AI Is a Function, Not a Ghostwriter

I record a lot of conversations with people who run things. The useful signal is when unrelated guests say the same thing without having heard each other.

That happened this summer. Joe Rockey, who works with small business owners, said AI is a function and never a relationship, and that the line between those decides who wins with it. Craig Leonard, on personal branding, said use AI as an advisor and never as the writer. Noah Landow, thirty years into running an IT firm, said the interesting use inside a company is removing drudgery instead of removing people. Susan Evangeline Bagh, who advises on transformation, said most of the current wave is hype and infrastructure built ahead of proven need.

Four people, four industries, one boundary, and it maps onto what I do for a living.

What can AI do in book work?

A great deal. I use it daily, so this is not a purist argument.

Transcription and organization. A book project generates twenty or thirty hours of recorded interviews. Turning that into searchable, sorted text used to be weeks of somebody’s life and is now a background task, and that change is good.

Research triage. Give it forty sources and ask which five address a specific question. It will be wrong sometimes and it will save hours anyway, provided you check.

Structural interrogation. Paste an outline and ask what is missing, what repeats, what a reader would ask that goes unanswered. It is a competent reader, as I said in my piece on AI-written profiles, and reading is a real skill to have on tap.

Mechanical passes. Consistency checks, chronology, whether the name of a company is spelled the same way in chapters two and nineteen. Machines are good at this and humans are terrible at it.

All four are drudgery. Noah’s formulation is the right one: ask your team what they hate doing, then hand that part over.

What can it not do?

Be you, and that is the product.

A memoir or an authority book sells on the fact that a specific person with a specific history is saying this. Strip that and you have a competent article on a subject, available free everywhere, with no reason to be a book.

A model cannot supply three things, and no prompting changes that.

It was not there. It does not know what the floor smelled like, what your boss said in the parking lot, or which of the two versions of that story is the true one. Those specifics are the difference between a book people finish and one they abandon at page forty.

It does not know what matters. Given a career, it cannot tell which decision was the hinge. It will weight by how much text you gave it, which correlates with what you find easy to talk about, which is usually the opposite of what the book needs.

It smooths. Output moves toward the familiar, because that is what a probability model does. Your material is valuable where it is unfamiliar, and that is the part the machine sands off first.

Joe Rockey’s version of this, about business generally, applies without modification. Anyone buying something important to them wants a human. An AI sales bot only sells to people who do not care. A book somebody puts their name on for the rest of their life is something important to them.

Where is the line in practice?

I draw it at the sentence.

Everything up to the sentence is fair game: recordings, transcripts, sorting, checking, questioning the structure, catching inconsistencies. A person who has spoken to the author for hours writes the sentences themselves, and knows which of their phrases are theirs.

That line is where the value sits. A client is paying for their voice on a page, and a voice consists of sentence rhythm, word choice, and what the writer leaves out. Those are the three things a model normalizes.

It also has a practical edge now. Readers have developed an ear for the register, quickly, and so have search and answer systems. Publish text under your own name that reads like the average of the internet, and you trade a durable reputational cost for a short-term saving. I argued that in what AI does to writing when nobody is watching.

Then why offer an AI-assisted service at all?

Because done properly it is good, and hiding from the word helps nobody.

My AI-assisted books service exists for people who want a shorter authority book at a lower price, and it lowers the price by using machines for the drudgery. That cuts hours, not the author. The interviews still happen. The structure is still decided by a person. The final pass is still mine. It runs at roughly half the per-word rate because the mechanical part got cheaper, and I set out the details in the AI-assisted book at half the cost.

What I will not sell is a book a machine wrote. There is principle in that, and there is also the fact that it would not work. The client would get a fluent, agreeable, forgettable document with their name on it, and the result defeats the reason to publish.

The disclosure question

People ask whether an author should disclose AI involvement, and my answer has hardened over the last two years.

Tell the client everything, in writing, before the contract. Which parts of the process use machines, which do not, and what they are paying a person to do. Any ghostwriter who gets vague when asked that question is telling you something. Craig Leonard applies the same test to profile writing.

Public disclosure on the book itself is a different question and depends on the platform, since retailers and audiobook distributors have their own rules and those rules keep changing. Check the current terms wherever you are publishing, because a policy violation discovered later is a far worse outcome than a line in the front matter.

My own position is simple enough to say in one sentence: the client knows exactly what we did, and no machine wrote a sentence in a finished book.

The transformation lesson

Susan Bagh made a point about organizations that applies to a single author. Transformation has to be business-led and top-down, or it fails, and roughly three quarters of them do fail.

Translated: decide what you are trying to achieve, then choose tools. The failure mode is the reverse, adopting a tool and looking for a use, which is most of what is happening with AI right now and most of why so little of it sticks.

For a book that means the same sequence I use for every project. Decide what the book is, who it is for, and what it has to do. Then decide which parts of making it are drudgery and hand those over. Anybody who starts with the tool ends up with a book shaped by the tool’s preferences, meaning the average of everything.

The short version

Use it for what nobody would miss doing. Keep the part that is you.

Four people with nothing in common arrived at that boundary independently. A rule of thumb rarely gets more confirmation than that. Their full conversations are on the podcast: Joe Rockey on AI as a function, Noah Landow on machines that stopped being predictable, and Susan Evangeline Bagh on seeing the whole chessboard, with more in the Leaders and Their Stories hub.

If you are weighing how to get a book made and where the machine belongs in it, the Feasibility Intensive settles what the book is first. That is the order that works.

Frequently Asked Questions