ElevenLabs: AI voice, done responsibly

What it is

ElevenLabs makes AI voice tools. The main ones are:

• Text to speech: paste a script, get a recording.
• Voice Library: thousands of ready-made voices.
• Voice cloning: a copy of a real voice made from recordings. Instant cloning works from short samples; Professional cloning trains a closer copy and is only allowed for your own voice.
• Dubbing: translates the speech in a video or audio file into another language.
• Speech to text: turns recordings into written transcripts.
• Conversational agents (ElevenAgents): voice assistants that answer people in real time, for example by phone.

Think of a recording studio with voice actors on call: fast and flexible, but the same rules apply as with real actors. You need permission, and listeners should know what they hear.

As of November 2026 (check the maker's page), the models include:

Why it matters

• Training narration: turn a procedure into a voice-over, and update it by editing text, not re-recording.
• Accessibility: audio versions of long documents, transcripts of recorded meetings.
• Translation: dub internal videos into colleagues' languages.
• Risk: a cloned voice can impersonate someone, such as a fake call from a manager asking for a payment (a deepfake).
• Law: in the European Union, the AI Act sets transparency duties for AI-made audio from 2 August 2026 (see "Going further").

How to do it

1. Check that your organisation has approved the tool and the account, and which data may go in (see guide 10).
2. Approve the script (yours or AI-drafted) before any audio is made (see guide 09).
3. Choose a voice from the Voice Library, or use a clone only with the speaker's written permission.
4. Generate the audio and listen to all of it: names, numbers and jargon are often mispronounced.
5. Label the result clearly, for example "This narration uses an AI-generated voice."
6. Have a native speaker review any dubbed version before it is shared.
7. Store the script, approval and consent record with the audio.

Example

HR wants a five-minute audio version of the new starter safety briefing in English and Spanish.

The safety lead approves the written script. A colleague pastes it into text to speech with a library voice, listens through and fixes two mispronounced product names by spelling them phonetically. The Spanish version is dubbed, then checked by a Spanish-speaking colleague. Both files start with this line:

This briefing is read by an AI-generated voice. The content was
written and approved by [team name] on [date].

Common mistakes

• Cloning a colleague's or manager's voice "for fun". Fix: never clone a real person without their written permission, and follow the maker's own rules, which go further (see below).
• Publishing without listening. Fix: a person listens to every file in full before release.
• Hiding that the voice is AI. Fix: say so at the start and in the description.
• Trusting a familiar voice on the phone. Fix: confirm unusual requests, especially payments, through a second, known channel.

Try it this week

1. Write a 100-word, non-confidential script, such as how to book a room (4 minutes).
2. In an approved account, generate it with two library voices (4 minutes).
3. Listen to both and note every mispronunciation (4 minutes).
4. Add the AI-voice disclosure line and ask a colleague which version is clearer (3 minutes).

Going further

• Maker's safeguards, as ElevenLabs states them: Professional clones need a verification step to confirm the voice is yours, and you cannot professionally clone someone else even with their consent; every Eleven v4 clone needs verified consent from the voice's owner; cloning of celebrity and other high-risk voices is blocked; it uses standards such as C2PA (content provenance: a label showing where media came from); and it has released a public AI Audio Classifier to detect its audio. Its usage policy requires disclosure when people are interacting with AI.
• Data use, as the maker states it: Enterprise customer data is not used for training by default; other users can switch off "Improve the models for everyone". Select Enterprise customers can use Zero Retention Mode.
• EU AI Act, Article 50 (European Union): from 2 August 2026, providers of systems that generate synthetic audio must mark it in a machine-readable way, and organisations that use AI to create deepfake content must disclose it clearly, for example with a visible or audible label. Systems already on the market before that date have until 2 December 2026 for the marking duty. This is general information, not legal advice; ask your legal team.
• Voice agents should hand complaints, payments and anything unusual to a person (see guide 26).

Key words

• Text to speech: software that reads written text aloud.
• Voice clone: an AI copy of a real person's voice.
• Dubbing: replacing speech in a video with another language.
• Deepfake: AI-made audio, image or video that resembles a real person and could seem genuine.
• Disclosure: telling the audience that content was made by AI.

Sources

Examples and figures in this guide are illustrative. AI tools change often: check what your organisation's approved tools offer today. Use approved tools only, and a person checks and decides.

Prefer to print or share it? Download this guide as a PDF (283 KB), or the complete AI Field Guide (2.5 MB).