Write a Book from Your Own Research: Turning Notes, PDFs, and Interviews into Chapters

Most people who want to write a non-fiction book are not starting from nothing. They have a folder of notes. A few years of newsletter archives. Client reports, workshop slides, interview transcripts, a half-finished draft from two summers ago. The problem was never a shortage of material — it's that the material is scattered across forty files in three formats and has no shape.

This is exactly the situation where generic AI writing goes wrong. Ask a model to "write a chapter about pricing strategy" and you get the internet's average opinion on pricing strategy: competent, forgettable, and indistinguishable from every other book on the shelf. The version worth publishing is the one that draws on your case studies, your numbers, and the specific things you learned the hard way. That only happens if the material actually goes in.

Why "from your material" changes the output

There's a meaningful difference between two prompts that look superficially similar:

1. "Write a chapter about onboarding new freelance clients."
2. "Write a chapter about onboarding new freelance clients, using the attached intake questionnaire, the three project post-mortems, and the transcript of my workshop on scope creep."

The first produces plausible general advice. The second produces a chapter with your intake questions in it, your post-mortem findings as examples, and your workshop's framing of scope creep — the things a reader can't get anywhere else. That specificity is the entire reason someone buys a book from a practitioner instead of reading a listicle.

Practically, this also solves the hallucination problem for a large class of books. When the source of truth is a document you wrote, the model isn't recalling facts from training — it's working from text you can point at and check.

Step 1: Gather before you outline

The instinct is to outline first and hunt for supporting material later. Reverse it. Spending thirty minutes collecting what you already have, before the outline exists, changes the outline itself — you'll discover you have far more on some topics than you assumed, and nothing at all on others.

Do a quick inventory pass and pull together:

• Notes and drafts — anything you've already written on the topic, even in fragments
• Primary material — interview transcripts, survey results, internal reports, case notes
• Reference documents — PDFs, whitepapers, or standards you'll be citing or summarizing
• Prior published work — blog posts, newsletters, talk scripts, course material

You are not trying to be exhaustive. You are trying to get the 80% that carries the book's actual substance.

Step 2: Get it into a usable format

Source material only helps if the tool can read it. In Ebook Creator, you attach documents to a book directly and the text is extracted for you — PDF, DOCX, EPUB, TXT, and Markdown, up to 25 MB per file, with a total budget of 600,000 characters across all documents for a single book.

That total budget is worth understanding rather than ignoring, because it shapes how you should select material. 600,000 characters is roughly 100,000 words — which sounds enormous until you upload three annual reports. The material is passed to the writing model as context, so the constraint is real, and the practical implication is simple: curate, don't dump.

If you're over the budget, or close to it:

• Cut boilerplate — legal appendices, repeated headers, tables of contents, disclaimers
• Prefer the primary document over three near-duplicate versions of it
• Trim transcripts to the sections that actually contain insight, not the small talk
• Drop material that supports a chapter you already decided to cut

A tightly curated 150,000 characters usually produces better chapters than a sprawling 600,000, for the same reason a well-briefed ghostwriter beats one buried in documents: signal density matters more than volume.

Step 3: Let the outline come from the material

Once your documents are attached, the outline step has something to work with. Instead of generating a generic chapter structure for your topic, it can propose a structure that reflects what your material actually covers — and this is where you'll immediately see the gaps.

Read the proposed outline against your inventory and ask, chapter by chapter: do I have real material behind this, or is this a chapter the model invented because books on this topic usually have one? Chapters in the second category are the ones that come out thin and generic. You have three options for each: cut it, merge it into a neighbour, or go find the material before you write it.

This is also why the outline is fully editable before you spend anything on writing — you can reorder, rewrite, delete, add, and set a target word count per chapter, and none of it costs a writing credit. See our guide on chapter length for benchmarks on what those targets should be.

Step 4: Write, then check against the source

When chapters are written, your source material goes to the writer alongside the outline and the running continuity context — the style guide, the accumulated entity notes, the summaries of previous chapters. The result is chapters that reference your examples and use your terminology rather than drifting into generic phrasing by chapter nine.

Your editing job changes shape as a result. Instead of "is this true?", the question becomes "is this my version of true?" — did it represent the case study accurately, did it flatten a nuance you care about, did it use the right term for something your field is picky about. That's a much faster read than fact-checking invented content.

For anything that's close but not right, targeted fixes are cheaper than regenerating a whole chapter: rewriting a selected passage costs nothing, and you can edit the Markdown directly for small corrections.

What this approach is not

Two honest limits worth stating.

It isn't a citation engine. Your documents inform the writing; they don't produce a footnoted bibliography with page references. If your book needs formal academic citation, you're doing that layer yourself.

It isn't a search index over an unlimited library. The material is provided to the model as context, within the character budget above — it isn't a system that retrieves from a bottomless archive on demand. For most practitioner non-fiction that's a non-issue; for a book that genuinely depends on tens of millions of words of corpus, it's the wrong tool.

Being clear about this is the point. A book written from a curated 100,000 words of your material is a real book. A book that claims to have digested your entire company drive is a claim nobody should believe.

A realistic workflow

Put together, a first pass looks roughly like this:

1. Half an hour: inventory and collect your material into one folder
2. Ten minutes: cut it down to the documents that actually carry substance
3. Ten minutes: describe the book — reader, promise, tone — and attach the documents
4. Twenty minutes: read the generated outline critically, cut invented chapters, set chapter lengths
5. Then: approve and let the chapters write, and spend your real time on the edit pass

The part that used to take months — turning a pile of unstructured material into an organized, evenly-paced manuscript — is the part that compresses hardest. The judgment about what's worth saying stays with you, which is exactly where it belongs.

If you've got the material and never found the time to shape it, start a book with your own documents — every new account gets 4 free credits, enough to outline and write a complete short book before paying anything. If you're at an earlier stage, how to write an ebook with AI covers the full workflow from scratch.