Is AI coming for audiobooks?

A new audiobook edition of “The Odyssey” came out this June, capitalizing on the summer blockbuster. Few audiobooks get any sort of national attention, but this one carried made-for-internet outrage: it was made with AI.

The narrator was Academy Award-winning actor Michael Caine — or at least, an AI recreation of his voice — along with a cast of AI-voiced characters.

The company behind the audiobook, ElevenLabs, specializes in AI voices. It’s not the first time it has made headlines, either. A couple years ago, it recreated the long-deceased Judy Garland’s voice and later released “The Wonderful Wizard of Oz” AI audiobook with it. It has partnered with numerous celebrities — living and dead — to clone their voices, including Matthew McConaughey, Stan Lee, Maya Angelou, and John Wayne.

One purpose of these celebrity imitations is for ElevenLabs to sell its AI audiobook service — and it’s not the only company interested in the artificial possibilities for the audiobook world.

A growing industry

The Pew Research Center reported that in 2025, 26% of adult Americans said they had listened to an audiobook in the past year. That share had more than doubled since 2015.

The Audio Publishers Association (APA) annual survey reported that audiobook sales grew 9% last year to $2.43 billion. That growth has been pretty consistent since 2021, when audiobook sales jumped during the pandemic.

While those numbers are significant, audiobooks are still one of the smaller players in the publishing industry. The Association of American Publishers found that audiobooks account for about 7.5% of the industry’s overall profit — dwarfed by the print side.

Will Damron has narrated over 800 audiobooks and was recently inducted into Audible’s Narrator Hall of Fame. He has been in the industry for over a decade now, and although it has grown, he said it’s still pretty small.

“Even today, even with how much it's blown up, there's only like maybe a couple hundred — and that's being generous — of us who really do this full time and really make a living at it,” Damron said.

He said that it’s easier to break into audiobook narration than TV or movie acting, but that doesn’t make it easy. It took him two years to break into the industry back in 2012, and he estimated that full-time narrators need to do 50 to 80 audiobooks each year. That’s no simple feat — they put a lot into each project.

“You have to prep the script, and that's what takes the most time,” Damron said. “If it's a fiction book, I make a list of every character that's in it — what vocal notes there are for them, if they have accents, if they have a gravelly voice, if they have a high voice. [I note] any unknown words, names that I might not know, names of places that are hard to find, names that can be pronounced [in] different ways.”

Once the publisher approves that research, the narrator starts recording, producing a couple hours of finished audio each day. Everything gets edited, the narrator records whatever corrections the publisher requests, and finally the audiobook comes out.

Why AI?

Human-narrated audiobooks take around six months to make, from casting to publication. They also require thousands of dollars to pay narrators and producers. AI audiobooks demand far less — they tend to cost closer to $200 or $300 and can take hours or days to finish.

Joseph Nassise is a New York Times best-selling author. He is also the business development director for Spoken, an AI audiobook company which licenses ElevenLabs’ software.

Nassise’s work has been popular enough to warrant at least 40 human-narrated audiobooks. He said that beyond the time and cost, he didn’t like how little control he had over the production.

“The human-narrated process through a major publisher is entirely hands-off on the author’s side,” he said. “Sometimes, we get the opportunity to decide, do we like this narrator or not? [But] that isn't always the case. Sometimes, you're just presented with a completed audiobook.”

On the other hand, he said, AI audiobooks are entirely hands-on. He chooses character voices and checks every pronunciation and intonation himself. He estimates he has made 24 AI audiobooks.

“[It’s about] being able to make sure that the inflection and the way that the story and the dialogue is being presented is correct,” he said. “I want to make sure I'm delivering the best story possible.”

Mind the gap

Jane Love, director of audiobooks for SAG-AFTRA — the union that also represents Marketplace reporters — estimated that roughly 74,000 audiobooks come out in the U.S. each year, compared to around 4.2 million print and ebooks. That gap is massive, and as long as AI companies operate in that space, Love said it won’t affect business for most narrators.

“What we are seeing is not a decline in human production and not really an impact on human production,” she said. “... While I see some [AI] development for sure in this space, it seems to me that at least so far, it’s not the titles for which our members would be hired anyway.”

Of those 4.2 million books that came out last year, 3.5 million are self-published, according to Bowker, the official ISBN agency for the U.S. That represents a 38.7% increase from the 2024 — roughly a million books — likely due to an AI-induced self-publishing frenzy.

But that still leaves a large volume of human self-publishers. For the time being, those authors are the primary target of these AI companies, because they don’t have a major publisher to front the cost of a human-narrated audiobook.

ElevenLabs has its own app called ElevenReader, a subscription-based audiobook service that offers both human- and AI-narrated content. When it came out, it was heralded as a direct competitor to Audible, which offers a subscription service at a similar cost.

The two are hardly comparable, though: Audible features most new and best-selling books, while ElevenReader features a few older or classic titles, along with an extensive catalog of self-published books.

And thus far, AI audiobooks haven’t seen much financial success. The APA’s report found that AI audiobooks made up about .03% of the audiobook industry’s revenue in 2025 — about $729,000.

Attack of the AI voice clones

ElevenLabs’ voice clones are not just for celebrities. The company also advertises its AI software as a tool for voice actors to make money off of their voices without having to do the work of recording. Spoken is among the companies who license some of those clones.

“We hire narrators, and we pay them for the use of their voice, and every time that voice gets used, they get paid,” Nassise said. “They can be paid and credited on multiple projects that [aren’t] impacting their time.”

Performers pay ElevenLabs at least $11 to clone their voice and provide 30 minutes to three hours of speech. After that, ElevenLabs said that most performers will make $0.03 per 1,000 characters, which comes out to roughly $1.20 per hour of use.

The company calls this passive income, but it takes a while just to break even with the money and effort spent to make the clone. Once in the library, performers don’t always know what projects their voices will be used for ahead of time; ElevenLabs didn’t tell performers that their voice clones would be used for characters in “The Odyssey” until they got paid.

Jeff Bennett, chief legal officer for SAG-AFTRA, indicated that as long as actors know what they’re getting into, the union has no issue with them selling clones of their voices.

“It really revolves around informed consent,” Bennett said. “If somebody is given the opportunity to do something like this [and] they know exactly what their voice is going to be used for, how it's going to be used — the mediums and the content that it's going to be used in — we're very comfortable that people can make that decision for themselves.”

Love added that performers should only agree to union contracts, where SAG-AFTRA could protect them if the company involved breached the agreement. Some narrators are interested in these digital replicas.

“We represent a really large and diverse constituency,” Love said. “Some people are like, ‘absolute no, I'm never doing it.’ Other people are AI-curious.”

Damron falls into the former category.

“The moment we all start selling off clones of our voice,” he said, “we devalue the integrity and the economic value and the artistic value of what we're doing.”

AI voice ethics

Like most of the industry, voice companies made their AI omelets by breaking a whole lot of eggs.

In 2024, ElevenLabs’ technology was used to make an AI clone of then-President Joe Biden’s voice to urge people not to vote. The company quickly banned the account responsible, made a list of off-limits voices of public figures, and added a consent check to its cloning process. It also now moderates content to prevent criminal use, but its privacy policy specified that it doesn’t review everything.

In April, Senator Maggie Hassan requested information from AI voice companies, including ElevenLabs, about what they were doing to stop scammers after the FBI reported that in 2025, AI-related scams totalled over $893 million in losses.

In May, a group of Chicago journalists, voice actors, and audiobook narrators sued nine big tech companies, including ElevenLabs, for using their voices to train AI models. They filed the suit in Illinois, which has biometric data laws that protect a person’s vocal imprint. Bennett said that SAG-AFTRA is closely monitoring cases in Illinois and is in contact with at least one firm involved in many of those suits.

“We expect there might be some decisions coming before the end of the year that will give us some more insight into whether or not that's an angle that's worth pursuing,” he said. “And certainly, because we do a lot of policy work, if it looks like those are going to be successful, then that’s a law we would look to potentially bring to other jurisdictions to help people protect their voice rights from that initial stage of ingestion and training.”

It’s not the first time ElevenLabs has been sued, and it’s certainly not a first for AI companies in general. Bennett estimated that there have been around 130 AI copyright lawsuits. Although Nassise works with a company that uses ElevenLabs’ technology, he’s sympathetic to the legal grievances.

“I'm part of the Anthropic lawsuit,” Nassise said. “They used 29 of my books to train this stuff without permission. … I think that was wrong, but the tool is here, and my view has always been, ‘Hey, use the best tool for the job.’ And so I don't see any difference in a human-narrated audiobook versus an AI-narrated audiobook, because what comes out the back end? An audiobook with a story people can enjoy.”

Anthropic settled the suit, and Nassise will be a part of the $1.5 billion payout. He said that Spoken’s approach to AI voice is very ethical and does not allow AI to train itself with any material he uploads.

For voice actors working with ElevenLabs, that isn’t the case. ElevenLabs uses the personal data that users upload for voice clones to train its models.

Nassise also agreed with Love and Bennett that digital replicas should be properly labeled.

“Right at the start of my audiobooks they all say, ‘Made with Spoken,’ and the listings on the various retailers … say ‘Created with a digital voice,’” Nassise said. “Okay, great. Now, the consumer knows what they're getting, and if they like it, terrific. They don't like it, terrific. That's entirely up to them. But I think it's important to identify that fact — to be upfront about it.”

Part of a broader AI scheme

Companies like Spoken and ElevenLabs don’t necessarily need AI audiobooks to be profitable right now, because they’re selling subscriptions to make this content, rather than the content itself.

And while Spoken is focused solely on AI audiobook production, ElevenLabs is not. Its audiobooks and celebrity endorsements tend to make the most buzz, but they’re just a piece of what ElevenLabs offers with its voice model.

“The voice model is the engine; the voice model is not the application,” said Nick Patience, vice president and practice lead for AI platforms at The Futurum Group. “If you look at [Large Language Models] … they are valuable, but they are not the end user tool. You have to build other things around them to make them useful.”

That is exactly what ElevenLabs has done. The company recently posted a $22 billion valuation with $500 million in revenue, making it one of the world’s most valuable AI labs. It hasn’t gone public yet — and declined a request for an interview for this story — so it doesn’t have to report how it makes its money, but an AI audiobook industry that racked up a measly $700,000 last year is probably not powering that growth.

“From what I can see,” Patience said, “it's the voice agents used across customer service, customer support, hiring, sales — all those kinds of things — that's really growing the business recently.”

ElevenLabs has held two summits in the past year, one in San Francisco and one in London. They both featured the tagline “Technology leaders on the future of voice and agents.” Neither lineup featured a single speaker from the audiobook industry.

It remains to be seen what these AI voice companies will do, but Patience said that investors want to see a return on their money.

“There's only two ways that really happens: you either get bought or you go public,” he said. “I'm sure that's what they want to happen. It's just a narrower business than OpenAI and Anthropic have.”

Patience said that ElevenLabs is focused on making AI voice models and almost nothing else — and it’s exceptionally good at it. That could make it a nice acquisition target but could also mean it has a better chance at survival than these full-stack AI companies, because it isn’t trying to do everything.

What listeners want

The APA’s consumer report noted that 16% of audiobook listeners had listened to an AI audiobook, but 61% were willing to try it — although that figure was down from 70% the year prior, amid a broader public backlash against AI. ElevenLabs sponsors the APA, and Jim Dinegar, executive director of the APA, declined an interview request.

AI audiobooks have earned mixed reviews, in part because many of them are low-quality pirated versions released on Youtube. But even the nicer ones haven’t escaped criticism. Critics credited ElevenLabs’ “The Odyssey” for the quality of Michael Caine’s AI voice but said emotional scenes fell flat.

Right now, Love doesn’t think that AI audiobooks can perform to the same standard as human narrators.

“I'm a voracious audiobook listener, and I can tell you it just does not compare yet,” she said. “I mean, there's just not the same nuance [or] emotion. … Sometimes it sounds like the [AI voice] doesn't know what [it’s] talking about.”

Spoken funded a survey this year which found that the number of respondents willing to listen to multi-character AI audiobooks more than doubled after they heard a sample, compared with a human voice. It also found that 61% of listeners could not distinguish an AI narrator from a human.

“The quality of the voices themselves are just fabulous,” Nassise said. “We're a long way from the Microsoft read [aloud] kind of approach to voices. What they've achieved in the last several years is just, to me, mind-blowing. … I have been very pleased. And I wouldn't release my works out to the public … if I hadn't been very satisfied with the way they come out.”

An independent, peer-reviewed study from German researchers earlier this year found that listeners slightly preferred human narrators over AI but concluded that the difference isn’t big enough to matter in daily life.

Notably, these surveys only presented a couple minutes of audio to listeners, because of the constraints of an experimental environment. Damron thinks the difference between a short clip and a full audiobook matters.

“It’s a very decent mimic,” Damron said. “It's like a well-trained parrot. And to a lot of people, I do think it sounds convincing enough. It certainly sounds convincing enough if you're listening to small snippets of text. What we do is long-form narrative acting.”

Are audiobooks art?

It remains to be seen whether consumers view audiobook narration like Damron does. The German study suggested that listeners might understand audiobooks “primarily as a functional medium that enables access to the text, rather than as the core locus of artistic creativity.”

In other words, listeners may view narrators as vessels — rather than creators — of art. Damron pointed out that the most dedicated audiobook listeners do not think this way.

“The fans follow authors and narrators,” Damron said. “They follow the voice that’s interpreting the work to them the same way that you might be like, ‘I really want to see a Matt Damon film. I don’t want to see an AI Joe film.’ You follow the artist that you want to follow.”

While Nassise does all his own writing without AI, he views audiobooks as a communication channel for his creative work, and he thinks major publishers will eventually catch onto it.

“AI is just a tool like anything else,” he said. “I wrote my first novel long-hand on legal pads with a Bic pen. … We didn't even have personal computers back when I started this process. I wrote on a Brother word processor, and then I wrote on a personal computer — all of that just becomes another tool in the process. And I view AI the same way.”

But Damron is adamant that AI cannot replace human narration.

“There's something uncanny about it,” he said. “It's that soulless thing. … When I put something out in the world to you, whether it's in written form or in vocal form — or as an actor in a performance, or I sing something — and it's true, it resonates with you. We all feel that. That's what we're all drawn to in our favorite music, our favorite voices — and that is why every narrator who does this got into this business. … I don't think that is the purpose behind anyone making AI audiobooks.”