AI for Podcasters in 2026: Booking, Edits, Sponsors and Voice Clones

AI can do much of a podcast's paperwork in 2026: find and pitch guests, turn a transcript into cuts, show notes and chapters, and tidy sponsor reports. It can't get a guest's consent, hold an opinion for you in an ad read, or decide what's fair to cut. Those stay with you.

Which AI tools do which podcast jobs?

Start with the row eating the biggest part of your week. Prices are from each company's own page and change often.

Booking guests: let AI build the list, then write the ask yourself

AI handles the list well and the reason someone should say yes badly. A database such as Rephonic can pull shows, listener numbers and contact details quickly.

For the invitation itself, name one specific moment from the guest's own work and say why your listeners would care. A guest on a busy press run reads a lot of these, and a generic first line is easy to skip.

Do you have to tell guests you're recording?

Yes, in practice. US federal law permits recording if a single participant agrees, but states including California, Florida, Maryland, Pennsylvania and Washington are listed as requiring every party's consent. A remote interview often crosses state lines, and the Digital Media Law Project says it's hard to know in advance which law applies, so its advice is to get everyone's consent.

The stakes are real. The Reporters Committee's California guide, last updated in August 2021, lists civil damages of $5,000 or three times actual damages, whichever is greater, plus criminal penalties. Those rules aren't limited to published audio, so a pre-interview call you run through a note taker deserves the same consent line as the episode.

Consent to record isn't consent to everything afterward. Your release should separately cover editing, clips and AI use, because a guest who agreed to a conversation may not have agreed to a generated sentence in their voice. Have a lawyer check your final wording. A starting point:

On tape, before the interview:
"We're recording this for [Show name]. Are you OK with that, and with us editing it for length and clarity?"

In the release:
"We may edit the recording, publish it in audio, video and text, and cut clips for promotion. We will use AI tools for transcripts, cleanup and show notes. We will not generate or alter your voice or likeness with AI unless you approve that specific use in writing."

Editing from a transcript: the cuts to listen back to

Transcript editors such as Descript let you edit a podcast by editing text, and they strip filler words automatically. For most interview shows that's the first tool to try, and it brings a specific risk: you stop hearing what you cut.

• Listen to every cut that changes meaning. A removed "I don't think" or "allegedly" reads as a small tidy-up on screen and sounds like a different claim on air.
• Don't patch a guest's words with AI without the written approval your release promises. Patching your own word with your own clone is a separate question, and Apple's disclosure rule can still apply to it.

Can you clone a guest's voice for your podcast?

Only with that guest's specific consent, and some tools refuse even then. Descript's guidance says someone else's voice can't train an AI voice unless that person specifically and verbally consents. ElevenLabs is stricter on professional clones: you can only clone your own voice, even with the other person's consent.

The law is moving the same way. Tennessee's ELVIS Act, in force since July 1, 2024, protects a person's voice whether the sound is real or simulated, and it isn't limited to musicians. Keep the signed approval with the episode file.

What Apple Podcasts, Spotify and YouTube require when AI touches your audio

Apple Podcasts asks the most of the three: audio or video generated with AI has to be disclosed in both the content and the metadata. Check Apple's guidelines before publishing a synthetic segment, because Apple says it may label or remove content that doesn't comply.

Show notes and transcripts: what to fix before publishing

Show notes generators are fine to use on every episode, provided someone reads the result. Look for the errors listeners and guests notice: a misspelled guest name, a link to the wrong book, a summary that promises something the episode never says. Apple's guideline 1.1 also requires metadata to represent the episode accurately.

Apple Podcasts generates its own transcript and labels it "Automatically generated." To correct one, download it from Podcasts Connect, edit it, link the file in your RSS feed, and set the episode to Display transcripts I provide. The label then reads "Provided by" your show name, and guideline 5.1 requires that transcript to match the audio, so proofread names first.

Sponsor reads and ad reports: disclosure, opinions and recaps

An obvious commercial read near the start of an episode needs no extra payment disclosure under an example in the FTC's Endorsement Guides, because listeners would expect the host was paid. If the wording suggests your own view, you must actually hold that view.

That matters for AI scripts. A chatbot given a sponsor's brief can easily write "I've used this every morning for a year." If you haven't, cut it. Casual mid-conversation mentions are harder for listeners to recognize as paid, and the FTC's guidance says a spoken claim needs a spoken disclosure.

Reporting is where AI saves the dullest hours. Paste your host's export and the sponsor's insertion order into a chatbot and ask for a one-page recap: episodes, air dates, downloads, code redemptions. Then check every figure against the source, since a wrong number here goes straight to the person paying you.

What Merlin does for a podcaster, and what it won't

Merlin, our Mac assistant, looks after the people side of a show: the guests, publicists and sponsors in your Gmail account and on your Google Calendar. Open People and Needs you lists anyone whose direct email has sat unanswered, such as the publicist who offered a date two weeks ago. In Meetings, press Prep me on the next pre-interview and send Merlin the chat request it writes: you get a sourced brief on the attendees, drawn from your earlier calls and your mail with them, plus a suggested agenda.

It can also record pre-interview and sponsor calls on your Mac. Paid plans add transcripts that separate speakers, and it may message you the to-dos afterward. A follow-up or booking confirmation you request in chat appears as a card, and nothing goes out until you've approved it, edits included. Routines, the jobs you set to repeat, work differently: nobody reviews a scheduled run, so one you allow to send email sends it without a card. Because the recording happens on your own Mac, telling the other person is your job.

Merlin isn't podcast software. Don't record your episode in it: each meeting saves one audio file, capturing the other side's audio is best effort, and there's no export. It won't edit audio, clone voices, publish to your feed, sell ads or place phone calls. Use a recording studio and a transcript editor for the show.

Try it on your next episode

1. Write the consent line and release first. Add the AI clause before any new tool touches a guest's audio.
2. Run one past episode through a show-notes tool. Compare it with what you published, and note every error it made. That list is your review checklist.
3. Cut the next interview from the transcript, then listen back to every edit that touches a claim or a qualifier.
4. Replace Apple's auto transcript on that episode with your corrected version, and check the attribution changes.
5. Draft the sponsor recap from real exports.