AI voices in your app or website: rights and the AI Act

01 — Give the voice one clear job

Most products start with one of three uses. Each asks something different of the voice:

• A listening version of articles and guides: long narration in one consistent voice
• Voice-over for product videos, onboarding and app screens: expressive delivery in short fragments that change with your copy
• A voice agent for customer contact: a real-time conversation, a fast response and a clear handover to a person

02 — Choose the model for the job

ElevenLabs covers all three uses on one platform: a text-to-speech API for your app or website, ElevenCreative Studio for long-form narration and ElevenAgents for conversational voice agents on the web, in apps and on the phone. Its documentation lists Dutch as a supported language for its current multilingual models.

Choose expressiveness for narration and voice-over, and speed for conversation. On 19 October 2026 ElevenLabs described Eleven v4 as its most emotionally rich, expressive speech model and Flash v2.5 as an ultra-fast model optimised for real-time use. Test your own script before you decide: names, numbers and product terms tell you more than a demo sentence.

Generate fixed audio once and store it with the page or screen it belongs to; regenerate it when the text changes. Use the API in real time only where the text is truly dynamic. That keeps your app fast and your usage predictable.

03 — Make sure you may publish it

Commercial use needs a paid plan. ElevenLabs states that its free plan does not include a commercial licence and cannot be used for any commercial purpose, while all paid plans include one, provided you are not using beta services. Under its terms of use you keep the rights to your output.

Cloning a voice needs consent. For Instant Voice Cloning you confirm that you have the right and the consent to clone the voice. Professional Voice Cloning is limited to your own voice, which ElevenLabs verifies. Its Prohibited Use Policy, updated on 9 October 2026, bans cloning or impersonating the voice of a living person without their explicit and informed consent or other valid authorisation.

A voice is also personal data. The European Data Protection Board calls voice data inherently biometric personal data, and the Belgian Data Protection Authority treated voice prints as biometric personal data in a decision of February 2025. When you clone the voice of a founder, employee or actor, agree in writing what the clone may be used for and when it is deleted.

04 — Tell people it is AI

The transparency rules of the EU AI Act (Article 50) have applied since 2 August 2026. A system that talks directly with people, such as a voice agent, must let them know they are interacting with AI, unless that is obvious. Say it clearly, at the latest at the first interaction.

Providers of systems that generate synthetic audio must mark the output in a machine-readable format, so it can be detected as artificially generated. For generative systems placed on the market before 2 August 2026, Regulation (EU) 2026/1744 gives providers until 2 December 2026 to apply that marking. ElevenLabs states that it embeds imperceptible watermarks in the audio it generates.

If you publish a deepfake, you must disclose it: audio that resembles existing persons, objects, places, entities or events and would falsely appear authentic or truthful. For evidently artistic, satirical or fictional work, a disclosure that does not spoil the work is enough. A short line such as “Read by an AI voice” next to your player keeps a listening version clear either way.

ElevenLabs' own terms for ElevenAgents go further: before anyone can use your agent, they must be told that they are talking to AI and that the conversation is recorded and may be shared with ElevenLabs and its language-model providers. Build that notice into the start of every conversation.

05 — Build the voice into the experience

A listening version adds value, but it does not replace an accessible page. The W3C notes that text-to-speech software only works well when content is coded properly. Keep the written text, give the player clear controls and never start audio on its own: WCAG requires a way to pause, stop or turn down audio that plays automatically for more than three seconds.

For a voice agent, decide in advance which questions it answers, what it may not do and when it hands over to a person. Test it with real customer questions in both languages before it goes live, and keep a written channel for people who prefer not to talk to AI.

Check where your data goes. ElevenLabs offers zero retention mode and EU data residency for enterprise customers. On other plans, read the privacy terms before you send customer conversations or personal data to the API.

Compare before you choose