Resources
Who owns the rights to your podcast?
You made the podcast, so you own it, right? Usually not all of it. Here is who holds which rights, with a worked example, before you license an episode for AI training.
Published 2026-07-22 · 6 min read
Key takeaways
- One podcast episode is a stack of rights: the recording, any underlying work, every performance, and voice and likeness where speakers are identifiable.
- Owning the show does not mean owning every layer. A network deal or a guest performance can sit with someone else.
- A release covering publication does not automatically cover AI-training use. Check the scope.
- Music almost always carries separate rights. If you cannot clear it for training, remove it.
- Chain of title is the documented line from each contributor to you, and it is exactly what buyers inspect.
I made it, so I own it. That is the assumption behind most podcast licensing questions, and it is usually wrong. A single episode is a stack of rights held by different people. The host. The guests. Maybe a network. Almost always the music.
Before you can license an episode for AI training, you need to know which of those rights you actually hold. This is chain of title: the documented line from each contributor to you. Get it clear and licensing is simple. Skip it and a buyer’s diligence will find the gap.
The layers in one episode
Start by separating the layers. There is copyright in the recording itself, the sound recording. There is copyright in the underlying work, the script or the composition where one exists. There are the performances of everyone who speaks. And where a speaker is identifiable, there are voice and likeness interests on top.
One episode can carry all of these at once. Ownership of the feed does not automatically mean ownership of every layer inside it.
Host, guest, network
The host usually owns the recording and the show, but not always. If a network or studio produces the show, the contract may assign rights to them. Read the production agreement before you assume.
Guests are the layer most often missed. A guest performs on the episode. Without a release or appearance agreement, their permission to reuse that performance for a new purpose, such as AI training, may not exist. A standard release covering publication does not necessarily cover training. Where the guest is identifiable, their voice and likeness are in scope too.
Music is its own problem
Music is where clean chain of title goes to die. An intro track, a stinger, or a bed usually carries separate rights held by a composer, a publisher, and a label. A licence to use music in a podcast is not a licence to include it in training data.
If you cannot clear the music for training, the practical answer is to remove it. An episode stripped to speech is far easier to license than one carrying third-party music you do not control.
A worked example
Take one interview episode. You host it, so you own the recording, subject to your network deal if you have one. Your guest spoke for an hour, so you need a release that covers AI-training use of their performance and, because they are identifiable, their voice. Your intro uses a licensed music track, so those rights sit with the composer and label, not you.
To license this episode cleanly, you hold the recording, you secure the guest’s consent for training, and you remove or separately clear the music. What remains is an asset with a documented chain of title. That is licensable. The raw episode, with an uncleared guest and third-party music, is not.
Building a clean chain of title
Work backward from the finished episode to every contributor. For each, ask a simple question: do I have documented permission that covers AI-training use? Where the answer is yes, keep the record. Where it is no, get consent or remove that layer.
This is exactly what a buyer inspects in diligence. Owners who can show the chain move quickly. Owners who cannot get stuck at the first question.
Frequently asked questions
I own my podcast feed. Is that enough to license it?
Not by itself. The feed does not carry every layer inside an episode. You still need to account for guests, any network agreement, and music. Licensing for AI training requires rights in each layer, not just ownership of the show.
Do I need consent from past guests?
If a guest is identifiable and their performance is in the recording, generally yes, and specifically for AI-training use. A publication release may not cover training. Where you cannot secure consent, that episode is harder to license cleanly.
Related resources
Want data that clears this in diligence?
Whether you're building a model or sitting on an archive, the first conversation is short and specific.
Send a brief