Somewhere between “I wrote a book with AI” and “I have a book” sits a step almost nobody explains: getting the words out as a manuscript rather than a transcript. The confusion is understandable, because “export” means two completely different things depending on where your book currently lives — and only one of them gives you something you can publish.
A ChatGPT data export is a transcript; a Canvas export is a document; a book is a manuscript
ChatGPT now has more than one export path. A privacy or account data export gives you conversation history—the prompts, replies, rejected directions, and corrections that record how the work happened. Canvas can export an individual document to PDF, Markdown, or Word, and its version history can restore earlier document states. Those are useful capabilities and should be credited accurately.
A manuscript still answers a different question: which chapter versions are accepted, in what reading order, with which front matter, and with no abandoned alternatives mixed in? One canvas can be a clean chapter or even a full document, but several canvases and chats do not automatically become one authoritative book. Manuscript export begins after acceptance and ordering have been resolved, not after the first downloadable file appears.
The question is not only “can this screen download a document?” It is “which accepted chapter versions, in which order, make up the book?”
Four formats, four jobs — match the file to what happens next
Once the accepted text exists in one place, the format question has a clean answer, because each common format is good at exactly one job. Choosing by what you are about to do — not by which format sounds most “final” — saves you from the classic mistake of sending a print PDF to an editor who needed to mark it up, or uploading a fixed-layout PDF as an ebook that then reads badly on every phone.
| Format | Reach for it when | What it is |
|---|
| EPUB | Publishing an ebook | Reflowable — adapts to any screen and font size |
| PDF | Making a print proof | Fixed-layout — pages exactly as designed |
| DOCX | Sending to an editor or beta reader | Editable — tracked changes and comments |
| Markdown | Keeping a portable source | Plain text with light structure — opens anywhere |
Retail-platform format acceptance follows each platform’s current documentation (see Sources). Verify before you upload.
The clean-manuscript checklist (what a good export file has)
A manuscript export should lint clean before it goes anywhere. These are the checks that separate a file an editor or a retailer accepts without complaint from one that bounces back. None of them is about fancy design — they are about the file containing the book, in order, and nothing else.
- One document, in reading order — no interleaved prompts, no model chatter, no “here is your chapter” framing.
- Real chapter breaks the format understands — headings or page breaks, not blank lines you eyeballed.
- Front matter separated from the story — title, copyright, and dedication are not chapter one.
- Consistent straight or curly quotes, not a mix the conversion mangles.
- No stray markdown symbols or asterisks left in the prose.
- For print (PDF): the trim size, margins, and final page count are what you intend — because page count sets spine width and cost.
- For ebook (EPUB): no hand-forced page breaks mid-chapter — reflow decides pages, not you.
Manual export and a publication snapshot are not the same thing
It is worth being precise about two BookWriter features that sound similar and are not. A manual export is a working file: generate an EPUB, PDF, DOCX, or Markdown any time, on any plan, from the current state of the manuscript. Use it constantly — to read on your phone, to send to a friend, to check a print proof.
A publication snapshot is a release artifact: a unique, immutable EPUB or PDF captured as a fixed record, available on Pro with a limited number per billing period. The difference is intent. A manual export changes whenever your book does; a snapshot is frozen on purpose, so the version you published stays exactly the version you published. Keep them straight and you will never wonder which file you actually shipped.
Run a three-file exit drill before you trust any writing system
Export three files from the same accepted manuscript state: a DOCX for editorial exchange, an EPUB for ebook reading, and a PDF for print inspection. Open the DOCX in a tool that supports comments or tracked changes. Open the EPUB in at least two reading systems or a validator. Inspect the PDF at the intended trim size. A successful download is not proof that the file is usable downstream.
Compare the files against a one-page control sheet: title and author metadata, chapter count and order, first and last sentence of every chapter, front-matter order, image count, footnote or endnote count, and total accepted word count. The presentation will differ by format, but those manuscript facts must agree. If they do not, stop before upload and identify whether the source state or one renderer is wrong.
Keep the control sheet, the source-state identifier, the export timestamp, and file checksums together. Then make a harmless manuscript change and export again. The new files should differ; the old release files should not. This small drill proves portability, repeatability, and rollback—the three properties that matter when the writing platform is no longer open in front of you.
- DOCX: comments and tracked-edit handoff works without flattened text.
- EPUB: table of contents, chapter order, reflow, links, and images survive.
- PDF: trim, margins, fonts, page count, and blank pages match the print plan.
- Control: metadata, accepted text, order, counts, timestamp, and checksum are recorded.
Build a release manifest so every file points to the same book
A folder containing “final.docx,” “final-final.epub,” and “print-v3.pdf” is not a release record. Create a manifest for every serious export. Record the book title, author name, edition, language, source manuscript identifier, accepted chapter count, accepted word count, export timestamp, application version where available, filenames, formats, and checksums. Add the first and last sentence of each chapter or a content hash so chapter boundaries can be compared without rereading the whole book.
Use one release identifier across all formats generated from the same accepted state. The EPUB, editorial DOCX, and print PDF may render differently, but they should agree on the manuscript facts in the manifest. If the PDF contains a late typo correction that the EPUB does not, they are no longer one release. Update the accepted source, regenerate every affected format, and issue a new identifier rather than patching one downstream file invisibly.
Keep the manifest with the exported files in storage independent of the writing product. Verify each checksum after copying or downloading. The checksum does not prove the prose is correct; it proves the file you inspected is still the same file. That distinction matters when a retailer, editor, or printer asks which artifact was supplied weeks later.
- Identity: title, author, edition, language, and release identifier.
- Source: accepted-state identifier, chapter count, word count, and export time.
- Artifacts: filename, format, size, and checksum for every delivered file.
- Content controls: chapter order plus first and last line or chapter hash.
- Verification: application used to open each file and the result of the check.
Preflight the content separately from the format
Start with a format-neutral manuscript check. Confirm title and author metadata, front-matter order, chapter names and sequence, accepted text, intentional blank sections, image inventory, notes, links, and the final sentence. Then run a format-specific check. Separating the layers prevents a rendering problem from hiding a source-state problem—or a correct-looking PDF from disguising a missing chapter.
For DOCX, verify headings, section and page breaks, comments, tracked-change behavior, footnotes, and image anchoring in the editor your collaborator will use. For EPUB, inspect navigation, reflow at several font sizes, paragraph indents, scene breaks, images, links, and accessibility metadata in more than one reader or a validator. For print PDF, check the intended trim size, embedded fonts, margins, page sequence, blank-page placement, image resolution, and final page count because that count affects the cover spine.
Do not treat retailer acceptance as the entire quality test. A file can upload successfully and still have broken navigation, orphaned headings, a nearly blank page, or an image that is unreadable on a phone. Use the retailer’s current Previewer after upload, then compare its rendered beginning, chapter transitions, images, and ending with the independently verified export. Preserve the exact submitted file in the release bundle.
| Layer | Questions it must answer |
|---|
| Accepted source | Is this the correct text, order, metadata, and front matter? |
| DOCX | Can an editor revise it without broken styles or flattened structure? |
| EPUB | Does navigation and reflow work across reader settings? |
| Print PDF | Do trim, margins, fonts, images, pagination, and blank pages print correctly? |
| Retail preview | Did platform conversion preserve the independently verified file? |