and do mornings or evenings work better?

Three to five questions is plenty. Ask only what changes who takes the job or which slot you book, and collect the rest at the appointment.

Stage 6: Checking the calendar and booking

The caller prefers evenings, and the AI checks real availability instead of guessing. Google Calendar's API, for instance, has a free/busy query that returns the time ranges during which each calendar should be regarded as busy, for several calendars in one request. That matters when "the next free physio" means checking three diaries.

The AI offers two concrete options ("Thursday at 5:30 or Monday at 6:00?"), books the one the caller picks, reads it back, and confirms the name and mobile number. Two specific options keep the call shorter than an open "When suits you?"

Stage 7: The wrap-up your team sees

Next morning, the front desk finds the booking in the calendar with a short summary: who called, the problem in one line, the answers to the qualifying questions, the slot and a link to the transcript. Nobody has to replay a recording.

What an AI front desk handles, and what it hands to a human

The handover rules decide whether an AI receptionist helps or hurts. Write them down before you switch it on. A sensible starting split:

There are two ways to hand over. A warm transfer puts the caller through to a person while they are still on the line, ideally with the summary on screen. A callback task ends the call politely and drops a task with the transcript into your team's queue. Use transfers in opening hours and callbacks after hours, and only promise a callback time your team can keep.

AI receptionist for a medical office: the extra rules

Health practices get a lot out of an AI receptionist, because so much of their phone traffic is booking. They also carry the most rules.

In the United States, HIPAA allows a covered entity to let a business associate create, receive, maintain or transmit protected health information on its behalf only if it obtains satisfactory assurance that the information will be safeguarded, and that assurance must be documented in a written contract or other written agreement. An AI that hears a patient's name, number and symptoms and stores the transcript does exactly that, so get the vendor's business associate agreement signed before the first patient call.

Beyond the contract, three practical rules apply to any health practice:

• No medical advice. The AI collects symptoms to route and book. It never diagnoses or tells someone whether they need care.
• A fixed emergency script. If a caller describes something urgent, the AI gives one instruction you wrote (for example, to hang up and call the local emergency number) and alerts staff. It does not try to triage.
• Only the minimum. Ask what you need to book the right appointment with the right clinician, not a full history.

In the EU, the AI disclosure rule applies here too. Health details are highly sensitive, so check where the vendor stores recordings and transcripts, and for how long.

AI receptionist pricing: what you are actually paying for

Prices vary between vendors and change often, so here is how the money breaks down instead. Every AI receptionist bill has three layers.

1. The phone line. This is the cheap part. On Twilio's published US pay-as-you-go rates, receiving a call on a local number costs $0.0085 per minute and the number itself $1.15 a month. Recording adds $0.0025 per minute and Twilio's transcription $0.05 per minute. For example, 200 calls a month at four minutes each is 800 minutes: about $6.80 in inbound minutes, or roughly $50 once the number, recording and transcription are added.

2. The AI. Speech recognition, the language model that decides what to say, and the synthetic voice. Vendors usually fold this into their own per minute or per call rate.

3. The product around it. Calendar and CRM connections, handover logic, the app where your team reads summaries, setup and support. This is where vendors differ most.

The common pricing models look like this:

Compare cost per booked appointment, not price per minute, and set it against what a booking costs you today in staff time, or in callers who hit voicemail and never tried again.

AI receptionist app, free plans and trials: what to check

Most products come with an AI receptionist app or web dashboard where you read transcripts, edit the knowledge base, see bookings and pick up handovers from your phone. Judge the app as hard as the voice: if your team cannot see a summary soon after the call, handovers fail.

Searching for an AI receptionist free option usually turns up one of three things: a free trial with a time or minute limit, a free tier capped at a small number of calls, or a do-it-yourself build on developer platforms where you pay only for usage. All three are fine for testing. Before real customers call one, check:

• Does it book into the calendar you actually use, or only take messages?
• Can it transfer a live call to a person, and to whom?
• What happens when you pass the free limit: do calls fail, go to voicemail or start billing?
• Where are recordings and transcripts stored, and can you delete them?
• For US health practices: will the vendor sign the written agreement HIPAA requires?

Setup checklist: what to give your AI receptionist before the first call

The voice is the easy part. Call quality depends on what you write down first. Before going live, prepare:

1. Business basics: hours, locations, service area, parking and payment methods.
2. What you do and do not do, in plain sentences. The "do not" list prevents bad bookings.
3. Three to five qualifying questions per type of enquiry, in the order you want them asked.
4. Booking rules: appointment types and lengths, buffers between them, which staff take which jobs, and how far ahead people can book.
5. Handover rules from the table above, with names, numbers and hours for transfers.
6. The greeting, including the AI disclosure and any recording notice.
7. An emergency instruction, if your business ever gets urgent calls.
8. Ten test calls made by you and your staff: a clear booking, a vague caller, a complaint, someone asking for a person, a strong accent, a noisy line from a car, a price question, an after-hours emergency, an existing customer and a wrong number.
9. A review routine: read the first 50 transcripts yourself, then a sample every week. Every wrong answer points to a missing line in your knowledge base.

Calls are one channel: the same flow for forms and messages

The pattern above (answer at once, qualify, book, hand over with a summary) is not specific to phone calls. The same enquiry often arrives as a lead ad, a web form or an Instagram message, and those sit unanswered after hours just as easily.

Tools such as Konverto apply that pattern to written enquiries: they pick up leads from Facebook and Instagram lead ads, website forms and website chat, reply typically within a minute by SMS and email, ask your qualifying questions and book a call or visit into your calendar. Konverto also hands the conversation to a person when a lead asks for one or a call-back, or raises a complaint.

Whatever the channel, the test is the same as on the phone: how fast does a new enquiry get a useful answer, and how cleanly does it reach a person when the AI should step aside?