Audible's AI Features Are Built Around the Cast

On October 1 Audible announced three AI features for its player: Character Guide, Interactive Story and Visual Explorer. None of them narrates a book. Two of the three are about characters: one shows who is speaking, in real time, while you listen, and the other lets you talk to a character and hear it answer. Both arrive on productions with a full cast. Our read: Audible is rebuilding its player around the character, and that says what the platform now takes an audiobook to be made of. It is made of voices that belong to someone in the story. One narrator reading everything is no longer the assumed shape.

For two years the AI question in audio has been who reads the book. This announcement answers a different one: what a listener can do with a story once it has a cast.

Everything in the table is from Audible's release. Andy Tsao, Audible's chief product and analytics officer, says in it that AI lets the company build interactive experiences that listeners, younger ones especially, are asking for. TechCrunch reads the launch as a response to new competition from Spotify.

The most useful sentence in the release is the one explaining why Character Guide exists. Audible says its customers report that keeping track of a large cast of characters is harder in audio than in print, where the page gives them something to look at.

That is the listener's side of a production problem every author of a dialogue-heavy book already has. On the page, a name and a paragraph break tell the reader who is talking. In audio, only the voice can.

Audible's answer lives on the screen. The older answer lives in the production: give each character a voice of their own and keep it the same from the first chapter to the last. Listeners ask for exactly that. In Edison Research's 2026 listener study, the one behind our post on blinded listening tests, 81% of frequent listeners said they are interested in distinct voices for each character.

The two answers are not rivals. A card that puts a name to a voice helps most when there are many voices to tell apart. But a screen only helps while the listener is looking at it, and the voice works with the phone in a pocket.

Look at where the two character features launch. Dracula is the adaptation Audible released on October 1, directed by Mary Lambert with a cast led by Jonathan Bailey and Maggie Gyllenhaal; we covered it last week. 1984 is Audible's 2024 dramatisation with Andrew Garfield, Cynthia Erivo, Andrew Scott and Tom Hardy. Exoplanet, due November 5, has a cast led by Anthony Mackie, Ariana DeBose and John Leguizamo, and was the subject of our August post on what premium audio now means. The remaining title is a nonfiction book, and it gets the feature that is not about characters.

The design itself makes the same point. A character card pairs each character with the performer behind the voice. That presumes more than one performer. On a single-narrator edition every card would carry the same face.

This is the third signal since August that points the same way. First Audible commissioned Exoplanet as produced audio: cast, scored, designed. Then publishers began re-recording books they already sell as full-cast editions. Now the player is being built for a cast. Format, then catalogue, then interface.

None of the three features reads a line of the book. In the three fiction titles the performance belongs to named actors, and the AI sits around it: labelling who is speaking, answering questions about the production, extending one character into a conversation.

That is worth saying plainly, because the public argument about AI and audiobooks is almost entirely about narration. Marketplace's report on September 30 is a fair map of it: narrators, their union and the companies selling synthetic voices, all arguing about who reads. One day later Audible put its AI somewhere else.

We use the technology for a different job. Audible puts it around a studio cast; we use it so that a book without a studio budget can have a cast at all. The jobs are different and the conclusion about the listener is the same: people follow characters, and characters need voices.

Interactive Story is the new thing here. A recorded performance is fixed: an actor said those lines once. A character that reacts in real time to whatever the listener says is speaking lines nobody performed. Audible's release says creators define each story's opening scenes, turning points and the character's persona, and that its product design keeps the experience faithful to the creator's vision. It does not go into how the character's voice is produced.

The point is general, and it is not a question about Audible. Once a character can say new lines, the voice itself is the asset, separate from any recording of it. The rule for that asset is being written this season, and it is consent. Jeff Bennett, SAG-AFTRA's chief legal officer, told Marketplace that for digital replicas of performers "it really revolves around informed consent": a performer who knows where and how the voice will be used can make that decision for themselves. On September 30 the Tokyo District Court held that a person's voice can be protected by publicity rights, a first for Japan, in a suit a voice actor brought over videos narrated by an AI imitation of him. The court also dismissed his request to have the videos deleted, because they had already been taken down.

We made the same argument from the UK side in August, and it is how the product works. In a Midsummerr full-cast production every character gets a voice cast for that book, and using a real person's voice requires an explicit confirmation of rights and consent before that voice is made.

Start with what it does not change. These features are in beta on Audible's own Originals. An author cannot switch them on for a title, and ACX, Audible's self-publishing route, still requires human narration unless Audible authorises otherwise. Nothing in this post is a claim about where a given audiobook can be sold.

What transfers is the assumption. The listening experience is being designed around characters, so the first question for a manuscript is whether a listener can tell its characters apart by ear. Read your own book for dialogue density and speakers per scene before you decide; not every book needs a cast. If yours does, how many voices it needs is a casting decision with a real answer, and in a series the cast has to hold from one book to the next.

Then the cost, because that is the part that used to end the conversation. On Midsummerr the Full Cast format, a distinct voice for every character with the dialogue assigned to the cast, is $3.75 per thousand words. Full Production adds music, sound effects and the mix at $5 per thousand words. For a 90,000-word novel that is $337.50 or $450. You can review and change any voice on the cast sheet, and the rights stay with you. Frankenstein is free to stream if you want to hear a gothic novel produced that way before you compare anything.

Three caveats. The features are a limited beta in two countries on a handful of titles, and we have not used them, so this post makes no claim about how well they work. Early coverage is divided on whether listeners want a conversation with a character at all (Engadget has the doubtful end of it). The claim here is about what Audible chose to build on, which is the cast.

The productions carrying these features have names an independent title cannot hire, and names bring an audience. That is real, and it is not the format.

And a character guide is a retailer's feature. No author controls whether a player shows who is speaking. The voice is the part of that problem you do control.

• The full-cast edition just became the premium edition: the publisher side of the same shift.
• Listeners don't prefer AI voices, they prefer a cast: the blinded test and the 81% figure.
• How many character voices does an audiobook need?: the casting question a character guide exists to solve.
• AI voice cloning, the law and consent: where the consent line comes from.
• Full-cast vs single-narrator audiobooks: the trade-offs in one place.