None of this asks you to write differently for the page. The habits that make a scene easy to follow by ear are the same ones editors have always asked for. In audio, they decide whether a line reaches the right voice.

Below: why tags carry more weight in audio, the habits that make attribution clear, whether an existing manuscript needs changes, and how to check the cast before you generate anything.

Why do dialogue tags matter more in audio?

Because the speaker has to be decided before a line is voiced. In a full-cast audiobook each character has a voice of their own, so someone, or something, must work out who says every line.

AudioForge is a production tool that turns a manuscript into a multi-voice audiobook, with a distinct voice for every character. Its parser reads the prose and assigns a speaker to each line from the same clues a reader uses: tags, names, paragraph breaks. When those clues are missing, the line is marked Unsure and left for you to assign.

Clear attribution therefore saves you review time. The fewer lines the parser has to guess at, the fewer you have to check.

Which habits make the speaker clear?

Five small ones cover most scenes.

1. Name the speaker early in a scene. The first line or two should leave no doubt about who is present and who is talking.
2. Keep one speaker to a paragraph. A new paragraph signals a new speaker. Two people sharing a paragraph is the most common cause of a misattributed line.
3. Anchor long exchanges. After three or four untagged lines, add a tag or a beat. With three or more people in the room, tag almost every line.
4. Use names when pronouns collide. Two characters who are both "she" need their names more often than feels natural on the page.
5. Mark thoughts as thoughts. "She thought" or a consistent convention keeps inner voice apart from speech.

Here is the same exchange written two ways.

Do you need to rewrite your manuscript?

No. A manuscript written for readers usually works as it stands, and the fix for a misattributed line belongs in the script, not in the book.

When the parser assigns a line to the wrong character, you change the speaker on that line in the Director Studio. Your manuscript stays exactly as you wrote it. The help center shows where that happens in the workflow.

The habits above pay off most for chapters you have not written yet. If you publish a serial and plan to release audio alongside it, a tag here and a name there costs nothing while drafting and saves a correction later.

How do you check the cast before generating audio?

Go to the uncertain lines first, then read one crowded scene. That finds most problems in a few minutes.

• Jump through the lines marked Unsure and assign each one.
• Pick the scene with the most speakers and skim it line by line. Crowded scenes are where attribution slips.
• Look at any exchange with no tags at all for a long stretch.
• Generate that one scene and listen. A wrong voice is obvious at once.

If a line still comes out in the wrong voice, change its speaker and regenerate it alone. Unchanged lines are reused, so a correction never means redoing the chapter. You can try this on a chapter of your own from the home page.

The bottom line

Write so that a listener always knows who is speaking: name the speaker early, give each speaker their own paragraph, and anchor long exchanges with a tag or a beat. Your existing manuscript does not need rewriting, because a misattributed line is fixed in the script. Check the uncertain lines and one crowded scene before you generate, and the cast will be right.

AudioForge is a production tool that turns a manuscript into a multi-voice audiobook, with a distinct voice for every character. The steps below are the same ones shown in the help center, with what to look for at each stage.

Below: what you need to start, how the script is made, how casting and direction work, and how the finished chapter comes out.

What do you need before you start?

A chapter of your manuscript and a free account. You create a book, then add text in whichever way suits you.

• Paste a chapter straight in.
• Upload a chapter as a .docx or .txt file.
• Import a whole manuscript as .docx, .txt or PDF. It is split into chapters, which you can rename, reorder, split or join before anything else happens.

Each chapter becomes its own audio file. That is why serials are a natural fit: you can produce a chapter as soon as it is written.

How does a chapter become a script?

The parser reads your prose and writes it out as a screenplay. Every line gets two labels: who speaks it, and how it should be delivered. A nervous aside and a shouted warning from the same character end up with different directions.

The parser is not always certain. When it cannot tell who said a line, the line carries an Unsure tag, and "Next uncertain line" takes you from one to the next. Checking those lines is the most valuable few minutes you will spend, because a wrong speaker is much easier to fix in the script than to notice in the audio.

How do you cast the voices?

You pick a voice for the narrator and for each character from the Voice Library. Casting is saved for the whole book, so a character who appears in chapter one sounds the same in chapter forty.

Two things help with a long series:

• Mark voices you like as favourites so you can find them again.
• Add invented names to the book's pronunciation glossary once, so they are said the same way every time.

How do you direct a line?

You edit it. In the Director Studio any line can be changed: the speaker, the delivery, the text itself, or the pause after it. What is in the script is what gets voiced, so a correction on the page is a correction in the audio.

You do not need to direct every line. Read through a tense scene, adjust the deliveries that matter, and leave the rest. Edits save automatically, and undo and redo work as you would expect.

How do you get the finished audio?

You generate the chapter and download one file. Generation runs in the background, so you can leave the page and a notification tells you when it is done.

Listen to the result. If a line sounds wrong, change it and regenerate. Only the lines you changed are produced again, and everything else is reused.

The bottom line

A full-cast chapter is four steps: parse, check, cast, generate. The parser does the heavy lifting and flags what it is unsure about, so your time goes into decisions rather than recording. Start with a single chapter, listen to it, and fix individual lines until it sounds the way you hear it. You can try it with one chapter for free during early access.

Rules in this area change quickly. Every claim below links to the store's own page and reflects what that page said in October 2026, so follow the link before you plan a launch around it.

The question matters most if you produce the audio yourself. AudioForge is a production tool that turns a manuscript into a multi-voice audiobook, with a distinct voice for every character. You download the finished files and decide where they go, which means the stores' rules decide your options.

Below: the short answer in a table, what each store asks for, how to disclose the narration, and where else the audio can go.

Which stores accept AI narration?

Two say yes in writing, one says no, and two only describe programs of their own.

What do Spotify and Kobo ask for?

Both ask you to say plainly that the voice is synthetic.

Spotify's help page on digital voice narration says it accepts audiobooks narrated this way. When you upload through Spotify for Authors, you select the option "This audiobook uses digital voice narration", and Spotify adds a short sentence to the book's description so listeners know. One limit: Spotify says it does not currently share digitally narrated audiobooks with its referral partners.

Kobo Writing Life's page on AI narration says Kobo accepts it too. The narrator has to be listed as a synthesised voice, next to the author. Kobo does not ask for exclusivity, so the same audiobook can be on sale elsewhere.

Does Audible accept AI narration?

Not if you upload the audiobook yourself. ACX is the service independent authors use to reach Audible, and its audio submission requirements state that a submitted audiobook must be narrated by a human unless otherwise authorized. Unauthorized text-to-speech, AI or automated recordings are prohibited.

That may change. The same page says Audible is working to accept text-to-speech content from other publishers and creators, and will share timing later. Until it does, treat Audible as closed for an audiobook you produced with AI voices. The rule is explicit, so do not try to slip one through.

What about Apple Books and Google Play?

Both sell AI-narrated audiobooks, but the pages they publish describe their own tools. Apple offers digital narration to independent authors and small publishers, produced from an ebook file. Google Play Books runs an auto-narrated audiobooks program that does the same for an ebook already on sale there.

Neither page states a rule for audio narrated with other AI voices. Audiobooks usually reach Apple Books through a distributor, and distributors set their own conditions. One large distributor, for example, accepts digital narration only from providers it has approved.

Before you count on either store, ask the distributor or the store's support in writing whether they will take your files and how they want the narration labelled.

How should you disclose AI narration?

Say it in the place each store gives you, and in your own words as well.

1. Select the store's digital or synthetic narration option when it has one.
2. Credit the narrator as a synthesised voice, never under an invented human name.
3. Mention it in your description, so listeners who care can choose.

Never submit AI narration as human narrated. Distributors spell this out in their terms, and it is the one mistake a clear disclosure cannot fix afterwards.

Where else can the audio go?

Anywhere you control. With AudioForge the audio files are yours, as the IP policy sets out, so you can release chapters to patrons on Patreon, post them on YouTube, or sell the audiobook from your own site.

That suits serial fiction well: a chapter a week, in audio, for readers who already follow the story. Each chapter exports as its own file, and the help center shows the steps from manuscript to download. Stores can come later, once their rules are clear for your book.

The bottom line

Publish where the rule is written down. Today that means Spotify and Kobo, each with a clear disclosure, plus any channel you run yourself. Audible stays closed to AI-narrated uploads through ACX until it says otherwise, and Apple Books and Google Play need a direct question before you rely on them. Decide on stores after you have heard a chapter: producing one is free during early access.

Ask authors what they think of AI audiobooks and you hear the same words: flat, one-dimensional, dull. Listeners notice within a minute, and the book is judged before the story has a chance. The verdict is deserved, but it is a verdict on one way of making audio, not on the idea.

Below: what a single voice cannot do, what changes with a cast, how the work is done, and how to judge the result yourself.

What can a single voice not do?

It cannot show who is speaking. On the page, a reader knows from the tags, the paragraph breaks and the rhythm of the scene. They hear each character differently in their head, and the narration in between has its own calmer register.

Hand all of that to one voice and the structure collapses. Two characters arguing become one person talking to themselves. The narrator's dry aside arrives with the same energy as the line before it. Thoughts are indistinguishable from speech. Every word is pronounced correctly, and the scene still does not land.

A human narrator solves this by acting: shifting pitch and pace between characters, dropping their voice for the narration, leaning into the tense lines. A text-to-speech voice cannot, because all it is given is words.

What changes when every character has a voice?

Three things change at once: who speaks, how they speak, and what you can fix.

• Every character gets a distinct voice. The listener no longer works out who is talking from context, because the voice tells them. The narrator has a voice of their own.
• Every line gets a direction. A line marked nervous is delivered nervously, and a line marked as a whisper is whispered. The same character can be furious in one paragraph and exhausted in the next.
• Every line can be fixed on its own. When a delivery misses, you change that line and regenerate it, not the chapter.

This is what audio drama producers have always done. What kept it out of reach for an independent author was the cost of a studio, a cast and an engineer for a sixty-chapter serial.

How is a full-cast audiobook made from a manuscript?

The chapter becomes a script first, and the script is what gets voiced. AudioForge is a production tool that turns a manuscript into a multi-voice audiobook, with a distinct voice for every character, and the work runs in four steps.

1. Add a chapter. Paste it or upload it. Each chapter becomes its own audio file.
2. Parse it into a script. Every line is tagged with who speaks it and how it should be delivered. Lines the parser was unsure about are flagged for you to assign.
3. Cast the voices. The narrator and each character get a voice, saved for the whole book, so a character sounds the same in chapter one and chapter forty.
4. Generate and export. The lines are voiced, stitched with the pauses you placed, and downloaded as one file.

The parsing step is the one single-voice tools skip, and it is where most of the difference comes from. The help center walks through each step.

How do you know if it is good enough?

Listen before you decide. Scepticism about AI narration is reasonable, and an argument will not settle it. A sample will.

Play the full-cast sample on the home page and wait for the moment two characters speak to each other. Compare it with the last single-voice AI audiobook you gave up on.

Then try a chapter of your own. It will not be perfect on the first pass. The parser can be unsure who said a line, and a delivery can miss. That is why uncertain lines are flagged, and why a single line can be redone without touching the rest. During early access it costs nothing to find out, and you keep every right to the text and the audio.

The bottom line

Flat AI narration is a production choice, not a limit of the technology. One voice reading a whole chapter cannot tell characters apart or change its delivery, so it sounds dull. Give every character a voice and every line a direction, and the same book sounds like a performance. Hear it on your own chapter before you judge.