If you have priced an audiobook this year, you have heard some version of the number: AI cuts the cost by around 90%. A white paper released this month by Frankfurter Buchmesse put that number next to what the people doing the work actually report, and the two are not close. As Publishing Perspectives reported on September 7, technology providers may promise 85–95% cost savings; respondents reported 20% and 50% in the second and third year of adoption.
The easy reading is that the technology under-delivers. We think that reading is wrong, and the study says why in its own pages. The saving went missing in the workflow, not in the voice.
What the white paper measured
Who is Narrating the Future? AI Adoption, Use and Outlook in Audio Production was researched and written by Dosdoce.com and published by the Frankfurt Book Fair, a trade fair with an obvious interest in the subject. Dosdoce spoke to 85 industry experts: half from audio production studios, 36% from publishers and 14% from streaming platforms or distributors, across Europe, Latin America, North America, Africa and the Middle East. It covers audiobooks, audio dramas and podcasts. Dosdoce founder Javier Celaya told Publishing Perspectives that respondents shared data only on condition that it be anonymized and aggregated, so treat every figure as a survey answer, not an audit.
| Finding | Figure |
|---|---|
| Cost savings technology providers may promise | 85–95% |
| Cost savings respondents reported, second and third year of adoption | 20% and 50% |
| Use AI somewhere in their audio production chain | 85% |
| Have integrated AI across more than six production phases | 17% |
| Focus entirely on human-only production | 15% |
| Cite voice quality and rights transparency when choosing a vendor | 86% |
| Cite data privacy when choosing a vendor | 79% |
| Expect disruptive or moderate growth attributable to AI within three years | 86% |
Source: Dosdoce.com for Frankfurter Buchmesse, as reported by Publishing Perspectives, September 7, 2026.
Ready to try it on your own book?
Start your first chapter free →The study explains its own gap
Two rows in that table belong together. Eighty-five percent of respondents use AI somewhere. Seventeen percent use it across more than six phases of production. Celaya put it plainly: "The headline number and the operational reality are two different stories."
Publishing Perspectives reports that the white paper draws the connection itself: the difference between promised and reported savings "may be due to using AI in one or two stages of the production process not in the complete production workflow." Most of the industry has automated a step. Very little of it has automated a production.
Why one stage cannot pay for the rest
An audiobook is not a narration. It is casting, direction, performance, editing, proofing against the text, pickups, mastering and a final quality check, with a hand-off between each. Narration is the most visible line on that budget. It is not the whole budget.
The arithmetic is unforgiving. Suppose, purely for illustration, that narration is 40% of what a studio spends per finished hour. Replace it with something 90% cheaper and the total falls 36%. Make the narration free and the total falls 40%, and not a point more. A stage can never save more than its own share, and that illustrative 36% sits squarely inside the range the respondents reported.
Partial automation also adds work back. A synthetic read dropped into a human workflow still has to be directed, checked line by line, corrected and mastered by people working at human rates. The vendor's 90% is a true statement about one line of the invoice. The studio's 20% is a true statement about the whole invoice.
The other end of the curve
The 85–95% class of saving does show up, and it shows up where the whole pipeline was rebuilt. Pocket FM, which makes serialized audio fiction rather than audiobooks, told TechCrunch on September 10 that AI is used in 99% of its new content, that production is about 80 times cheaper, and that 100 hours of audio that once took about a year can now be made in a day. Those are chief executive Rohan Nayak's own figures, given in an interview, not audited numbers, and the format is not ours. The shape of the claim is the point: the large number belongs to a company that owns every stage, not to one that bought a voice and kept everything else.
What this changes when you price an audiobook
Ask what arrives at the end. A price for narration is a price for one stage. The question that predicts your real cost is what you still have to pay for after the files land: editing, proofing, music, mastering, a retail-ready export. If the answer is "most of it," you are in the reported column, not the promised one, whatever the pitch said.
Buy the production, not the step. This is the argument for a finished audiobook produced end to end, and we can make it from the position of doing it. On Midsummerr a manuscript goes in and a cast, scored, mixed production comes out at one price: $5 per thousand words for the full production, so a 90,000-word novel is about $450. There is no second invoice for the stages around the voice, because they are not separate stages you staff. Our production cost breakdown sets that against human studio rates line by line, and The Murder of Roger Ackroyd is free to stream if you want to hear what the money buys before you compare anything.
Cost is not the only column. The respondents rank voice quality and rights transparency (86%) and data privacy (79%) among their vendor criteria, and half of the barriers to scaling come from limits set by literary agents, publishers and authors. Celaya's closing line is the one to keep: "Trust, once spent, is far more expensive to rebuild than any single title was to produce with AI tools." A cheap production that listeners abandon as unfinished work saved nothing.
The honest version
This is 85 anonymized respondents, not a census, published by an organisation that hosts the industry it surveyed, and "20% and 50%" is what people said about their own books, not what an accountant found in them. The direction is what matters, and it is consistent from both ends: the saving scales with how much of the workflow changed. Eighty-six percent of the same respondents expect AI to grow the market within three years, by making backlist, academic, niche and non-English titles viable that never were before.
They are probably right. But that growth will not come from a cheaper narrator inside the old production line. It will come from productions that no longer have one.
Further reading
- Audiobook production cost: human vs AI in 2026: the line items, with sources.
- Audiobook titles grew 43% last year. Revenue grew 9%.: why a crowded catalogue makes production quality the differentiator.
- Listeners don't prefer AI voices. They prefer a cast.: what the blinded listener research says about format.
