The short answer
Yes, if you own the recordings and can show that hosts and guests agreed to AI training. Long-running shows with kept masters, separate microphone tracks and transcripts are the easiest to license. Guest releases are the usual obstacle, and episodes you cannot clear are left out.
On this page
A weekly show that has run for eight years can hold four hundred hours of natural, unscripted conversation. That is exactly the kind of speech that is hard to stage in a recording booth, and it is why podcasters are being asked whether their back catalog is for sale. The answer depends less on the audio than on the paperwork behind it.
What do buyers want from a podcast archive?
Conversation. Scripted narration is easy to commission; real people talking over each other, laughing, correcting themselves and changing topic is not. A podcast archive offers that in bulk, often with a consistent setup over years.
What buyers check first is the spec sheet: sample rate, format of the masters, number of speakers, hours of speech, whether transcripts exist, and whether every voice is cleared. The voice dataset spec sheet sets out each line.
What helps an archive
- Original WAV or FLAC masters, not just the MP3s you published.
- Separate tracks for each microphone, if you kept them.
- Consistent recording setup across long stretches.
- Transcripts, ideally checked by a person.
- A short list of regular voices who are easy to reach.
What hurts it
- Theme music or beds under speech.
- Ad reads for sponsors, which may carry their own rights.
- Phone-in or remote guests on poor connections.
- Guests you cannot identify or contact.
Who owns the recordings?
Usually the show’s owner, but not always. If the podcast was produced by a network or studio, read the production agreement. Co-hosts may have a claim. A show recorded for a brand as paid content probably belongs to the brand. Licensing is limited to what you own outright or hold written permission for.
Do guests have to agree?
Yes, and this is where most archives stall. A typical guest release lets the show record, edit, publish and promote the episode. Licensing the guest’s voice to a third party to train an AI system is a different use. Some releases are written broadly enough to cover it; many are not. Your lawyer should read the wording.
Buyers take this seriously. Labelbox, for instance, says on its Prism page that it licenses consented voice data and excludes voice without explicit, timestamped consent. Expect any serious buyer to ask for the releases, not just your word.
The practical route for most shows:
- List every guest and the episodes they appear in.
- Check what each one signed.
- Write to reachable guests with a short release that names AI training and says what they receive.
- Drop episodes with guests who decline or cannot be found.
| Episode type | Usually clearable? | What to check |
|---|---|---|
| Hosts only | Yes, between the hosts | Co-host agreement and ownership |
| Regular guests | Often | A new release naming AI training |
| One-off guests | Sometimes | Contact details; many will not reply |
| Call-ins or listener audio | Rarely | Consent from members of the public |
| Sponsored segments | Depends | The sponsor contract |
Does the law treat a voice as personal data?
Recording laws decide whether an episode was lawfully made: some states need one party’s consent, and about a dozen need everyone’s. For a podcast where every guest knew they were being recorded, that is rarely the issue.
Biometric laws are the newer question. Illinois’s Biometric Information Privacy Act counts a voiceprint as a biometric identifier and requires written consent before one is collected; Texas and Washington have their own laws. If a buyer might use your archive to recognize or clone voices, ask your lawyer how these apply.
What is a podcast archive worth?
No buyer we track publishes a per-hour price for podcast audio. Labelbox publishes license ranges for datasets in general, starting around $50K for smaller targeted sets, but that is not a price for your episodes. Video podcasts have one reported reference point. A Bloomberg report of 10 January 2025 put what AI companies paid for video footage at between $1 and $4 a minute. That covered video rather than audio alone, and real deals frequently settle below reported highs.
What raises the value is the same as for any voice data: clean masters, per-speaker tracks, checked transcripts and full releases. What lowers it is music under speech, mixed setups and voices you cannot clear.
How do you prepare the files?
Do not send anything to anyone yet. Start with a description.
- Count hours of speech per year of the show, excluding intros, ads and music.
- Note the master format and sample rate for each period.
- List the speakers with an anonymous ID, the hours each contributes and their release status.
- Note which episodes have transcripts and how they were made.
When a buyer is interested, it will ask for samples under its own agreement. Before you share even one episode, remove guest contact details from show notes and anything personal said about people who never agreed to be recorded.
If the archive passes that check, you are in a good position to sell voice data on terms you understand. If it does not, the work is mostly letters to past guests, and that can start this week.
An archive is only as sellable as its least-documented guest.
Sources for this post
- Labelbox, Prism licensing site (read on 7 November 2026)
- Bloomberg, Aisha Counts, report on creators licensing unused footage to AI firms (10 January 2025)
- 740 ILCS 14 (BIPA), Illinois Compiled Statutes