Table of Contents
TL;DR: The hardest part of writing a book is not the writing. It is getting the raw material out of your head and onto a page in a form a writer can work with my ghostwriting process. AI handles this surprisingly well when used the right way, by helping you talk a book into existence the AI-assisted book and then organizing the talk into usable material. Here is the actual process, the specific tools and prompts that work, and the rules that keep you from contaminating the raw material with the average prose AI produces by default.
The bottleneck nobody names
Authors who hire a ghostwriter or who try to write a book themselves both hit the same wall, and it is not the writing. The wall is that the book exists in your head as scattered memories, half-formed ideas, lessons learned the hard way, opinions you have not put into clear words, and stories you have told dozens of times but never written down. Until that material gets out of your head and onto a page, nobody can do anything with it. For more, see using AI for research without getting burned. Not your writer, not you, not the editor, not the publisher. For more, see isn't using a ghostwriter cheating?. The material has to exist outside your head before the book can be built from it.
Most authors underestimate this stage and assume the writing is the work. The writing is twenty percent of the work. Sixty percent is getting the raw material out of your head. The remaining twenty is editorial shaping. The bottleneck on most book projects is the first stage, and it is the stage AI handles better than any other tool ever available to authors, when used the right way.
Talk your book into existence
The right way is to talk. Most people are far better at talking through their thoughts than at writing them down. A speaker can hold a complex idea in their head while a sentence runs, can interrupt themselves with the better version of the point, can pause and circle back, can use the tones and gestures that carry meaning before sentences do. On paper, all of that gets stripped out and you are forced to produce final-form sentences, which is why writers stall. Speech sidesteps the stall because you do not have to produce final-form anything when you talk. You just have to say what you actually think.
Talk into a voice recorder, your phone, or any of the dictation tools that exist. Talk for thirty minutes about one piece of the book. Then transcribe what you said using one of the cheap or free transcription services available, or paste the recording into an AI tool that handles transcription. What comes out is a messy, rambling, redundant version of what you actually know about the topic. That mess is gold, because it contains the specific material in your specific voice that no other process can produce.
Where AI earns its keep here
The mess that comes out of a transcription is unusable as it stands. Sentences are incomplete. Topics overlap. Ideas double back on themselves. Tangents go nowhere. This is exactly the kind of mechanical organizational work AI handles well. Drop the raw transcript into the model with a clear prompt: organize this into topics, preserve the speaker’s specific words and phrases, do not paraphrase, flag anything that seems unclear or contradictory. The machine returns a structured version of the mess that is now usable as source material for the actual writing.
The critical rule is “do not paraphrase.” This is where most authors using AI for this purpose go wrong. They let the machine clean up the transcript, and the cleanup smooths out the specific phrasings that contain the voice. The machine turns “I always tell people this is the trap” into “It is important to note that this represents a common pitfall,” and the voice is gone before the writing even starts. The instruction has to be explicit: preserve the speaker’s words, do not improve the prose, flag confusion but do not solve it. Run that prompt and the output is a structured raw material file that contains your actual voice waiting to be shaped into chapters. I have written elsewhere about why AI prose is structurally shallow, and the shallowness comes from exactly the paraphrasing instinct the machine defaults to. The first move is to turn it off.
Prompt yourself, then prompt the machine
You also need to know what to talk about. The fix is to prompt yourself. Sit down with a list of questions, generated by you or by AI from your topic, and answer them out loud one by one. Talk about the worst project you ever ran, the lesson that changed your career, the client who taught you something, the moment you knew you were wrong about something important. Each prompt produces five to thirty minutes of voice material, and the prompts unlock the specific stories and arguments the book needs.
AI can generate the question list well, once you have given it the topic and the audience. A useful prompt is something like “generate twenty interview questions for a memoir about [topic], focused on specific moments, decisions, and turning points rather than general reflections.” The list comes back as a working interview script. You answer them out loud over the course of a week, an hour a day. By the end, you have ten hours of transcribed voice material covering the territory of the book in your own words. The writer or you, depending on the engagement, now has source material that is genuinely usable. The Book Discovery Intensive runs a version of this process formally, where the interviews and material extraction happen in a structured engagement before any larger writing commitment.
What to do with the material once it exists
Once the raw material is in a structured file in your own voice, several paths open. A ghostwriter on a full or AI-assisted engagement now has source material that lets them write quickly because the substance is already there in your words. Book coaches can work with you to shape the material into a manuscript while you keep doing the writing. DIY authors have the raw foundation to build chapters from directly, and the chapters will sound like the author because they are built on material that already sounds like the author.
The savings on time and money across every path are substantial, because the bottleneck has been cleared. The author who has talked their book into existence and organized the transcripts can move through the writing process at perhaps three times the speed of the author who is still trying to figure out what the book is about. That speedup is real, and it is the single biggest practical benefit of AI on a book project for most authors.
The rules that keep this from going wrong
Three rules keep this process honest. First, preserve your actual words at every step. The transcription captures what you said. Organization preserves the phrasings. Each writing step builds on the phrasings rather than replacing them. Any step that smooths your voice out of the material has just damaged the source. Second, verify the AI’s organization against your recordings. Did the machine drop something? Did it group two unrelated things together? Could it have introduced a claim you did not make? The verification is fast and catches the errors before they propagate. Third, never let the machine generate raw material from nothing. The output of “write about [topic]” is the average. The output of “organize what I said about [topic]” is your material. Those are completely different things, and only the second is useful.
Get those three rules right and AI becomes the most useful tool ever invented for the front end of a book project. Get them wrong and you spend a month producing a contaminated source file that smells like AI and never makes the book any better. The discipline is small, the upside is large, and the difference is in whether you treat the machine as a writer or as an assistant for organizing your own work.
Frequently Asked Questions
Related: the AI-assisted book