How to Make an AI Audiobook: Pick the Store First, Then the Tool

The decision that shapes everything is which store you want to sell in, because each store takes different tools and different file packages. Pick the store first, then pick the narration tool that store accepts, then follow the same short workflow: clean the text, generate, fix, listen to all of it, package, label, upload and check the live listing.

Which store you sell in decides your tools

As of 2026-11-08, each major route has its own rule about AI audio, and the rule decides which tools and files work. This table gives the rule in one line. Each store page holds the detail, and the cross-store comparison puts them side by side.

Two consequences follow. If Audible is your main market, an AI file is shut out of ACX by default, so read the human versus AI cost page before you spend anything. INaudio's FAQ says distribution is limited to partners that accept digitally narrated titles and names none.

Four questions that choose the route

These four answers narrow the route before you pay for a plan, and each one removes options.

1. Must the book be on Audible through ACX? If yes and you hold no KDP invitation, stop here. ACX's page requires human narration unless otherwise authorized, so your realistic choices are a hired narrator or your own recorded voice.
2. Do you hold the audio rights to an in-copyright book? Google requires audio rights for auto-narration. INaudio's digital voice workflow, KDP Virtual Voice and Apple's program all exclude public-domain titles. If a publisher holds any of your rights, read that contract before you generate anything.
3. How far does the AI edition need to travel? Spotify says it does not currently share digital voice titles with its referral partners, so that sharing will not widen your reach. Reaching other stores needs a separate route, and the distributor comparison sets those side by side.
4. Will you touch the audio outside the tool that made it? If you plan to master, trim or swap a file, INaudio's digital voice route is closed to you. Choose a route that takes ordinary audio files.

Next, match the route to what you must supply. This reflects the pages read on 2026-11-08, and "not restricted" means only that the pages read do not forbid it.

Check the text itself against the store

The book's words can trip a rule even when the voice passes. KDP's content guidelines require disclosure of AI-generated text, images or translations, and Draft2Digital's content guidelines (October 27, 2025) do not accept content generated entirely by AI that has not had extensive human editing. Apple's formatting guidelines and Author's Republic's FAQ both treat AI-written content as a separate label from AI narration. Know your answer before you reach the upload form.

Step 1: Clean the manuscript for the route you chose

Every route starts from your text, and most start from an EPUB. Google needs a live EPUB, Spotify's creation tools take EPUB only, and KDP Virtual Voice needs a reflowable ebook with a navigation table of contents. ElevenLabs' Studio docs call EPUB the best import format and say each chapter heading should be formatted as Heading 1 so chapters split automatically. They also list PDF, DOCX, TXT, HTML and URL imports. The manuscript preparation checklist covers the cleanup, with the sourced requirements first and practical checks after.

Do this on a copy, and keep the audio edition's text separate from your ebook file. Spoken-only changes should not leak into the book you sell.

Step 2: Generate the narration

Generate one chapter before you commit the whole book, using the method in the same-chapter comparison. A dialogue-heavy chapter with your hardest names shows more than chapter one does. How you generate after that depends on the route.

• Spotify's own tools: per Spotify, the Audiobook Creation Tools are for the US, English and EPUB only, with a limit of five audiobook projects per account and a US payment profile. You pick one voice for the whole book. Our Spotify creation tools review covers the route.
• ElevenLabs Studio: ElevenLabs' docs, as listed on 2026-11-08, give limits of 500 chapters per project, 400 paragraphs per chapter and 5,000 characters per paragraph. Credits are the cost driver, so check what ElevenLabs costs per book before you pick a plan. See ElevenLabs plans.
• Google Play auto-narration: Google's page says there is no program fee during the Beta, and you must own the audio rights. You can download the result as a .zip or .lpf.
• KDP Virtual Voice: only if you hold an invitation. KDP asks participants not to share beta details, and we are not participants, so this page stays at rule level.

Three ElevenLabs details, from its Studio docs and voices overview as read on 2026-11-08, change how you plan:

• Pick a voice that will last. The overview says all Default voices expire on December 31, 2026 and will then be inaccessible. Do not start a long book or series on one without checking its replacement.
• Bulk export has its own retries. The docs say exporting a whole chapter or project checks for problems such as mispronunciations and missing words, and regenerates up to twice at no extra cost. That reduces errors, but it does not replace listening.
• Free retakes have conditions. Up to two regenerations are free if you do not change the text, voice or voice settings. After that, generating costs credits.

If you want character voices or your own voice, read multi-voice fiction audiobooks or voice cloning for authors at this point, because both change the cost and the labelling.

Step 3: Fix pronunciation and pacing

Names, invented words and long sentences are the usual things to check; we have no measured error rates, so budget for corrections before you export. Pronunciation and pacing fixes lists the documented controls per tool. If you go through INaudio, make fixes inside the generation tool, because its FAQ requires an unmodified package.

Step 4: Listen to the entire book

Proofing means listening to all of it. A 90,000-word book is about 9.7 finished hours by ACX's 2013 rule of thumb of 9,300 words per hour (90,000 ÷ 9,300), so a single full listen takes at least that long, before any retakes. Use the proofing checklist to know what to listen for. On Spotify's tools you can change the voice until you submit, and after publishing you need a new version, so proof before you press publish.

Step 5: Export and package for the store

The package is where routes diverge, so check the current instructions before you export.

• ElevenLabs to Spotify for Authors: ElevenLabs' help article describes Export, then Publish, then Spotify in Studio, which produces an MP3 you upload through Spotify for Authors. You choose a free-preview chapter and can edit the narrator credit. Instant voice clone projects must be verified first.
• ElevenLabs to Voices by INaudio: ElevenLabs' older blog post (February 20, 2025, with a later note that exporting has changed) describes downloading an LPF file and uploading it as Digital Voice Narration. Its current Studio docs list a separate Export, Publish, InAudio path that generates an LPF. INaudio's FAQ still requires an unmodified LPF. ElevenLabs' descriptions differ by date, so confirm which options your account shows before you pick.
• Google Play to wider distribution: Google's download page says LPF files let you distribute through INaudio.

Check export quality before you pay for a plan, because ElevenLabs' own pages disagree. On 2026-11-08 its pricing page plan cards added 192 kbps at Pro ("Everything in Creator, plus ... 192kbps") while its comparison table showed "128 & 192 kbps" from Creator up, and its Studio docs gave new projects 128 kbps MP3 on Free, Starter and Creator, and 192 kbps MP3 from Pro up. The pages disagree or the table is ambiguous. If your route needs 192 kbps, confirm it in Studio's Export menu, where the docs say hovering over Format shows a project's quality, before you generate a whole book on Creator.

The same docs say uploaded audio "cannot be published to distribution platforms". If you add a human-read intro or an imported music file inside Studio, confirm the Publish route still works before you rely on it.

For bit rates, loudness and file rules per store, see the audio specs guide.

Step 6: Label it and upload

Labelling is a store rule, not a courtesy. On Spotify you tick "This audiobook uses digital voice narration" and Spotify adds a sentence to the description. On Kobo you list the narrator contributor with a synthesized-voice label. INaudio sets the narrator credit and adds a disclosure itself. Author's Republic asks whether the audiobook is narrated by an AI voice and adds a sentence to the description if you answer yes. The AI audiobook labelling guide collects the wording by store.

Apple's Audiobooks Formatting Guidelines 1.3.2 (sections 1.15, 3.9 and 5.14) require audiobooks narrated by AI to state this in the title or narrator field and to reference it in the description. That is a label rule. No Apple page read says whether third-party AI audio is accepted.

Prepare the rest of the listing while you are there, since these fields differ by route:

Step 7: Confirm the live listing

An upload is not finished until the listing is correct, and each route states a different wait.

When the listing appears, check that the narrator credit and description disclosure are right, the sample plays, the chapters match your table of contents, and the price is the one you set. Record which store received the book through which route, and the date.

Send each store the book through one route only. Author's Republic's FAQ says delivering the same audiobook to a retailer through several distributors violates its terms and can bring complaints or takedowns.

What it costs, and what we cannot tell you about time

Cost depends on the tool and on how many retakes you need. The AI audiobook cost calculator works out credits for your word count. We have no measured figure for how long correction takes, and we do not quote one. Treat any single-afternoon claim as unmeasured.

The fixed waits are easier to plan. As an estimate for a 90,000-word book on INaudio's route, one full listen takes at least 9.7 hours, plus up to 60 days for distribution, so set any launch date at least nine weeks after you expect to finish proofing. That sum allows no time for corrections or a second listen.

Who should not follow this route

Skip it if Audible is where your listeners are and you have no Amazon invitation. Skip Spotify's own tools if you need more than one voice, outside the US or English. Skip INaudio's route if you plan to master the audio in your own software. Skip AI narration altogether if you cannot afford a full listen, because every route on this page leaves the checking to you.

For the full set of AI narration guides, return to the AI audiobooks hub.

Last verified 2026-11-08. Rules and prices change: check the linked primary page before you rely on this.