Does Article 50 apply to voice bots?

Yes. The Article 50(1) disclosure duty covers any AI system that interacts directly with natural persons; that explicitly includes voice bots and phone-based virtual assistants, not just chat interfaces.

The required form for voice is a spoken statement, delivered before any substantive content, at the start of every call. The verified pattern (from the Commission's draft guidelines) is a simple opening line; in English, for example: "This call is handled by an AI-powered virtual assistant." Equivalent verified wording exists in German and French.

For long calls, add a mid-call reminder: the disclosure should be persistent where appropriate, not just a one-time opener that a caller might miss or forget. If a human agent takes over partway through the call, a handoff line ("You are now connected with a human agent") is good practice; it's not a strict legal requirement, but it keeps the caller correctly informed about who they're speaking to.

One nuance specific to branded conversational AI: a branded voice bot is, in principle, also a generative text system in the sense that matters for the separate Article 50(2) marking duty, but the Commission's draft guidance doesn't explicitly address conversational or spoken output either way, and text/audio watermarking for live conversation is technically weak. This is a genuinely unsettled area, not a clear yes or no; the defensible position is to document what your underlying voice/model provider does (if anything) and revisit once the final guidelines land.

For evidence: keep a dated record of your call script or IVR configuration showing the opening disclosure, the same way you'd keep a screenshot for a chat widget.

The same accessibility principle that applies to visual disclosures applies here too: if children or other vulnerable groups are a foreseeable part of who calls in, simplify the spoken language accordingly rather than relying on a single generic script for every caller.