Blog · October 26, 2026 · 16 min read

AI Receptionist: Providers, Pricing, Law

Graphic title card for the article “AI Receptionist: Providers, Pricing, Law” with a stylised motif: a telephone handset with sound waves, leading to a note with an appointment and a handover to a person.
Graphic: HumanITy

An AI receptionist is software that picks up your business's phone and talks with the caller freely instead of playing an announcement: it greets, takes down the request, answers standard questions from your knowledge base, books appointments if a calendar is connected, transfers to a human when needed, and sends you a summary afterwards. An AI phone assistant can pay off where many calls meet few hands: in the trades, medical practices, law and tax firms, and service businesses. This page is a market overview based on what the vendors say, not my own test, written for businesses in Germany and the EU: everything about the providers comes from their own websites, retrieved on 26 October 2026.

I run my business with fourteen AI employees, but I do not use an AI receptionist; my employees work with text and files. How I manage them is on AI Employees: Tasks, Costs, Getting Started.

How this overview was made: I retrieved the websites of eight providers on 26 October 2026 and used only what they state there. I did not test or rank any of them. Prices change quickly; the provider's current offer counts. The legal section describes German and EU law, because that is the framework I work under.

What an AI receptionist is

It also goes by AI phone assistant, AI answering service or voice agent. Unlike an answering machine or a "press 1" menu, it holds a conversation. Technically it is an agent with a very narrow job: it has tools such as a calendar or call transfer and decides during the call when to use them. How that differs from a chatbot and an assistant is in AI Agent, Chatbot, or Assistant?.

How it works technically

ElevenLabs and PUNKU.AI describe the same chain: the call arrives via a provider number, your phone system or call forwarding; speech recognition turns it into text; a language model works out the request, searches your knowledge base and calls tools such as calendar, CRM or transfer; speech synthesis reads the answer out. Afterwards you get a transcript and a summary. The links often come from different companies: PUNKU.AI names Twilio for telephony and OpenAI for the language model, fonio uses Twilio as its carrier. Every link processes call data.

Where these systems hit limits:

  • Names, numbers, jargon. fonio offers a glossary with pronunciation for this; only a test call shows whether yours are understood. Retell AI warns that speech recognition confuses closely related languages, such as Cantonese and Mandarin.
  • Made-up answers. A language model can phrase a plausible answer even when the knowledge base has none; Vapi's documentation names higher hallucination rates for assistants with long, conflicting instructions. Prices, deadlines and promises do not belong in its hands.
  • Emergencies. Doctolib's assistant recognizes possible emergencies by keywords, while the medical assessment stays with the practice team. I recommend that division of labor to any business.
  • Missing features. PUNKU.AI openly lists what its product cannot do, such as calendar booking or transfer to more than one number. Ask every provider for such a list.

Market overview: eight providers as stated by the vendors

The selection mixes three groups: ready-made assistants for smaller businesses in the German-speaking market, platforms for large contact centers, and international toolkits for building your own voice agents. The order follows the groups, not a rating. All details as of 26 October 2026.

Provider Target group Core features German Data and servers Price
fonio businesses, own pages for medical practices, lawyers, trades and more answer questions, book appointments, transfer, transcript by email yes servers in Europe, telephony via Twilio Solo €119 monthly or €99 billed yearly (1,000 minutes), Team €359 or €299, net
PUNKU.AI businesses, example: trades answers from the knowledge base, transfer to one number, transcript yes data in AWS Frankfurt, telephony Twilio, language model OpenAI Voice Lite €99 (500 minutes), Voice Standard €475 (2,500 minutes), net
Doctolib (with Aaron.ai) medical and therapy practices patient recognition, booking in the Doctolib calendar, summary to the team yes (offer for practices in Germany) European servers not stated; sales consultation
Parloa enterprise customer service AI agent platform with per-agent safety controls more than 140 languages, German not listed individually data stays in the region you choose, EU data residency per minute, custom
Synthflow enterprises, contact centers no-code builder, inbound and outbound calls, appointments, transfer yes, 28 languages "regional hosting" enterprise from $30,000 per year
Retell AI developers and businesses toolkit for phone agents, appointments, transfer, analytics yes no services in the EU, DPA with standard contractual clauses $0.07 to $0.31 per minute
Vapi developers and teams toolkit with free choice of transcriber, model, voice yes (transcription) EU region only on the Premier package $0.05 per minute plus models and telephony
ElevenLabs (ElevenAgents) creators and businesses of all sizes voice and chat agents, integrations incl. calendar, CRM yes stored in the US by default, EU only for enterprise Free 15 minutes, Starter $6 to Business $990 per month, model extra

Ready-made assistants. fonio, based in Austria by its own account, offers call recording with deletion options and custom data processing agreements on its largest tier. PUNKU.AI, a GmbH from Heilbronn, stores the transcript but no audio and names OpenAI (US) and Twilio (US and EU) as subprocessors.

Industry solution for practices. Doctolib says Aaron.ai is now on board. The page states no price and still invites practices to sign up as first users; availability is settled in the sales consultation.

Contact center platforms. Parloa, founded in the EU by its own account, charges per minute based on complexity; Synthflow, a GmbH from Berlin, states contracts from $30,000 a year.

Toolkits. With Retell AI, Vapi and ElevenLabs, you or a service provider assemble your own agent. Check data location closely: Retell currently operates no services in the EU, Vapi's EU region needs the Premier package, and ElevenLabs stores data in the US by default and notes that processing outside a chosen region can still occur.

What an AI receptionist costs

I am not giving an average price; the models are too different. The overview contains four kinds:

  • Monthly fee with included minutes: fonio, PUNKU.AI. PUNKU.AI explicitly states no setup fee.
  • Per-minute usage pricing: Retell AI, Vapi, Parloa.
  • Plan plus minutes plus language model: ElevenLabs.
  • Custom enterprise contract: Synthflow (from $30,000 a year according to its pricing page), Parloa (volume package with a custom pricing model) and the enterprise tiers of the toolkits (price on request).

Watch the side costs: at ElevenLabs, external telephony providers bill you directly, Vapi lists transport fees separately, fonio charges carrier costs for outbound calls. Extra minutes cost €15 per 100 at fonio, added automatically, and €39 for 150 at PUNKU.AI, which rounds every call up to the full minute. fonio and PUNKU.AI quote net prices, the toolkits charge in US dollars.

To compare flat fees and per-minute prices fairly, you need three numbers: calls per month, average call length, and the share the assistant should take.

Law and privacy: five points before you start

I'm a developer, not a lawyer, and this isn't legal advice. These are the points I would settle first, under German and EU law.

1. Recording. § 201 of the German Criminal Code (StGB) makes it an offense to record, without authorization, the privately spoken word of another person on a sound carrier. If your assistant records audio, the safe rule is the caller's prior consent; switching audio off is simpler. A transcript alone still needs a legal basis under Art. 6 GDPR and the information from point 2. More on transcripts in Transcribing Calls: What German Law Says.

2. Informing callers. Under Art. 13 GDPR you inform people "at the time when personal data are obtained", among other things about purpose, legal basis and recipients. On the phone, a short announcement that points to your privacy policy, which lists the assistant and its providers, is the obvious approach.

3. Disclosing that an AI is speaking. Art. 50(1) of the EU AI Act requires AI systems intended to interact directly with people to be designed so that those people "are informed that they are interacting with an AI system", unless that is obvious; under Art. 50(5) at the latest at the first interaction. The obligation is addressed to the provider of the system and has applied since 2 August 2026. If you build an assistant with a toolkit and run it under your own name, you may become the provider yourself; have that reviewed. In practice, the disclosure goes into the first sentence: "You are speaking with the digital assistant of Example Painters."

4. Data processing agreements. Under Art. 28 GDPR you only use processors "providing sufficient guarantees", with a contract; further subprocessors need your authorization. Ask for the contract, the subprocessor list and every location in the chain. How I decide what an AI gets to see is in AI and Privacy: What the AI Gets to See.

5. Professional secrecy. § 203 StGB makes it an offense for doctors, lawyers, notaries, tax advisors and others to disclose secrets without authorization; contributing service providers may receive them only as far as necessary and must be bound to secrecy. Health data are also special categories under Art. 9 GDPR. A real case from a tax firm is in AI Agents for Tax Firms: A Real Case.

Outbound advertising calls fall under a different area of law: § 7(2) No. 2 of the German Unfair Competition Act (UWG) requires prior express consent for advertising with an automated calling machine; more in Automating Cold Outreach: Limits for AI.

Build or buy

Buying means a ready-made assistant runs at the provider; building means assembling your own agent with a toolkit, with more control and more responsibility for telephony, models and data flows. Without a development team, you will more likely buy.

What you write down for it matters more than the provider. I manage my AI employees by a fixed pattern: a personnel file with a clearly defined job, rules in three levels (alone, by procedure, only with approval), an escalation path that sends everything uncertain to a human, and a result someone checks. Applied to the phone:

Building block For an AI employee For an AI receptionist
Job description job in the personnel file greeting, knowledge base, questions it may answer
Rules alone, by procedure, only with approval take down and answer yes; prices, promises, diagnoses, legal advice never
Escalation path uncertain cases go to a human transfer on emergency, complaint, uncertainty or on request
Report result for approval summary per call that someone reads the same day

One difference is the most important point on this page. With my employees, nothing leaves the house before I approve it; Sonja writes emails in my voice, I read them and hit send. On the phone, the assistant talks directly to the customer, so its job has to be narrower: answer, take down, handle standard questions, transfer. What would be "only with approval" in my business goes on its "never" list. Renting versus building in general is covered in AI Employee Platforms: Buy or Rent?.

Checklist: questions for the provider

  • Can I test with my own calls before signing whether our accent and jargon are understood?
  • Which companies process audio, text and model requests, and where; do I get a data processing agreement and, as a practice or firm, a confidentiality undertaking?
  • Is audio stored or only text? Can I set recording and deletion periods myself?
  • How do callers learn at the start that they are speaking with an AI?
  • What can the assistant explicitly not do, in writing?
  • How many numbers can it transfer to, what happens if nobody picks up, how does it recognize emergencies, and which calendars and CRMs are connected out of the box?
  • What does a month at my call volume cost, including telephony, model, extra minutes and tax?
  • How do I get out of the contract, and in what form do I take transcripts and settings with me?

Frequently asked questions

How much does an AI receptionist cost?

The smallest plans of fonio (billed yearly) and PUNKU.AI cost €99 net per month, as of 26 October 2026. Toolkits charge per minute: Retell AI states $0.07 to $0.31, which its price calculator builds from language model, voice, platform and telephony; at Vapi, models and telephony come on top of the $0.05 platform fee.

Is an AI receptionist GDPR compliant?

That depends on your setup, not the label: caller information, a data processing agreement, subprocessors and data location. This isn't legal advice.

Does the assistant have to say it is an AI?

In the EU, Art. 50(1) of the AI Act has required since 2 August 2026 that people are told when they interact with an AI system, unless it is obvious. The obligation is addressed to the provider; in practice, the disclosure belongs in the first sentence of the greeting.

Can an AI phone assistant record calls?

Under German law, unauthorized audio recording is an offense under § 201 StGB; get prior consent or switch audio off. A transcript alone still needs a legal basis under the GDPR. This isn't legal advice.

Can an AI receptionist book appointments?

Many can with a connected calendar, fonio and Doctolib among them; PUNKU.AI's cannot. Ask about your specific calendar.

Is there a free AI receptionist?

For trying out, yes: ElevenLabs has a free plan with 15 minutes, Retell AI a starting credit of $10, Vapi of $5. For real operation with a phone number, you pay everywhere.

Where to go from here

Before you contact a provider, write the job description: what it may answer, what it never says, when it transfers and who reads the summaries. For a desk-based employee, that is in Creating an AI Employee: How to Start. Which roles fit a trades office is in AI Agents for Tradespeople: Roles, Not Tools.

If you want your rules read by others before you start: in my German-speaking community we go through personnel files and rules together on the calls: Claude Practitioners.

Kevin Welter

Kevin Welter

Developer, IT architect, author of technical books (Kubernetes, cloud infrastructures) and speaker. Runs his business with an AI workforce of fourteen AI employees and shows solo business owners in his community how to hire their first AI employee.

More about AI employees

Your first AI employee up and running within an hour

Try it free for 7 days (in German)