Publishing Has an AI Problem, but I Think We're Asking the Wrong Question
There is an interesting article in the Wall Street Journal today about the problems generative AI is creating inside publishing.
A number of very substantial book deals have now fallen apart after questions were raised about whether manuscripts were actually written by the authors whose names appeared on them. In one case discussed by the WSJ, literary agency Europa Content withdrew its support for Jerry Falade's crime novel after saying it could no longer verify that the manuscript had been wholly written by him. Falade has denied using AI to write the book. (The Wall Street Journal)
There are all sorts of difficult issues sitting underneath individual cases like that, particularly because AI detection is still imperfect.
The part that interests me more is that publishing still doesn't appear to have agreed on a much more basic question.
What exactly counts as unacceptable AI involvement in a book?
That sounds as though it should have a fairly simple answer.
It doesn't.
“Did you use AI?” isn't a particularly useful question anymore
Imagine four authors.
One asks ChatGPT to help research a subject before writing a chapter.
Another writes a chapter themselves and asks AI to identify gaps in their argument.
A third gives AI an outline and asks it to produce a first draft which they then rewrite.
A fourth asks AI to generate almost the entire manuscript.
All four could truthfully answer yes if somebody asked whether they had used AI.
Yet I don't think most readers, authors or publishers would regard those four situations as equivalent.
This is where publishing seems to be getting stuck.
The WSJ reports that the major publishers have largely resisted introducing sweeping rules around generative AI, despite the growing number of controversies surrounding it.
You can see why.
The technology is already embedded in research tools, editing software, search, transcription and all sorts of ordinary working processes. Drawing a line around “AI use” quickly becomes rather difficult.
Some organisations are starting to draw much more specific lines instead.
The Authors Guild has drawn one boundary
The Authors Guild now operates a Human Authored certification programme.
Its definition is quite interesting.
A book can still qualify when AI has been used in relatively limited ways, including research and tools that assist with spelling and grammar. What cannot happen is AI generating the actual authored text beyond a very small amount. (Authors Guild — Human Authored FAQ)
That feels much more useful than simply declaring a book either “AI” or “not AI”.
It is beginning to distinguish between using a tool during the process of writing and delegating the authorship itself.
The Authors Guild has also updated its AI best-practice guidance to reflect this more nuanced position. It now talks explicitly about different levels of AI use and the legal and professional risks writers should consider when using these systems. (Authors Guild — AI Best Practices for Authors)
Other parts of publishing are drawing harder boundaries.
Simon & Schuster's rules for its Avid First Novel Award state that entrants cannot use generative AI, machine learning or large language models to create the content they submit. (Simon & Schuster — Official Rules)
So we already have different standards depending on the context.
I suspect we are going to see much more of this.
I think provenance is going to matter
For an expert author, there is another layer to this that I think is particularly important.
A book is often carrying much more than words.
It may contain twenty years of experience with clients, a methodology that has gradually evolved, mistakes that changed the author's thinking, original frameworks, judgement and the thousand small observations that eventually became something worth writing down.
That is intellectual property.
If AI helps somebody organise that thinking, interrogate it or notice that something is missing, I'm fairly comfortable with that.
I already work with AI in ways that help me think.
Where I become much more cautious is when the AI begins quietly supplying the thinking itself and the distinction becomes invisible.
Because at that point, whose intellectual property are we actually reading?
That question matters enormously for expert-led books.
This also explains why I make a distinction around AI Book Companion™
I've been thinking about this a lot while developing the AI Book Companion™ framework because there is an important distinction that could easily get lost.
An AI Book Companion™ does not need to write the author's book.
The book already exists.
The author's intellectual property already exists.
The Companion is built around that defined body of thinking and gives the reader a different way to engage with it.
They might finish a chapter and realise they don't completely understand one of the author's ideas.
They can ask.
They might want to explore how the framework could apply to their own business.
They can work through it.
They may have read the book six months ago and suddenly find themselves in a situation where one particular idea becomes relevant.
They can return to it.
The AI is sitting between the reader and a body of human-created intellectual property, helping the reader interact with it.
For me, that sits in a very different place from asking an AI to manufacture the original intellectual property and then putting an author's name on the front.
We probably need better language
I suspect part of the problem publishing has at the moment is that our language hasn't caught up with what people are actually doing.
We talk about an “AI-written book”.
We talk about an “AI-assisted author”.
We ask whether somebody “used AI”.
Those descriptions are becoming too broad to be particularly useful.
We may eventually need to describe the role AI played in creating a book with the same sort of precision that we already use for editors, ghostwriters, researchers, illustrators and translators.
Because there is a substantial difference between AI helping an author test an argument and AI creating the argument.
There is a substantial difference between an author asking an AI to proofread their own prose and asking it to produce 60,000 words.
And there is a substantial difference between using AI to create a book and using AI afterwards to help a reader get more from the author's existing work.
The publishing industry is now being forced to work out where those boundaries sit.
I think that conversation is overdue.
Three Key Insights
- Publishing still lacks a shared standard for what constitutes acceptable AI involvement in authorship, and different organisations are beginning to establish different boundaries.
- Simply asking whether an author “used AI” increasingly tells us very little about who actually created the intellectual work.
- For expert authors, being able to explain the provenance of their ideas, methodology and manuscript may become an increasingly important part of establishing trust.
Three Questions to Consider
- If you use AI while writing, could you clearly explain which parts of the thinking and writing remain yours?
- Where would you personally draw the boundary between AI assistance and AI authorship?
- If AI is used after a book has been written to help readers interact with the author's thinking, should we regard that differently from AI being used to create the book itself?
Sources
The Wall Street Journal
AI Has Plunged the Book Publishing Industry Into Utter Chaos
Published: 18 August 2026
https://www.wsj.com/arts-culture/books/generative-ai-book-publishing-be79a287
Authors Guild
Human Authored FAQ
https://authorsguild.org/human-authored/faq/
Authors Guild
AI Best Practices for Authors
Updated: 11 May 2026
https://authorsguild.org/resource/ai-best-practices-for-authors/
Authors Guild
Authors Guild Launches Expanded “Human Authored” Certification to All Authors
Published: 2 March 2026
https://authorsguild.org/news/human-authored-certification-expands-to-all-authors/
Simon & Schuster
Avid First Novel Award — Official Rules
https://www.simonandschuster.com/p/avid-first-novel-official-rules
EmpowerAi™ | Real AI. Real Impact. Smarter Business Decisions.
AI Book Companion™ helps expert authors turn a completed book into a governed, interactive knowledge asset built around their own ideas, methodology and boundaries.