AI phone assistants and WhatsApp: the 2026 rules.

At a glance

• In Switzerland, an AI assistant on the phone or on WhatsApp is allowed. There is no dedicated law: the FADP applies.
• It must say it’s an AI straight away. If you have customers in the EU, the AI Act requires this too from 2 August 2026.
• If it transcribes and summarises calls, it says so first. The exception for orders and bookings only covers keeping a record as evidence.
• On WhatsApp Business, a bot that serves your own customers is allowed. Meta’s crackdown was aimed at general-purpose assistants.

It’s 19:12, the garage is closed and a customer calls about changing her tyres. An AI phone assistant picks up: “Good evening, this is the garage’s AI assistant. How can I help?” A minute later, Thursday’s appointment is in the calendar and the owner has the summary on his phone.

This is an example, the scene we play in our demos. It can be done in Switzerland too, on three conditions: say it’s an AI, know what gets recorded and know where the data ends up. And on WhatsApp your company’s bot is still allowed, whatever January’s headlines about Meta “banning chatbots” suggested.

What Swiss customers accept

Comparis’s annual survey on AI assistants (Innofact, March 2026, 1,035 people from every region) also asks when people would rather deal with a machine than a person1. For simple, written questions, the answer is clear.

On the phone, the enthusiasm fades. Only 33.8% of respondents like the idea of talking to a chatbot on the phone, barely up from 31.4% in 2024. And 62.9% are not prepared to forgive a chatbot more mistakes than they would forgive a person.

What that means for an SME: a voice assistant works if it does a few things well (opening hours, availability, appointments, requests to pass on to someone) and hands over to a person as soon as it needs to. A voice that pretends to know everything doesn’t work.

Rule one: say it’s an AI, straight away

Switzerland has no law on artificial intelligence. It has the revised Federal Act on Data Protection (FADP), in force since 1 September 2023, and since 2023 the Federal Data Protection and Information Commissioner (FDPIC) has repeatedly made clear that it applies to AI too2. On transparency, it says:

“In the case of intelligent language models that communicate directly, users have a legal right to know whether they are speaking or corresponding with a machine and whether the data they have entered is being processed to improve self-learning programs or for other purposes.”

On 28 January 2026, Data Protection Day, the FDPIC said it again: users “must be able to recognise when they are dealing with AI”3.

In practice, the first sentence is enough: “Good evening, this is the joinery’s AI assistant.” You don’t need a robotic voice or a minute-long disclaimer. What matters is that nobody discovers halfway through a call that they’ve been talking to a machine.

• A real person’s voice. Want the assistant to speak with your voice? According to the FDPIC, the use of programs that can fake the voice messages of identifiable people must always be clearly indicated2. A neutral voice avoids the problem.
• Customers in the EU. From 2 August 2026 the EU AI Act (Art. 50) requires providers of chatbots and voice assistants to tell people they are talking to an AI, unless it is obvious. This also applies to providers outside the EU if their service reaches people in the EU4.

Rule two: recording and transcribing calls

To transcribe and summarise a call, the assistant has to listen to it and usually store it. This is where the Criminal Code comes in.

Recording a non-public conversation without the consent of the other participants is a criminal offence, even if you are taking part in it yourself (Art. 179ter SCC)5. There is one exception: in business dealings, recording phone calls about orders, contracts, bookings and similar transactions is not an offence. But those recordings may only be used as evidence (Art. 179quinquies SCC).

The FDPIC explains that the exception is a narrow one6:

• it covers “bulk business” with a degree of urgency, such as a booking line;
• it doesn’t cover a complaint, and for a broader negotiation conducted by phone it is reasonable to say at the outset that the call is being recorded;
• if recordings are used to train staff or to monitor sales, people must be told beforehand, and they cannot be used for marketing.

An assistant that transcribes, summarises and writes into your business software uses the conversation for far more than keeping evidence. The simple route is to say so at the start, in the same sentence in which it introduces itself as an AI, and to state it in your privacy notice (Art. 19 FADP)7.

This is a guide, not legal advice. For your own situation, ask your fiduciary or a lawyer.

Rule three: WhatsApp Business after Meta’s crackdown

In October 2025 Meta changed the WhatsApp Business terms: from 15 January 2026, AI providers could no longer use it to offer assistants where the assistant is the main product rather than an ancillary service8. Hence the headline “WhatsApp bans chatbots”. For your company’s bot, that isn’t true: Meta made clear that the rule doesn’t affect businesses that use AI to serve their own customers, such as a travel agency with its customer service bot.

WhatsApp and AI assistants: the key dates

Date	What happened
15 October 2025	Meta changes its terms: rival general-purpose AI assistants are out8.
24 December 2025	Italy’s competition authority orders Meta to suspend the new terms9.
15 January 2026	The new terms come into force.
4 March 2026	Meta lets rival assistants back in, but for a fee10.
9 June 2026	The European Commission orders Meta to reopen free access for rival assistants in the European Economic Area until its investigation is complete10.

It’s a battle between Meta and the providers of general-purpose assistants such as ChatGPT or Perplexity8. For a bot that books appointments, nothing changes. What applies to you are the WhatsApp Business rules, updated on 23 September 202611:

• you can only message people who have given you their number and agreed to receive your messages;
• you can use automated replies within 24 hours of the customer’s last message, but you must offer a quick, clear way to reach a person: an agent in the chat, a phone number, an email address, your website or your shop;
• outside those 24 hours you can only send messages based on templates approved by Meta.

As for costs, since 1 July 2025 Meta has charged per message, and only for template messages12. Ordinary replies within the 24-hour window opened by the customer are free, as are “utility” templates (an appointment confirmation, say) sent within that window. Marketing messages are always charged. So a bot that replies to people who message you costs you nothing in Meta fees; a reminder sent after the window has closed does.

Where voice, text and data end up

AI phone assistant, virtual receptionist, automated answering service: the names vary, but there is almost never just one program behind them. Usually there’s a chain: the phone carrier, a service that transcribes speech, a language model that writes the answer, a synthetic voice and finally your calendar or business software. Each link can be a different provider, often abroad. The FADP asks for a few specific things7.

• A contract with whoever processes data for you (Art. 9). The provider only does what you yourself would be allowed to do, guarantees data security and only brings in others with your permission.
• Adequate countries, or safeguards (Art. 16). Data can go to countries the Federal Council considers adequate, such as those in the EU. In the United States, since 15 September 2024, it can go to companies certified under the Swiss-U.S. Data Privacy Framework13. Otherwise you need safeguards, such as contractual clauses recognised by the FDPIC.
• The privacy notice (Art. 19). Anyone who calls or writes must be able to find out who processes their data, for what purpose, who it is passed on to and, if it goes abroad, to which country.

Anyone who wilfully sends data abroad without meeting these conditions, or entrusts it to a provider without meeting the conditions of Art. 9, risks a fine of up to CHF 250,000 (Art. 61). That requires a complaint from the person concerned, and intent. It’s no reason to panic, just a reason to ask the right questions first.

With Digitam Voice, before we switch it on, we tell you in writing where calls are routed, where transcripts and summaries are kept, which AI model reads them and how long they are stored. If a provider can’t answer, that’s an answer in itself. The same questions apply to any AI tool: we explain them in our article on ChatGPT and the FADP.

When an AI phone assistant pays off, and when it doesn’t

It pays off where calls and messages are much the same and arrive when nobody can answer:

• the Ticino grotto that takes Saturday bookings on a Friday evening while it’s busy serving tables;
• the garage where the person who could answer has their hands under a car;
• the fiduciary firm that receives a request for a salary certificate on WhatsApp and passes it to the right person.

These are examples. A real case: on WhatsApp we work with an estate agency with five sales agents in the Lugano area. The automation transcribes customers’ voice messages and drafts a reply with suitable properties; requests land in a single dashboard instead of being scattered across the agents’ phones. And if the call is a request for a quote, with Digitam Quote the draft can be ready by the time you hang up.

If you want to hear how it sounds for your business, try the live demo of Digitam Voice: you’ll hear it answer, book an appointment and leave you the summary. Then you decide.

Sources

1. Bereits drei Viertel der Bevölkerung nutzen KI-Helfer wie ChatGPT & Co. (in German). Comparis, press release of 2 April 2026; Innofact survey of 1,035 people from every region of Switzerland, March 2026.
2. AI and data protection. FDPIC, page published on 24 September 2025, with the statements of 9 November 2023 and 8 May 2025.
3. Data Protection Day 2026: Uses of generative AI and data protection issues. FDPIC, 28 January 2026.
4. EU AI Act Article 50 Transparency Obligations Go Live. Baker Botts, 10 September 2026.
5. Swiss Criminal Code (SCC), SR 311.0. Fedlex, Arts. 179bis, 179ter and 179quinquies.
6. Recording conversations. FDPIC, information page.
7. Federal Act on Data Protection (FADP), SR 235.1. Fedlex, in force since 1 September 2023: Arts. 9, 16, 19 and 61.
8. WhatsApp changes its terms to bar general-purpose chatbots from its platform. Ivan Mehta, TechCrunch, 18 October 2025.
9. Meta AI: the Italian Competition Authority orders Meta to suspend the terms excluding competing AI Chatbots from WhatsApp. Italian Competition Authority (AGCM), 24 December 2025.
10. Commission imposes interim measures on Meta to preserve free access to WhatsApp for rival AI assistants. European Commission, press release IP/26/1276, 9 June 2026.
11. WhatsApp Business Messaging Policy. WhatsApp, updated on 23 September 2026.
12. Pricing on the WhatsApp Business Platform. Meta for Developers, per-message pricing in force since 1 July 2025.
13. Data Protection Ordinance (DPO), SR 235.11. Fedlex, Annex 1 (countries with adequate data protection; the United States for certified companies since 15 September 2024).

Written by Alexandru Ciobanu. Sources checked on 24 September 2026. If you spot a figure that doesn’t add up, email us at info@digitam.ch.