Mercury Logo - Full version with bubbles and lettering "Mercury.ai" Conversational AI Platform

Industries

Applications

Platform

Resources

Company

Prices

Mercury Logo - Full version with bubbles and lettering "Mercury.ai" Conversational AI Platform

Labeling AI on the Phone: Disclosure Requirements Under GDPR and the EU AI Act, in Every Language

Post

Compliance & Data Protection

Labeling AI on the Phone: Disclosure Requirements Under GDPR and the EU AI Act, in Every Language

Labeling AI on the Phone: Disclosure Requirements Under GDPR and the EU AI Act, in Every Language

Dr. Hendrik ter Horst, CPO of Mercury.ai

Dr. Hendrik ter Horst

Chief Product Officer @Mercury.ai

4 Min. read time

Multiple speech bubbles in various colors, each beginning with the same short note
Multiple speech bubbles in various colors, each beginning with the same short note
In this article

A voicebot must inform callers right at the beginning that they are speaking with an AI, unless this is obvious. This is mandated by the transparency obligation under Article 50 of the EU AI Act. In addition, there are the information obligations of the GDPR: who is responsible, for what purpose the data is processed, and where callers can learn more. If a call is being recorded, a notice about this must also be included right at the start. If the hotline is multilingual, every notice applies in each language in which the hotline is available. This article answers the most common questions on this topic and provides greeting scripts in five languages.

Does a voicebot have to state that it is an AI?

Yes, unless it is obvious to the callers. Over the phone, this is rarely the case: a natural voice and concise answers sound human. The safest approach is a sentence in the greeting before the conversation begins. Providing a notice only upon request is not sufficient, as many callers will never ask.

How should the notice be phrased?

Concise and unambiguous. "You are speaking with the digital assistant from Musterfirma" fulfills this purpose. Terms like "virtual employee" are vague and should be avoided. Whether the word "AI" is explicitly used is less critical than clarity: callers must understand that there is no human speaking on the other end.

What data protection information do callers need?

Under the GDPR, data subjects must be informed about who is processing their data, for what purpose, on what legal basis, and where they can exercise their rights. It is practically impossible to read all of this out over the phone. A layered approach has proven effective: a brief first layer in the greeting stating the controller and the purpose, accompanied by a reference to where the full information can be found, such as on the website. You should coordinate the specific wording with your data protection officer.

What applies if the call is recorded?

In that case, the recording requires a legal basis—typically consent—and a notice before the caller's first word. Without consent, recording a phone call is a criminal offense in Germany (Section 201 of the German Criminal Code, StGB). If you do not record, you do not need to say anything about it. How transcripts, recordings, and retention periods correlate is detailed in our article on call recording in voicebots.

How does this work in multiple languages?

Every notice applies in every language in which the hotline is available. An English greeting paired with a German data protection notice does not satisfy the obligation for English-speaking callers. In practice, this means: dedicated greeting scripts per language, carefully translated and reviewed by a speaker of the language. The greeting should also be spoken by a voice that matches the language.

Begrüßungstexte eines Voicebots in Deutsch, Englisch, Französisch, Spanisch und Italienisch, jeweils mit KI-Hinweis, Zweck und Verweis auf die Datenschutzinformationen

How long should the greeting be?

As short as possible. Every second prior to the actual conversation strains patience. Two to three sentences have proven ideal: AI disclosure, purpose, and reference to data privacy information, along with the recording notice if applicable. The prompt asking for the caller's request follows directly afterward.

Where should the notice be placed if the voicebot only acts as an overflow?

At the beginning of the segment handled by the voicebot. If it only takes over when no one on the team picks up, its part begins with the greeting, and that is where the notice belongs. It must be clear to callers from what point onward they are speaking with an AI.

What applies if the voicebot transfers to a human?

Callers should be informed that they are now being transferred, for example: "I will connect you with a colleague from customer service." If the agent receiving the call receives a prior summary of the conversation, this constitutes an internal transfer within the company for processing purposes. It must be included in the privacy policy, just like a handoff in chat.

How does Mercury Voice implement this?

In Mercury Voice, each voice channel has its own voice and greeting script, which is spoken immediately upon answering. This allows every language, phone number, or hotline to have its own customized notice. The voices cover European languages, including German with variants for Germany, Austria, and Switzerland, English, French, Spanish, and Italian. A pronunciation dictionary ensures that company and product names sound correct in every language variant. Mercury Voice announces transfers to employees with a customizable message, and the receiving agent hears a brief summary beforehand. Recording can be toggled per channel and is disabled by default. Learn more on the page about Mercury Voice and in the article on GDPR-compliant AI phone assistants.

Looking to set up your greeting scripts across multiple languages? Talk to us.

Dr. Hendrik ter Horst, CPO of Mercury.ai

About the Author

Dr. Hendrik ter Horst

Chief Product Officer @Mercury.ai

Dr. Hendrik ter Horst is Chief Product Officer at Mercury.ai. He holds a PhD in Computer Science from the CITEC Institute at Bielefeld University, with a focus on machine information extraction and data processing. For ten years, he has been working on dialogue-based AI systems, overseeing the conception, implementation, and validation of AI-powered applications from initial idea to production deployment.

Discover related posts
Four black dots on a white background as a symbol for interaction or user interface at mercury.ai

Experience Mercury.ai in a demo.

We show you how Mercury.ai answers customer inquiries.

Four black dots on a white background as a symbol for interaction or user interface at mercury.ai

Experience Mercury.ai in a demo.

We show you how Mercury.ai answers customer inquiries.

Four black dots on a white background as a symbol for interaction or user interface at mercury.ai

Experience Mercury.ai in a demo.

We show you how Mercury.ai answers customer inquiries.