Skip to content
Compliance · Switzerland

Swiss FADP and Chatbots: the 2026 Compliance Checklist

The revised Swiss Federal Act on Data Protection (revFADP, revDSG) has applied since September 2023. A practical checklist for running an AI chatbot under it, with the questions to ask your vendor before you sign.

Author
intoCHAT Team
Published
Last updated
Reading time
9 min read
Illustration: a padlocked shield, a vendor checklist and data flowing from a chat widget into a safe
On this page
  1. What the revised FADP means for a chatbot
  2. Personal data in chatbot conversations
  3. Tell visitors they're talking to an AI
  4. Where data is stored and where it travels
  5. Processor agreement (DPA) and sub-processors
  6. Retention, deletion and data subject rights
  7. Access control and an audit trail
  8. The practical checklist
  9. How intoCHAT helps with the checklist
  10. Frequently asked questions

If your business operates in Switzerland and runs an AI chatbot, you are processing personal data under the revised Swiss Federal Act on Data Protection (revFADP, in German revDSG or nDSG). It has applied since 1 September 2023 and brings Swiss law broadly in line with the EU's GDPR, with some Swiss specifics. This article is a practical checklist for a website chatbot, the kind of due diligence a Swiss legal or procurement team runs before launch. It is not legal advice.

What the revised FADP means for a chatbot

The points that matter most for a chatbot:

  • Duty to inform (Art. 19). When you collect personal data, you tell people who you are, why you process the data, who receives it and, for transfers abroad, which countries. A chatbot needs a place in your privacy notice.
  • Privacy by design and by default (Art. 7). Collect only what the chatbot needs, and choose defaults that protect visitors.
  • Processors (Art. 9). A chatbot vendor processes data on your behalf. You stay responsible, need a contract, and the vendor may use sub-processors only with your prior authorisation.
  • Record of processing activities (Art. 12). Companies with fewer than 250 employees are exempt unless they process sensitive data on a large scale or run high-risk profiling. Many Swiss SMEs keep one anyway, because it makes the other duties easier.
  • Data breaches (Art. 24). A breach likely to cause a high risk must be reported to the FDPIC as soon as possible.
  • Fines. Up to CHF 250,000 for intentional breaches of certain duties, aimed at the responsible individuals rather than the company.

Personal data in chatbot conversations

A chatbot conversation almost always contains personal data, even when no name or email is asked for. Visitors type names, order numbers, addresses and sometimes details about their health, money or legal situation. Some of that is sensitive data under the FADP and calls for extra care.

So every chatbot needs a defined purpose, a retention rule, a working process for access and deletion requests, and a mention in your privacy notice that covers what the chatbot collects and who receives it.

Tell visitors they're talking to an AI

The FDPIC has stated that the FADP applies directly to AI-supported processing, and that people should know when they are talking to a machine. Name the agent as an AI assistant in its greeting, and link your privacy notice from the chat window. If you also serve visitors in the EU, the EU AI Act requires the same: its Article 50 transparency duties apply from August 2026.

Where data is stored and where it travels

The revised FADP does not require data to stay in Switzerland. Transfers are allowed to countries the Federal Council lists as adequate, which includes the EU and EEA states. Since 15 September 2024, the list also covers US companies certified under the Swiss-US Data Privacy Framework. Transfers to other recipients need safeguards, typically standard contractual clauses.

For a chatbot there are two places to ask about:

  • Where conversations are stored at rest: the vendor's database and file storage.
  • Where messages are processed: the AI model provider that writes the replies. Most large providers (OpenAI, Anthropic, Google) process in the United States. Ask which provider your vendor uses, whether the provider trains on your data, and how long it keeps it.

Processor agreement (DPA) and sub-processors

A chatbot vendor almost always processes personal data on your behalf, so you need a processor agreement (often called a DPA, or Auftragsverarbeitungsvertrag in German). It should cover the categories of data, the purposes, the sub-processors and how you learn about new ones, the security measures, breach notification and help with data subject requests. A vendor that can't provide one on request is a risk.

Retention, deletion and data subject rights

Decide how long you keep transcripts and leads, and write it down. A common pattern is a few months for transcripts, the length of the customer relationship for leads, and anonymised statistics beyond that. Check that your chatbot lets you act on it: delete a conversation, delete a lead, and export what you hold about a person when they ask. Under the FADP you answer an access request within 30 days.

Access control and an audit trail

Who can read transcripts and change the agent? Give each team member only the access they need, require two-factor sign-in, and keep a record of who changed what. For banks, insurers and other regulated firms, single sign-on and a longer audit trail are usually on the vendor questionnaire.

The practical checklist

  • Processor agreement signed, with the sub-processor list on file.
  • Storage location and AI provider documented, including transfers abroad and their legal basis.
  • Privacy notice updated: the chatbot, what it collects, who receives it, which countries.
  • The agent introduces itself as an AI, and the chat window links your privacy notice.
  • Retention rule per data category, with a routine to delete old conversations and leads.
  • Access, deletion and correction requests tested end to end.
  • Team access limited by role, two-factor sign-in on, changes logged.
  • Agent limited to your own content and your own domains.
  • Breach process agreed with the vendor.

How intoCHAT helps with the checklist

No software makes a chatbot compliant on its own, and intoCHAT holds no certification such as SOC 2 or ISO 27001. What it gives you are the controls and the documentation for most of the points above.

  • DPA and providers. You can request a data processing agreement at privacy@intochat.ai. The privacy policy lists the providers: Vercel (hosting and file storage), Supabase (database), Redis Cloud in the EU (rate limits), OpenAI, Anthropic and Google (AI replies), Firecrawl (reading your website), Resend (email) and Stripe (payments).
  • Hosting. The database that holds what intoCHAT stores is hosted in Switzerland. There is no choice of region, and some providers process data in the United States and other countries, under standard contractual clauses where the law requires them.
  • Model and provider choice. Each agent writes its replies with Claude Haiku 5.5 from Anthropic by default, and GPT-6 Luna from OpenAI answers if Anthropic doesn't. Instead, you can pick GPT-6 Luna on every plan, a Gemini model from Google from Starter or a larger Claude model on Pro. OpenAI is still used to index your knowledge and for background features such as summaries, so name it in your privacy notice whichever model you pick. intoCHAT doesn't train AI models on your data or your visitors' data.
  • Answers from your content only. You decide which pages, files and Q&A the agent learns from. Built-in rules make it say what it can't confirm instead of inventing an answer, and it never asks for passwords or card details in the chat.
  • Deletion and export. Delete a conversation with its messages and files, delete leads and contacts, and export conversations, leads and contacts as CSV to answer an access request. With Let visitors delete their chat, visitors can delete their own conversation from the widget. There is no automatic deletion schedule, so plan a regular clean-up to match your retention rule.
  • Disclosure. Name the agent and write its greeting yourself, and use the widget's footer text for a link to your privacy notice.
  • Identity verification. On every plan, your server signs each signed-in user's id, so the agent knows who it is talking to and nobody can pretend to be another customer. You can let only verified visitors chat.
  • Team controls. Roles (Admin, Editor, Viewer) and access to selected agents from Starter, two-factor sign-in for everyone, and an audit log of account changes on Pro (90 days) and Enterprise (365 days). The log records who changed agents, sources, integrations, API keys and team members and who downloaded exports, never message content. Single sign-on with SAML and custom roles are on Enterprise.
  • Where the agent answers. Limit it to your own domains and block countries you don't serve.

For GDPR and FADP side by side, read our chatbot GDPR guide. For an overview, see the AI chatbot for Switzerland page.

This article is informational and not legal advice. For a binding interpretation of the revised FADP for your business, consult a Swiss data protection lawyer.

Frequently asked questions

Does the revised Swiss FADP apply to website chatbots?

Yes. A business that runs a chatbot in Switzerland processes personal data under the revised Federal Act on Data Protection (revFADP or revDSG), which has applied since 1 September 2023. Chat transcripts, contact details and free-text messages can all contain personal data, so the chatbot belongs in your privacy notice and, where you keep one, your record of processing activities.

Does the FADP require chatbot data to be stored in Switzerland?

No. The revised FADP does not require storage in Switzerland. Transfers abroad are allowed to countries the Federal Council lists as adequate, such as EU and EEA states, and to US companies certified under the Swiss-US Data Privacy Framework; other transfers need safeguards such as standard contractual clauses. intoCHAT's database is hosted in Switzerland, without a choice of region, and its AI providers process messages in the United States and other countries.

Do I need a data processing agreement with my chatbot vendor?

In almost all cases, yes, because a chatbot vendor processes personal data on your behalf. The agreement should cover the categories of data, purposes, sub-processors, security measures, breach notification and support with data subject requests. intoCHAT provides a data processing agreement on request at privacy@intochat.ai.

What are the penalties under the revised FADP?

The revised FADP provides for fines of up to CHF 250,000 for intentional breaches of certain duties, such as the duty to inform, the rules on transfers abroad and the duty of care when engaging a processor. The fines are aimed at the responsible individuals, such as managers, rather than the company. Documenting your chatbot processing and signing a processor agreement help reduce that risk.

Does intoCHAT make a chatbot FADP compliant?

No software makes a chatbot compliant on its own, and intoCHAT holds no certification such as SOC 2 or ISO 27001. It gives you the tools for much of the checklist: a data processing agreement on request, a database hosted in Switzerland, a choice of AI provider per agent (OpenAI, Anthropic or Google), deletion and CSV export of conversations and leads, a delete button for visitors, identity verification, team roles, two-factor sign-in, and an audit log on Pro. OpenAI is used to index knowledge whichever model you choose, so it belongs in your privacy notice.

Back to Blog
  • Illustration of five chat assistants facing a smartphone with a customer message, a clipboard with a bar chart beside them and the Alps behind
    Study · Switzerland

    Do AI Chatbots Understand Swiss German? We Tested 5 Models

    50 customer messages from 7 regions, 5 current AI models, 1,000 answers: every model understood written Swiss German almost as well as standard German. The differences are in how they reply.

    6 min read

  • Editorial illustration of a closed shop inside a browser window, with a warm chat light connecting it to a visitor on a laptop at night
    Lead generation

    The 11:47 PM Customer: Your Best Lead Is Still Awake

    Someone is on your pricing page tonight, one unanswered question away from leaving. Here is how to build an AI lead generation chatbot that helps them take the next step—with a reusable prompt, ten tests and a seven-day launch plan.

    10 min read