The phone rings while you're halfway through a haircut, under a sink or explaining a bill to the customer in front of you. You let it go to voicemail. Half the people who reach voicemail don't leave a message, and the ones who do often want something simple: are you open on Saturday, how much is it, can I book for Thursday.
So what is an AI receptionist? It's software that answers your business phone when you can't, talks to the caller in ordinary spoken language, answers their questions from information you've given it, and books appointments or takes their details. It works at any hour, and it doesn't put anyone on hold to finish what it was doing.
That's the short version. The rest of this guide covers what happens on a call step by step, what the technology is good and bad at, how it compares with a person or an answering service, what it costs, and how to set one up so it helps rather than annoys your customers. We make an AI receptionist, Klugon, and we use it as the worked example throughout, including the things it doesn't do.
What is an AI receptionist, in plain terms
An AI receptionist is a voice assistant that sits on your phone line. When someone calls, it picks up, greets them in your business's name and holds a conversation. The caller doesn't press buttons or choose from a menu. They say what they want, in their own words, and the assistant works out what they mean.
A typical call goes something like this. A customer rings a salon and asks whether there's anything free on Saturday morning for a cut and colour. The assistant knows the salon's services and how long each one takes. It checks the diary, offers two times, and books the one the caller picks. If the caller then asks whether they can bring their dog, it answers from the salon's own notes, or says it doesn't know and takes a message.
Three things make this different from older phone technology:
- It understands free speech. "Have you got anything Saturday, like before lunch?" works as well as "I would like to book an appointment on Saturday at 10 am".
- It can act, not just talk. It can check availability, make a booking, cancel one or save the caller's details for you to follow up.
- It answers from your information. Your prices, opening hours, policies and services, not a generic script.
Most AI receptionists also answer written messages. The same assistant that picks up the phone can often reply on your website chat, WhatsApp, Instagram or Facebook Messenger, so a customer gets the same answer whichever way they ask.
What an AI receptionist is not
The term gets stretched, so it helps to be clear about what doesn't count.
It isn't a phone menu. "Press 1 for opening hours, press 2 to book" is an interactive voice response (IVR) system. It routes callers along fixed paths and frustrates anyone whose question doesn't fit. An AI receptionist has no menu.
It isn't voicemail with transcription. Some phone systems now turn voicemails into text and email them to you. That's useful, but the caller still hangs up without an answer or a booking.
It isn't a human answering service. Answering services and virtual receptionists are real people, usually in a call centre, who answer on your behalf. Some now add AI to their tools, but you are paying for human time. We compare the two later.
It isn't a general-purpose chatbot. An AI receptionist is set up for one business. It should only answer from what you've taught it and should say when it doesn't know, rather than inventing a policy or a price.
It isn't a replacement for every part of a receptionist's job. A front-desk person greets people who walk in, deals with deliveries, calms an upset customer face to face and notices when something is off. An AI receptionist handles one part of that work: the phone and the inbox.
How an AI receptionist works on a call
You don't need the technical detail to use one, but knowing the moving parts helps you judge what it will and won't cope with. Our companion guide, can AI answer phone calls, goes further into the technology itself.
Getting the call to the AI
A call can only reach the assistant if your phone line points at it. There are usually three ways to do that:
- A new number. The provider gives you a phone number and you publish it. Simple, but you have to change your website, your Google listing and anything printed.
- Call forwarding from your existing number. You keep the number customers already know and set your carrier to forward calls to the AI's number. You can forward every call, or only the ones you don't pick up, or only out of hours, depending on what your carrier supports. Our post on keeping your number with call forwarding walks through the options.
- A SIP trunk. If your business already runs an internet phone system, it can send calls straight to the AI over SIP. This is the route for businesses with an existing phone system they want to keep.
With Klugon, you can buy a number, forward your existing number to a Klugon number, or bring your own SIP trunk on the voice Business plan or higher. Numbers cost $9 a month each. Klugon doesn't port numbers, so "keeping your number" always means forwarding or SIP.
Listening, understanding, acting, replying
Once the call connects, the assistant runs a loop for every turn of the conversation:
- It listens. Speech recognition turns what the caller says into something the language model can work with.
- It understands. A language model works out what the caller wants, using your business information as context: services, prices, hours, policies and common questions.
- It acts. If the caller wants to book, it checks your availability and makes the booking. If they want something it can't do, it can save their details with a note for your team.
- It replies. A synthetic voice speaks the answer back.
This happens fast enough to feel like a conversation, though it's not identical to talking with a person. There's a short gap before each reply, and how short depends on the provider, the call quality and how much the assistant has to look up. Try a demo yourself and judge whether the pace feels natural to you.
Good systems also handle interruptions. If the caller starts talking while the assistant is speaking, the assistant stops and listens. Klugon does this: as soon as the caller speaks, the AI's audio stops.
Where its answers come from
An AI receptionist doesn't know anything about your business until you tell it. Its knowledge usually comes from a mix of:
- Your services and prices, entered as a list with durations, so it can book the right length of slot.
- Your opening hours and closed dates.
- Documents such as a price list, menu, policies or FAQ sheet.
- Your website, read once or scanned for key pages.
- Free text where you write answers to questions people ask.
In Klugon, you can upload PDF, Word, text, Markdown and CSV files up to 20 MB each, paste text, import a single web page, or scan your website (the homepage plus up to seven key pages, such as contact and about). Scanned image PDFs aren't read, so a photographed menu needs typing up.
When a caller asks something, the assistant searches that knowledge for the relevant part and answers from it. When nothing relevant comes back, a well-set-up assistant says it doesn't know and offers to take a message. That behaviour depends heavily on the instructions you give it, which is why setup matters more than the technology.
What happens after the call
After the caller hangs up, the record of the call is what you work from. What you get varies a lot between providers. Ask each one exactly what is saved: a summary, a transcript of both sides, a recording, or just a log entry.
In Klugon, every call appears in your inbox alongside your chats and messages, with its duration and the messages from the conversation. Bookings it made appear in your bookings list, and callers it took details from appear as leads. Klugon doesn't record calls, and the saved conversation may not include every word the caller said, so don't plan around having a full two-sided transcript.
What an AI receptionist does well
These are the jobs where an AI receptionist earns its place.
Answering every call, at any hour
The most basic benefit is that the phone gets answered. Evenings, weekends, bank holidays, lunch breaks and the ten minutes when everyone is busy at once. People who ring a business out of hours usually want something routine, and an answer now beats a call back tomorrow, when they may have booked elsewhere.
Klugon answers whenever the voice plan is live and a number is routed to it. There's no setting that stops it answering outside your opening hours, so if you only want it after hours, set that up with your carrier's forwarding rules.
Booking, cancelling and checking availability
Booking is where most of the value is for appointment-based businesses. A caller who books on the first call is a customer. A caller who leaves a message is a maybe.
Klugon books into its own built-in calendar, using your opening hours, service durations, staff, buffer times, minimum notice and how far ahead you allow bookings. On the voice and social Business plans and above, it can also check Google, Outlook or Apple calendars for busy times, after you've connected them. On a call it can book and cancel, but not reschedule: a caller who wants to move an appointment would need to cancel and rebook, or you'd change it in the dashboard. In chat, it can reschedule. More detail is on our AI appointment booking page.
The same questions, answered the same way
Every business has questions it answers twenty times a week. Where are you, do you take cards, is there parking, how long does a colour take, do you do gluten-free. An AI receptionist answers these consistently, at any hour, without sighing at the twentieth asking.
Taking details when it can't help
When the caller wants something the assistant can't do, it can still collect their name, number and what they need, and save it so nobody is lost. In Klugon this is saved as a lead with a note. On the Business plan and above, you can have new leads posted to Slack or sent through Zapier, so someone sees them quickly.
Many conversations at once
A person can take one call at a time. An AI receptionist can usually handle more than one, so a Saturday-morning rush doesn't produce a queue of engaged tones. How many at once varies by provider, and few publish a hard number. If you get sharp peaks, ask.
What an AI receptionist can't do, or shouldn't
Every honest description of an AI receptionist needs this section. Some limits come from the technology, and some from the particular product you choose.
Limits of the technology
Judgement in unusual situations. The assistant follows your instructions and your information. When a call doesn't fit either, a person reads the situation better: a distressed caller, a complaint with history behind it, a request that's technically allowed but obviously a bad idea.
Emergencies and safety. It shouldn't triage a medical problem, assess a gas smell or advise on a legal deadline. Your instructions should tell it to direct those callers to the right emergency service or to a person, and you should test that it does.
Hard audio. Speech recognition copes with most accents and ordinary background noise, but a bad mobile line, a busy street or a caller spelling a foreign surname can still trip it up. Names, email addresses and postcodes are where mistakes cluster. A good setup has the assistant read back what it heard.
Saying things that aren't true. Language models can produce confident answers that are wrong, especially when asked something your knowledge doesn't cover. Good instructions and good knowledge reduce this a lot, but nobody can promise it never happens. Read your conversations in the first weeks and fix the gaps.
Limits of particular products
Features vary widely, and marketing pages sometimes promise more than the product does. Here is where Klugon stands today, so you can hold other providers to the same list:
- No live call transfer. Klugon can't put a caller through to a person. When a call needs a human, it saves the caller's details with a note and, on Business and above, notifies Slack or Zapier. Your team calls back. If you need warm transfers, look elsewhere or keep a person on the line for those calls.
- No outbound calls. It answers calls. It doesn't ring customers back or make reminder calls.
- No text messages. No SMS confirmations, reminders or missed-call text-backs.
- No automatic confirmations or reminders. The caller hears a booking reference on the call. Nothing is emailed or messaged afterwards.
- No payments on a call.
- No call recordings.
- No rescheduling on calls, as above.
None of this makes it useless. It makes it a specific tool: an assistant that answers, informs, books and takes details, with a person behind it for the rest.
A sample call
A sample call. The business and caller are made up, and this is written to show the flow, not a recording.
A few things to notice. The greeting says it's an AI. The assistant reads back the name and number. It answers the parking question from the practice's own notes. And the warning at the end isn't the AI's own judgement: it's a line the practice wrote into its instructions for any call about back pain, which is the right way to handle anything near clinical territory. Our guide on what to teach your AI receptionist covers how to write rules like that.
Beyond the phone: chat, WhatsApp and social messages
Many customers would rather type than call, especially younger ones and anyone at work. Most AI receptionist products now cover written channels too, with the same knowledge and booking rules as the phone.
In Klugon, the voice agent, the social agent (WhatsApp, Instagram and Messenger) and the website chatbot are separate products, each with its own plan, sharing one knowledge base. All conversations arrive in one inbox, and you can reply yourself on WhatsApp, Instagram and Messenger within Meta's 24-hour window. The website widget is a script tag you paste into any site. WhatsApp, Instagram and Messenger need some setup in Meta's tools first, and Instagram needs Meta's app review.
One practical limit is worth knowing for messaging: if a customer sends a photo or a voice note on WhatsApp, the assistant only sees that an image or voice message arrived. It can't look at the photo or listen to the voice note, so it can't quote from a picture of a broken boiler. See our WhatsApp AI chatbot page for the details.
AI receptionist vs a human receptionist vs an answering service
There are three common ways to make sure your phone gets answered. Each is the right answer for someone.
| AI receptionist | In-house receptionist | Answering service | |
|---|---|---|---|
| Hours | Any hour the line is routed to it | Their working hours | Often 24/7 |
| Calls at once | Usually several | One | Depends on staffing |
| Booking | Directly, if connected to your calendar | Yes | Sometimes, often an add-on |
| Judgement and warmth | Fine for routine calls | Best | Good |
| Transferring a caller to you | Depends on the product (Klugon: no) | Yes | Usually yes |
| Messaging channels | Often included or available | If they have time | Usually not |
| Typical cost | Monthly plan | A salary plus employer costs | Per minute or per call |
An in-house receptionist is best at judgement, warmth and the unexpected, and can do other work between calls. They take one call at a time, go home in the evening and cost a salary. For many small businesses, the receptionist is also the person who does the books, orders stock and opens up, which is why the phone gets missed.
An answering service gives you trained people without hiring. It suits calls that need human judgement, such as legal intake or sensitive enquiries. You usually pay per minute or per call, so costs rise with volume, and booking may be extra.
An AI receptionist answers at any hour, handles more than one caller at a time and costs a fixed monthly amount up to an allowance. It's strongest at routine work: questions, bookings and taking details. It still needs a person behind it.
Many businesses combine them: the AI takes routine calls and out-of-hours calls, and staff deal with the calls it hands over. We go deeper on the trade-offs in AI receptionist vs human receptionist.
What an AI receptionist costs
Pricing comes in a few shapes, and the shape matters more than the headline number.
The main pricing models
- Monthly plan with a minute allowance. You pay a fixed amount for a set number of voice minutes, and extra minutes are charged on top.
- Per call. Every answered call costs a set amount. Simple to understand, expensive if you get many short calls.
- Per minute. Common for human answering services. Long calls cost more.
- Per unique caller. Repeat calls from the same person in a month count once.
Check what sits outside the headline price: the phone number, booking, messaging channels, extra languages, integrations and what one extra minute or call costs. Our guide to AI receptionist cost compares real vendors' pricing models side by side.
Klugon's voice pricing
Klugon's prices are in US dollars:
| Voice plan | Monthly | Yearly (per month) | Voice minutes a month | Phone numbers |
|---|---|---|---|---|
| Starter | $99 | $79 | 1,000 | 1 |
| Business | $299 | $239 | 3,600 | 3 |
| Scale | $499 | $399 | 6,000 | Unlimited |
Phone numbers are $9 a month each. Extra voice minutes are $15 for each started block of 100. There's a 7-day free trial; a card is required and is charged when the trial ends unless you cancel. During the trial you test the assistant in your browser, and you can add a phone number once the trial ends. Integrations such as Google Calendar, Slack and Zapier need the Business plan or above. Full details are on the pricing page.
A worked example
Here's a rough comparison with stated assumptions. Change the numbers to fit your business.
Assumptions:
- 20 calls a day, 22 working days a month: 440 calls.
- An average call of 3 minutes: 1,320 minutes a month.
- One forwarded number on Klugon's Starter plan, billed monthly.
AI receptionist: $99 for the plan plus $9 for the number is $108. The 1,000 included minutes leave 320 minutes over. That's four started blocks of 100 at $15, so $60. The total is about $168 a month. If you regularly went well over, the Business plan's 3,600 minutes would become the better deal at some point, so check both.
A person: according to the US Bureau of Labor Statistics, the median pay for receptionists was $38,010 a year in May 2025, or $18.27 an hour. That's about $3,170 a month before employer taxes, benefits, holidays and sick cover. Our guide on how much a receptionist costs breaks that down further.
This isn't a like-for-like comparison, and it shouldn't be read as one. A person does far more than answer the phone. The fairer question for most small businesses is not "AI or a receptionist?" but "what happens to the calls nobody can answer now?" For those calls, the AI costs a small fraction of a person and covers hours no person would.
Who uses an AI receptionist
Any business that misses calls, or spends hours on the same questions, can use one. The fit is strongest where calls are frequent, routine and lead to a booking.
Restaurants and cafés
Calls peak exactly when the team is busiest: the dinner rush. An assistant takes table bookings, answers questions about hours, menus and allergens, and handles takeaway questions. Allergy questions deserve care: the assistant should answer only from your written allergen information and send anything uncertain to a person. More on our page for restaurants.
Salons, barbers and spas
The person who could answer the phone has scissors or a brush in their hand. An assistant books appointments against each stylist's real availability, with the right service length, and answers price and product questions. See AI receptionist for salons.
Clinics and practices
An assistant can handle appointment requests, opening hours and routine questions like parking and what to bring. It should never triage symptoms or give clinical advice; anything clinical or urgent goes to a person or to emergency services. Our clinics page covers what to keep away from the AI.
Trades and home services
A plumber on a job can't answer the phone with wet hands. An assistant answers, takes the job details and address, and books a visit or quote slot, so the caller doesn't ring the next name on the list. It can't send emergencies to your mobile, so emergency call-outs need their own plan. See trades.
Professional services, gyms and studios
Law firms, accountants and agencies use an assistant to take details from new enquiries and book a first consultation, with clear rules that it gives no legal or financial advice. Gyms and studios use one for membership questions, class times and booking a trial session.
Disclosure, privacy and the law
Rules about AI on the phone are developing quickly. This section isn't legal advice, but it covers what a small business should think about.
Telling callers they're talking to an AI
We think you should tell callers, early and plainly, and most callers prefer it. In the EU, Article 50 of the AI Act requires AI systems that interact directly with people to be designed so those people are informed they're dealing with an AI, unless it's obvious from the context. According to the Act's text, this applies from 2 August 2026. Check the current position for your country, as the EU's timetable has been subject to proposed changes.
In Klugon, the assistant is instructed that it's the business's AI phone assistant, but the default greeting ("Hello! Thank you for calling..., how can I help you today?") doesn't say so. You can write your own greeting, and we recommend you include something like "I'm the practice's AI assistant", as in the sample call above.
Outbound calls are a different matter
In the US, the FCC ruled in February 2024 that AI-generated voices in robocalls count as "artificial" voices under the Telephone Consumer Protection Act, which means consent rules apply to outgoing calls that use them. That ruling is about calls a business makes, not calls it answers. Klugon only answers calls, but if you're considering a product that rings customers, take advice first.
Personal data
Every call involves personal data: names, numbers, sometimes health or financial details. Under the GDPR and UK GDPR, you're responsible for how it's handled, including by the providers you use. The UK ICO's guidance on AI and data protection is a sensible starting point.
Ask any provider where data is processed, which sub-processors they use, how long conversations are kept and how you delete them. Klugon is operated by a German company under the GDPR, with transfers to its US sub-processors covered by standard contractual clauses, as listed in its privacy policy. You can export conversations, leads and customers as CSV and delete records yourself; full erasure of an account is by email request. Klugon doesn't record calls. It holds no HIPAA, SOC 2 or ISO 27001 certification, which matters if your sector requires one.
How to choose an AI receptionist
Ask these questions before you sign up for anything, including Klugon.
- Does it sound right to you? Call a demo or test it yourself. Listen for the pause before each answer, how it copes when you interrupt, and how it handles you changing your mind.
- Can it book where your bookings live? If you already run a booking system, find out whether the assistant books into that or into its own calendar, and what happens to double-bookings.
- What happens when it needs a person? Live transfer, a message to your team, or both? How quickly will someone see the message, and on which tool?
- What does it send afterwards? Confirmations, reminders and summaries vary widely. Don't assume.
- Which languages, and how well? Test each language you need with a real call, using real names and addresses.
- What does a busy month cost? Work out your bill at your actual call volume, including overage and the phone number.
- What's kept, and where? Recordings, transcripts, retention and deletion.
- Can you try it first? A trial lets you test with your own business information, which is the only test that matters.
Our page on Klugon's AI phone receptionist answers these questions for our own product.
Setup checklist
A usable setup is mostly writing, not technology. Work through this list:
- List the 20 questions you're asked most, with the exact answers you'd give.
- Gather services, prices, durations and opening hours in one place. Add closed dates.
- Write the rules: what it may do alone (answer, book, cancel), what it must never do (advise on health, law or money; promise refunds; quote for custom work), and when it must take details for a person.
- Write the greeting. Say who is answering, and that it's an AI assistant.
- Set up booking: services, staff, buffer times, minimum notice, how far ahead.
- Decide who sees handovers and how. In Klugon that's the inbox, plus Slack or Zapier on Business and above. Make sure someone actually watches it.
- Test it hard. Call it yourself. Mumble. Interrupt. Ask the awkward questions. Ask it something it shouldn't know and check it says so.
- Route calls gradually. Start by forwarding only unanswered or out-of-hours calls. Widen it once you trust it.
- Read every conversation in the first week and fill the gaps in its knowledge.
How to tell whether it's working
Give it a month and look at numbers, not impressions.
- Calls answered that would have been missed. Compare with your phone bill or missed-call log from before.
- Bookings made by the assistant, and how many of those people turned up.
- Handovers. How many calls needed a person, and why. A high number usually means a gap in its knowledge or rules.
- Questions it couldn't answer. Klugon's insights page lists "questions without a good answer" and the most asked questions of the last 30 days, which is a ready-made to-do list.
- Complaints. Any caller who says the assistant got something wrong is worth reading in full.
If after a month it books appointments and takes details you'd otherwise have lost, and the handovers are reasonable, it's working. If callers keep asking for a person, look at what they wanted and either teach it or route those calls elsewhere.
Frequently asked questions
Will callers know they're talking to an AI?
They should. Voices are natural enough now that some callers won't be sure, which is a good reason to say so in the greeting. In the EU, the AI Act's transparency rule expects people to be told unless it's obvious. In Klugon you write the greeting, and the default doesn't mention AI, so add it.
Can an AI receptionist transfer calls to a person?
Some can. Klugon can't: when a caller needs a person, it saves their details with a note and, on Business and above, sends an alert to Slack or Zapier so your team can call back. If live transfer matters to you, check for it specifically before you buy.
Do I need a new phone number?
No. Most businesses forward their existing number to the AI, either for every call or only for calls they miss. With Klugon you can forward to a Klugon number or connect a SIP trunk. Klugon doesn't port numbers.
What happens if the AI doesn't know the answer?
A well-set-up assistant says it doesn't know, takes the caller's details and passes the question on. You then add the answer to its knowledge. If it guesses instead, tighten its instructions.
Which languages can an AI receptionist speak?
It depends on the provider. Klugon answers calls in English, German, French, Spanish, Italian, Portuguese, Dutch, Japanese and Arabic. You choose the language for each assistant and should write the greeting in that language.
Does it record calls?
Some do. Klugon doesn't. Calls appear in your inbox with their duration and the conversation, but there's no audio recording. If you use a product that records, check your obligations under data protection law.
Is an AI receptionist worth it for a small business?
If you regularly miss calls, take bookings by phone or spend hours on the same questions, usually yes. If you get a handful of calls a week and always pick up, you probably don't need one yet.
Can I try one before paying?
Most providers offer a trial. Klugon's is 7 days, with a card required and no charge if you cancel before it ends. You test the assistant in the browser during the trial, and can add a phone number when it ends.
An AI receptionist won't run your front desk for you. It will answer the calls nobody's free to take, give callers a straight answer, and book them in or take their details. If that's the gap in your business, Klugon's AI receptionist is one way to fill it, and you'll know within a week of testing whether it suits you.