As the founder of a nonfiction publishing company, a longtime ghostwriter, and an author myself, I have been just as saturated as everyone else in the cultural conversations about AI. Because of my place in the publishing industry, I recently decided I needed to do a deep dive into what was actually happening with AI in the writing sphere so that I could give authors the information, thought processes, and mindset shifts they need to navigate it.
After doing that work, my professional recommendation is straightforward: do not use generative AI to write your book. I do not recommend using it to develop your book idea, build your structure, generate your prose, develop your voice, or make the creative choices that turn an idea into a manuscript.
I arrived at that position over time, and I do not share it lightly. I’m generally much more interested in understanding why people use a tool and what actually happens when they do than in drawing a quick line in the sand. But after reading research, listening to experts, looking closely at how large language models work, and—most importantly—watching authors use them, I have become increasingly convinced that generative AI short-circuits some of the most important work a writer needs to do.
The reason comes down to something I have been saying about writing for years: writing is a series of choices.
That idea is at the center of how I think about using AI to write a book, because every time an LLM writes or structures something for you, it is making some of those choices in your place.
Why I’m Skeptical of AI for Writers
Before I get into writing specifically, I want to be transparent that I already have plenty of concerns about AI more broadly. I worry about the environmental costs. I worry about the enormous amounts of money being poured into an industry whose economics have not always been particularly transparent. And I worry that potentially powerful technology is developing without anything resembling the kind of responsible oversight I would like to see.
I am also skeptical of some of the claims being made about what these tools can actually do.
A great deal of the cultural messaging around AI tells us that these tools are going to transform essentially everything. At the same time, I work with business owners and authors who are actually using the tools that exist right now, and what I see very often is that the results do not live up to what people have been promised.
That does not mean the technology is unimpressive. Of course it is impressive. But “impressive” and “good for your book” are two very different questions.
One of the reasons I wanted to spend more time looking at this issue is that conversations about AI have become incredibly black and white. A LinkedIn post I wrote about AI recently got far more attention than my posts normally do, and I was surprised because I had not thought of it as especially anti-AI or pro-AI. I was trying to ask what I think is a much more useful question: when people use these tools, what is actually happening to the work?
That is the question I want writers to ask too.
Writing Is a Series of Choices
At its core, writing is simply a series of choices.
Think about the very beginning of a book. First, you choose whether to write it at all. Then you choose what the book is about.
For memoir writers, I talk often about the importance of choosing a specific memoir topic and distinguishing that topic from your memoir’s message. Your topic could be adopting your third child, starting your family more broadly, your extended family’s trajectory, or only the period during which you decided whether to adopt.
Each of those choices creates a different book.
And once you make one choice, a whole new tree of choices opens in front of you. You choose the message. As the author, you decide which experiences belong and which do not. You decide what you want the reader to understand.
This is what we are doing when we develop a book idea. We are narrowing an almost infinite range of possibilities until we begin to understand the particular book we want to write.
Then we move into structure, where we make another enormous set of decisions.
Structure is a little different because the choices are more constrained by genre and publishing conventions. Certain books tend to use particular kinds of structures, and someone working outside the publishing industry may see far more possible options than an experienced publishing professional would.
That can be a blessing and a curse. Sometimes people who come from outside an industry make breakthroughs because they are willing to violate its expectations. But knowing what those expectations are is still crucial, arguably even more so if you plan to violate them.
That is why I talk so much about learning how to outline a nonfiction book and, for memoir writers specifically, how to build a memoir outline. The outline is not busywork. The process of creating it forces you to make the decisions that create the logic of the book.
And we still have not started writing sentences.
Once you begin drafting, the number of choices becomes almost impossible to quantify. You choose what belongs in each chapter, what example comes first, how you explain an idea, what sentence comes next, what word comes next, and even what form of that word you use.
That is the writing experience. We choose and choose and choose, and every choice opens new possibilities while eliminating others.
The fact that you are making those choices is what makes the finished book uniquely yours.
What Happens When an LLM Makes Those Choices Instead?
When people talk about AI and writing right now, they are usually talking about large language models, or LLMs.
There are many different forms of artificial intelligence, some of which we have been using for years. But when a writer says, “I used AI to write this,” they are typically talking about generative AI powered by an LLM.
At a simplified level, an LLM works by predicting what token—essentially a piece of a word—is most likely to come next based on what came before it. Then it predicts the next one and the next one and the next one.
In that sense, there is a superficial similarity between what an LLM does and what a writer does. I am choosing the next word while I write this sentence too.
The difference is what those choices are based on.
The LLM is operating through mathematical probability. Human beings bring an enormous number of cultural, social, professional, ethical, personal, and emotional constraints into every choice we make. From the time we are born, we absorb norms, experiences, relationships, beliefs, values, preferences, and expectations. Most of those influences shape our decisions without us consciously stopping to identify them.
Those are some of the things that make one human being choose a sentence another human being would never choose.
An LLM does not have your history. It does not have your relationships, your values, or your intentions for your life. It has no stake in what happens when the book enters the world.
It’s just generating likely language.
And that is why I do not recommend using AI to write a book. If writing is the act of making all of these choices, outsourcing the choices means outsourcing a substantial part of the writing itself.
There are two places where I see the consequences most clearly: voice and structure.
Do Not Outsource Your Writing Voice to AI
Voice is probably the most seductive place for authors to hand over control because so many first-time authors already feel insecure about it.
I hear versions of the same concern all the time: I know what I want to say, but I’m not sure I sound literary enough. (Or insert your favorite: professional enough, polished enough, interesting enough, etc.)
Then an AI tool comes along and effectively says, “Give me the idea. I’ll choose all the words for you.”
If you already doubt your ability to make those choices, that offer sounds wonderful.
The problem is that the way you get better at making choices is by making them.
If you consistently outsource word choice to an LLM, you have short-circuited the process that would have helped you become a stronger writer. I am much faster at writing today than I was earlier in my career because I make these decisions every day. Experienced writers become faster because they have practiced.
The same thing happens when you work on sentence-level editing. You start hearing your own patterns, which helps you notice when the rhythm is off. That helps you develop judgment because you are actually doing the work.
This is also why reading strong books matters. When you think seriously about what makes a memoir good, for example, you are studying the choices another writer made and developing a better sense of the choices available to you.
When AI consistently makes those choices instead, you lose that practice.
There is another problem too. Because an LLM is generating probable language, its choices tend toward language that is already common. That is part of why AI-generated prose can feel so clean and easy to read. Your eyes can move across it with almost no friction.
But friction is often where voice lives.
A strange turn of phrase, an unexpected rhythm, a sentence that makes you stop for a second—those things are not necessarily writing flaws. Often they are exactly what makes a writer recognizable.
I have written before about how AI can flatten your writing, and this is a major reason why. The language becomes increasingly probable and increasingly similar to the language the same system is producing for everyone else.
If your concern is that your voice is not good enough, my professional advice is not to have AI replace it. Your voice is the exact skill you need to practice.
Do Not Let AI Build Your Book Structure Either
The other place I strongly recommend keeping generative AI out of your process is structure.
I have not yet worked with an author who told me that AI built their book structure and then found that structure genuinely useful as they moved through the manuscript.
What I see much more often is an author arriving with an outline that looks reasonable on the page but does not actually support them when they begin writing.
They do not know why chapter four is chapter four or why a particular idea appears in one section instead of another. The chapter titles and bullet points are there, but the author–the author, y’ll–cannot explain the underlying logic of the book because they were not the person who worked through that logic.
That is exactly why outlining matters.
When you build your own outline, you are not simply producing a document that tells you what the chapters are called. You are deciding what the reader needs to know first, which ideas belong together, how the argument should build, how tension progresses, and what each chapter is supposed to accomplish.
Doing that work teaches you how your own book works.
If you ask AI to create the outline, you get the artifact without necessarily getting the understanding that makes the artifact useful.
Then you sit down to write chapter four and realize you do not know why chapter four is there.
And what is the easiest solution at that point? Ask AI to write chapter four.
This is how I have seen authors become progressively more dependent on generative AI. The tool makes one decision that the author does not fully understand, which creates a problem that can then be solved by asking the tool to make the next decision.
Before long, the writer is no longer really writing the book at all.
I know that is not what most people intend when they first experiment with using AI to write a book. They usually think they are saving time or getting help with a piece of the process that feels difficult.
But difficulty is not always a sign that something has gone wrong.
Sometimes difficulty is the work.
And the good news is that authors become better at this surprisingly quickly. I see it all the time in coaching. Someone who feels completely overwhelmed by structure or drafting at the beginning can become dramatically more confident within a couple of months because they have practiced making those decisions and received useful feedback.
There really is no way out but through.
What About Grammarly, Spellcheck, and Other AI Tools?
This is where people understandably start asking questions about what “counts.”
Someone recently asked me whether Grammarly was okay, and the anxiety in that question really stayed with me.
We have been using technologies that automate parts of writing for many years. Spellcheck is the obvious example. Most of us do not think we have compromised our authorship because Microsoft Word told us we misspelled something.
I do not think you need to become afraid of every tool that contains some form of AI.
The useful distinction for me is whether the tool is making your creative choices or helping you identify something that you are still capable of evaluating yourself.
I use tools to check spelling and grammar, particularly at the very end of a document when I want another pass for typos. I can look at a spelling suggestion and decide whether it is correct. The tool is not determining my argument, creating my structure, deciding what story belongs in the book, or choosing how I want to express an idea.
I am still exercising discernment.
That is very different from typing a prompt into a generative AI system and asking it to write the paragraph, rewrite your prose, create the chapter, generate the outline, or tell you what your book should be about.
So my recommendation is not “never touch technology.”
It is much more specific: do not use generative AI to perform the creative work of writing your book.
Pay Attention to the Places Where You Feel Insecure
One of the most important things I have noticed while working with authors is that AI becomes especially tempting at the point where someone already feels insecure.
If you are insecure about your spelling, a spellchecker can help you identify errors while still leaving you in control of the writing.
But if you are insecure about your voice, an LLM offering to rewrite your prose can very quickly take over a part of the process you already do not trust yourself to manage.
The same is true of structure. If you do not believe you know how to organize a book, an AI-generated outline can feel like relief.
If you do not believe your idea is good enough, asking AI to improve or expand it can feel reassuring.
And if the real insecurity underneath all of those things is that you are not sure whether you are good enough to write the book, a tool that offers to make the decisions in your place can become incredibly difficult to resist.
But those are the exact places where I want you to build more discernment, not less.
If you are unsure of your idea, spend time developing the book idea until you understand what makes it worth writing.
Same with the question of whether you have enough material. Work through whether you actually have enough for a book rather than asking a model to generate material to fill the gaps.
If structure feels daunting, learn the principles of structure and practice applying them. For memoir writers, that might mean learning the basic steps for outlining a memoir or looking at different ways to structure a memoir.
If you are unsure about the manuscript as a whole, get feedback from a person who can tell you what is working and what needs revision. That is exactly what our Manuscript Review is designed to do: give you the 10,000-foot assessment and a step-by-step revision plan without taking over the creative work of writing the book for you.
The insecurity itself is not a reason to outsource the decision. It is a sign that this is a skill you need an opportunity to strengthen.
My Professional Recommendation on Using AI to Write a Book
So let me make my position as clear as I can.
I do not recommend using generative AI for creative writing.
I do not recommend asking it to write your prose.
I do not recommend asking it to rewrite your prose so that it sounds “better.”
I do not recommend using it to generate your structure.
I do not recommend using it to decide what belongs in your book, what your argument should be, or what stories you should tell.
I know that sounds like a hard line, especially from someone who generally prefers nuance. Trust me, it is not a conclusion I reached quickly.
It is a conclusion I reached after watching authors become less able to move forward because they had outsourced decisions they needed to understand themselves. It is also a conclusion that follows directly from how I understand writing: if writing is a long series of choices, then the practice of writing requires you to make those choices.
You can absolutely use tools that help you catch an error while leaving the decisions in your hands. Spellcheck is not the same thing as asking an LLM to write a chapter.
But once the tool begins making the choices that create your topic, structure, voice, argument, or prose, I think you have moved into a fundamentally different kind of activity.
That does not mean I am interested in shaming anyone who has already used AI this way. I am not. A lot of writers have been told that these tools are simply the new version of spellcheck and that refusing to use them means falling behind. Others started using them because they were genuinely trying to solve a problem in their writing.
What I want is for you to understand what you are giving up when you hand those decisions over.
You Still Get to Make the Choice
My recommendation is clear, but the decision is still yours.
I do not need everyone to land on my side of every controversy, and I certainly do not want you adopting my position simply because I told you to. I would much rather give you information that you can combine with your own experience and values to decide how you want to work.
But if you are thinking about using AI to write a book, I want you to make that decision with a clear understanding of what the writing process actually asks of you.
Writing is the work of making choices.
Those choices begin with the topic and message. They continue through structure. They multiply through every chapter, paragraph, sentence, and word.
Making them can be frustrating. It can be slow. You will sometimes make the wrong one and have to go back.
But making those choices is also how you develop discernment. It is how your voice gets stronger. It is how you learn to structure an argument or tell a story. It is how you eventually become the kind of writer who can sit down with a difficult idea and know how to work your way through it.
And when you finish the book, you keep all of those skills.
That is why my professional recommendation is to keep generative AI out of the creative writing process.
You do not need a probability machine to make the book sound like everyone else’s best guess at what comes next.
You need the practice of making those choices yourself.
