The embarrassing failure in a generated episode is rarely the voice reading "bullet, bullet, bullet." It is the voice reading "TODO: ask Priya about the outage" in the same warm tone as the story of the outage itself. That note was for the host. Send raw show notes to a voice and the listener hears every note you left for yourself.
A podcast script for text-to-speech has to be the episode, not the plan for the episode. VoxAnvil's Author Podcast Studio, on Starter and Pro, is a separate workflow from a book's chapter list. Episode formats there include conversation, solo, and interview, among others. Those labels do not write the script. You still have to hand the studio words that should be heard.
Start from the listener's first ten seconds
Write the cold open as something a person would say if they sat down late. No "welcome back to episode 47" until after you have given them a reason to stay. One concrete scene, then the question the episode answers. Thirty to forty-five seconds is enough. Generate that open first, with the voice you intend to keep, and listen to it before you write the rest out loud. If the open is cute and unclear, the episode will be too.
Then write the body as short sections with spoken titles only if you would actually say them. "The night of the outage" can be said. "Segment 2" should not. Put timestamps and editor marks in a separate rundown. The spoken file should still make sense if the timestamps are deleted.
Show notes are not the same as the script. Show notes can hold links, spellings, and a summary for people who will not press play. The script holds the sentences. If a link matters in audio, say the name of the site in words.
Names, guests, and the second voice
Say a guest's name, and say how you want it said, the first time they appear. Store that spoken form if you will have them back. A guest who is "Priya Raman" in the notes and "Priya" in the audio needs that choice made on purpose. Do not leave an email address or a social handle in the spoken file unless you want it read.
A solo episode is one voice. An interview script can still be one voice if you are narrating a conversation you already had: "Priya told me the alert fired at 2 a.m." That is often cleaner than pretending a text model is the guest. If you assign a second voice, you are claiming the listener should treat that voice as a person. Use a voice you can tell from the host in the same room, keep it for that guest across episodes, and do not put the guest's voice on the host's jokes.
Character casting on Pro can store voice assignments. The podcast tools and the book tools are related, not identical. Do not assume a chapter credit and an episode are the same unit. Check the plan you are on. Starter and Pro include the podcast studio, an RSS feed, and a website embed. The free plan is aimed at trying chapter audio. If you need the feed, you need the plan that lists it.
Consent still applies. Do not clone a guest's voice because you have an email from them. Voice cloning asks you to attest that you are authorized to use the recording. A chat message that says "sounds good" is not a recording and may not be permission. When in doubt, narrate the interview in your own authorized voice or in a stock narrator, and quote them as a quote.
Clean the read like a script, not a post
Delete greetings you would not say twice, throat-clearing, and "in today's episode we dive into." Say the thing. Keep one signpost near the middle if the episode is longer than ten minutes: "That is the cause. Here is what they changed." Signposts are for the ear. They are not a table of contents read as a list of nouns.
Numbers in audio need spoken forms. "2 a.m." can stay if you have heard it said correctly, or you can write "two in the morning." Dates, versions, and error codes are where shows get sloppy. The numbers guide is the checklist. For a podcast, also decide whether you say the numbers at all. An error code can be "the auth error" in the story and the exact code in the show notes.
Leave expression tags off the open. A host who starts every episode on [excited] sounds like a commercial by week four. A single [pause:0.5s] before the answer to the episode's question is enough direction. The series voice note is how you stop the open from drifting.
Generate, then listen at section breaks
Generate the open, then each section, in order, in the same voice. Play across the boundary. Cut repeated sentences. If you pasted the open's last line into the body as a reminder, the voice will say it twice.
Listen once at normal volume while you do something else with your hands. A podcast is heard in kitchens. If a sentence only works when you stare at it, rewrite it. Then export or publish through the path you actually use. An RSS item with the wrong audio file is a production error, not a writing error. Play the file from the feed, not from the project, before you tell anyone it is live.
Work it through on a 12-minute solo episode about a failed deploy. The rundown has four bullets and a link to a status page. The spoken script has the open, three sections of about three minutes, and a close that tells the listener what to check first tomorrow. The status URL stays in the notes. The voice never says "todo." The open, played beside last week's open, is the same host at the same pace. That is a finished episode.
Frequently Asked Questions
Q: Can I generate a podcast from show notes alone? A: You will get a voice reading your notes. Rewrite the notes as sentences you would say, then generate. The studio does not turn a bullet list into a story by itself.
Q: How is a podcast episode different from an audiobook chapter? A: A chapter is part of a book project and is billed as a chapter credit, typically up to 2,500 words. A podcast episode is the podcast workflow, with its own script and, on Starter and Pro, feed and embed tools. Do not paste a chapter into the feed and call it an episode unless you meant to publish that audio.
Q: Should the trailer use a different voice? A: A trailer can, if you want it to sound like an ad. The episode should use the series voice. Do not let the trailer voice leak into the cold open.
Q: What belongs in the cold open? A: A specific moment and the question of the episode. Not your biography, not a list of platforms, not a request to subscribe. Ask for the subscription after the listener has a reason.
A good cold open gives the listener one specific moment and the question the episode answers. The name of the show can wait until they care.