Updated October 2026 · Voice rights, consent and deepfake labels
Disclose any AI-generated or cloned voice in a podcast clearly, near the start of the episode, with a plain sentence that says what was generated and whose voice it is based on. Repeat the note in the description, and keep the same wording across the show so listeners learn to trust it.
Podcasts are built on trust between a host and a listener. A voice that listeners think is human, and that turns out to be generated, can break that trust quickly. Disclosure is the simple way to protect it. Done well, it takes a few seconds and makes the show more credible. Done badly, it reads like a warning label and makes listeners uneasy.
This guide covers when disclosure is needed, how to word it, where to place it and how to keep it consistent across a season. It is practical advice for podcasters, and it sits alongside the legal points covered elsewhere in this series.
Disclose any voice that is generated, cloned or altered so that it could be taken as real. A narration voice generated from text, a host voice created from a recording and a guest voice produced by a tool all count. Cleaning noise, levelling speech or cutting a pause does not generally need a disclosure, because the speech is still the person’s own.
When you are unsure, disclose. A short note is cheap, and it prevents a complaint that could damage the show far more.
Use words that describe what happened. A sentence such as “the narration in this episode was generated by an AI voice model, based on a recording of the host” is clear and honest. Avoid vague phrases such as “enhanced audio”, which can mislead, and avoid jargon. The listener should understand the disclosure the first time they hear it.
If the voice is cloned from a real person, say so, and say whether that person agreed. That detail protects both the speaker and the listener.
Put the disclosure near the beginning, after the intro and before the first generated line. A listener who joins late, or who skips the intro, should still hear it before the generated voice matters. If the generated voice returns after a long gap, repeat a short reminder.
A disclosure at the very end of an episode is easy to miss, and many listeners never reach it.
A written note in the episode description gives listeners a place to check, and it helps the transcript and search record the disclosure. Use the same words as the spoken note. The written note supports the spoken one, but it should not replace it, because many listeners never read the description.
Use the same disclosure across every episode where the voice appears. Listeners learn to recognise it, and that recognition builds trust. Change the wording only if the tool, the voice or the way it is used changes, and say so on air when you do.
Automatic transcripts may not mark the generated voice. Add the disclosure to the transcript and captions by hand, and check the names and terms against the audio. A transcript that hides the disclosure undermines the note, even when the audio includes it.
A clip cut from an episode may reach people who never hear the full show. If the clip contains a generated voice, add the disclosure to the clip itself, in the audio or in a caption. The short clips post covers the wider practice.
Write the disclosure into your episode template, so it is added every time without a decision. Include the exact words, the place in the episode where they appear and the matching description note. A template removes the risk that a busy week leads to a missed label.
Review the template when the voice, the tool or the way it is used changes. Update the wording on the same day, so the old disclosure does not outlast the change.
Listeners often accept a disclosure more easily when they understand why it is there. A short line such as “we label generated voices so you always know what you are hearing” explains the purpose without a lecture. Tell listeners what the voice does in the show, such as reading a summary or translating a segment, so they can judge the use for themselves.
If the voice is a version of a real person, say whether that person agreed and how they are credited. Honesty about the process builds the trust that the disclosure is meant to protect, and it gives the audience a reason to keep listening.
A disclosure does not make a misleading use acceptable, and it does not replace consent from the person whose voice is used. Some uses may be restricted whatever the disclosure says. This is general practice, not legal advice.
Real voices deserve the same care as generated ones. The Vocal Checker shows clipping, noise and level problems in a recording, and the Tunedzy Go page covers cleaning a phone recording. For the labelling rules that apply in the EU, read the EU labelling post in this series.
Spot clipping, noise, rumble and level problems. Runs in your browser, nothing is uploaded.
Disclose any voice that could be taken as real. When in doubt, disclose.
No. Use a spoken note near the start, and add the written note to the description.
Cleaning noise and levelling do not usually need a disclosure, because the speech is still yours.
Yes, if a clip contains a generated voice, since viewers may never hear the full episode.
