How to Write a Book with AI (Without It Writing For You)

There are now over 200,000 AI-generated books on Amazon. Most of them are terrible. They read like longer versions of ChatGPT output: fluent, generic, and empty of anything a specific human being thought or experienced. They sell a few copies, collect a few one-star reviews, and disappear.

This is not the model for writing a book with AI. If you want to write something worth reading, AI should never hold the pen. But it can be an excellent editor, researcher, and thinking partner. The difference between those two approaches is the difference between a book that sounds like everything else and a book that sounds like you.

Why AI-Generated Books Are Bad

The problems with AI-generated books are not technical. The grammar is fine. The structure is acceptable. The problem is that the writing has no point of view. It has no lived experience. It has no surprise.

When a human writes about starting a business, they remember the specific phone call where a client said no. They remember eating ramen for three months. They remember the afternoon they almost quit. These details make the writing real. AI does not have these details. It has patterns from thousands of books about starting a business, and it produces the statistical average of all of them.

Sam Altman, the CEO of OpenAI, said in an interview with Tyler Cowen that even GPT-7 might only produce "a real poet's okay poem." Not great. Not good. Okay. And that is for a short poem. A book is 50,000 to 80,000 words. The flaws compound over that length. Repetitive vocabulary. Compulsive hedging. The same transitional phrases appearing every few pages. Readers notice. Readers always notice.

Patrick Collison and Tyler Cowen ran a grant program called "New Aesthetics" looking for great AI-created art. Their conclusion: they had not seen much great work that only uses AI. The ceiling on pure AI generation is structural. More parameters will not fix it because the problem is not capability. The problem is that good writing requires a specific person with specific experiences making specific choices. AI makes average choices by design. That is what prediction means.

What the Brain Research Says

The most important study on this topic comes from a joint research effort by MIT, Wellesley, and Massachusetts College of Art. Researchers put writers in brain scanners and measured what happened when they used AI in different ways. The results should change how every writer thinks about AI.

Writers who used AI freely from the start, letting it generate text and then editing the output, showed lower brain activity than writers who wrote without AI. Their brains were coasting. The AI was doing the cognitive work, and the human was passively reviewing it.

But writers who drafted first and then used AI to revise showed the opposite pattern. Their brain activity increased. It was higher than the no-AI group. Evaluating AI suggestions, deciding which to accept and which to reject, engaged the brain more than writing alone.

This is the key finding for anyone writing a book. The order matters. Draft first, then use AI to edit, and your brain works harder. You make better decisions. You produce stronger prose. Let AI draft for you, and your brain disengages. The quality drops. Your voice disappears.

Think about what writing a book really is. It is not typing. It is making thousands of small decisions: this word, not that one. This example, not that one. This structure, not that one. These decisions are the craft. They are also the thinking. When AI makes them for you, both the craft and the thinking atrophy.

The Right Workflow: Draft, Then Edit

Anne Lamott wrote about "shitty first drafts" in her book Bird by Bird. Her argument: all good writers write terrible first drafts. The magic is in the revision. The first draft is just you getting your ideas on the page in whatever ugly, disorganized form they come out.

This is exactly the right model for writing a book with AI. You write the first draft. AI helps you revise it. Here is what that looks like in practice.

Step 1: Write your messy first draft

Write each chapter without worrying about quality. Do not stop to fix sentences. Do not rearrange paragraphs. Do not worry about transitions. Just get your ideas, stories, and arguments onto the page. Write fast. Write badly. Write honestly.

This is the part AI cannot do for you. Your first draft contains your voice, your experiences, your point of view. It is messy and unpolished, but it is real. A polished chapter generated by AI has none of those things.

For most people, this means writing 1,000 to 2,000 words per chapter in a single sitting. Do not edit as you go. Editing and writing use different parts of the brain. Mixing them slows you down and produces worse results.

Step 2: Edit chapter by chapter with AI

Once you have a draft of a chapter, bring in AI. But not as a chatbot. Not in a separate window where you copy and paste paragraphs back and forth. Use an editor that shows you exactly what the AI wants to change.

This is where most writers go wrong. They paste their chapter into ChatGPT and say "make this better." ChatGPT rewrites the whole thing. The writer gets back a wall of text and has no idea what changed. Did the AI improve the argument? Did it cut an important detail? Did it replace a strong, specific verb with a weak, generic one? There is no way to tell without reading both versions line by line.

The better approach is using AI as an editor that shows inline diffs. You see the specific words and sentences the AI wants to change, highlighted in your document. You accept the changes that make your writing stronger. You reject the changes that flatten your voice or remove important details. This is the workflow the MIT research supports. You stay in control. Your brain stays engaged.

Step 3: Accept, reject, repeat

For each chapter, you might run three or four editing passes with different instructions. First pass: tighten the prose. Cut unnecessary words. Fix awkward sentences. Second pass: check the argument. Are there logical gaps? Missing evidence? Third pass: improve transitions between sections. Fourth pass: read for voice. Does it still sound like you?

Each pass, you accept some suggestions and reject others. Over time, you develop a sense for when AI suggestions improve your writing and when they flatten it. That judgment is itself a skill. The brain research says practicing it makes you a better writer.

Tools for Each Phase

Different phases of book writing benefit from different tools. Here is what works for each stage.

Brainstorming and outlining

ChatGPT and Claude are good thinking partners when you are working out the structure of your book. Describe your topic and ask for feedback on your outline. Ask what angles you might be missing. Ask what questions a reader would have after chapter three. This is conversation, not generation. You are using the AI to pressure-test your thinking, not to think for you.

At this stage, the chat interface works fine. You are not producing final text. You are exploring ideas.

Organization

Scrivener remains the best tool for organizing a book-length project. It lets you break your manuscript into chapters and scenes, move them around, and maintain notes alongside your text. Notion works too, especially if you want to link research notes to specific chapters. The key is having a tool that can handle the structural complexity of a book without forcing everything into a single linear document.

Drafting

Use whatever tool lets you write fastest. Google Docs, Word, a plain text editor. The tool does not matter at this stage. Speed and honesty matter. Some writers use voice dictation for first drafts because talking is faster than typing and produces more natural rhythms. Whatever gets words on the page with the least friction.

Editing

This is where the tool matters most. Athens is built for this workflow. You select a passage, describe the edit you want, and the AI shows you inline diffs. Green for additions, red for deletions. You accept or reject each change individually. Your document never gets rewritten out from under you.

For a book, this means working chapter by chapter. Load a chapter, run editing passes, and accept the changes that strengthen the writing. The key advantage over copy-paste workflows is that you always see exactly what is changing and why. No surprises. No losing your voice to a wall of AI-generated text.

Research

Perplexity is excellent for fact-checking and finding sources. It returns citations with its answers, so you can verify claims before including them in your book. NotebookLM is useful for processing large volumes of source material. Upload your research documents and ask questions across all of them.

A warning: always verify AI-provided facts and citations. AI can hallucinate sources that do not exist. For a book, every factual claim should be traceable to a real source you have read yourself.

A Chapter-by-Chapter Process

Here is the complete process for one chapter. It takes roughly four to six hours per chapter, depending on length and complexity.

Write the first draft. One to two hours. Get everything on the page. Do not edit.
Let it sit. A few hours or overnight. Distance improves your editing judgment.
First AI editing pass. Focus on clarity and concision. Accept changes that tighten the prose. Reject changes that remove your voice.
Second pass. Focus on argument and structure. Are the ideas in the right order? Are there gaps?
Third pass. Focus on voice. Read the chapter out loud. Does it sound like you? If the AI has smoothed out too many rough edges, revert some changes and keep the roughness. Roughness is often honesty.
Final read. One complete read-through without AI. Trust your own judgment for the last pass.

Multiply this by twelve to twenty chapters, and you have a book. The process is not fast. But the result is a book that contains your thinking, your experience, and your voice, polished to a level that would take much longer without AI assistance.

What Not to Do

Do not ask AI to generate chapters. You will spend more time editing out the generic phrasing, removing the hedging, and adding real details than you would have spent writing the chapter yourself. Generation feels fast but produces more work downstream.

Do not use AI for the parts of your book that require lived experience. Your personal stories, your original arguments, your specific observations about the world. These are what make your book worth reading. AI cannot supply them.

Do not let AI smooth away all the rough edges. Some of your best writing will be sentences that AI flags as awkward or unclear. Sometimes awkwardness is the point. Sometimes an unusual sentence structure is exactly what the passage needs. Develop the judgment to know when AI is right and when it is wrong.

The Bottom Line

AI can cut the time it takes to write a book by 30 to 50 percent. But only if you use it correctly. The correct use is: you write, AI edits. Not the other way around.

The writers who will produce the best books in the next decade are not the ones who generate the fastest. They are the ones who think the clearest, write the most honestly, and then use AI to polish what they have written. The book is still theirs. AI just helps them make it better.

Write your shitty first draft. Then let AI help you make it great. That is how you write a book with AI without it writing for you.