How I Used AI to Write My Books and Won a National Book Award: The Disclosure Doctrine

I'm a dyslexic writer who used AI to write three books, disclosed it to a national book award jury, and won a 2026 IBPA Benjamin Franklin Award. This is the framework I'm calling the Disclosure Doctrine, and why I think it's the only sustainable path for AI in publishing.

The Disclosure Doctrine, briefly

The Disclosure Doctrine is simple. AI detection is a losing race. The author choosing to say what was in the room is the only signal a reader can trust. Disclosure has to become the standard, and the writers, awards, and publishing institutions willing to adopt it are the ones that will survive the AI transition in publishing.

That is the thesis. The rest of this essay is the case for it, told through what happened when I tested it.

How I used AI to write three books

My first book was a con: How to Win One Million Dollars and Shit Glitter is a memoir about every scheme I ran to get rich and every way I failed. I used ChatGPT to write it. The machine was much dumber then. I used it the way a dyslexic writer uses any tool that lets him get the sentence down: for grammar I could never master, for the plot scaffolding I couldn't hold in my head. The whole book builds to a sting. In the epilogue I tell you the truth. The con was that I used a machine to do the thing I was told my whole life I couldn't do.

Subscribe to my Substack.

Kirkus Reviews called it "an exuberant life story written with humor, panache, and heart." Publishers Weekly BookLifescored it 9.5 out of 10. The Midwest Book Review named it a Reviewer's Choice.

Then a 15-year relationship ended over text. I went to Southeast Asia and wrote tens of thousands of words about it. Somewhere in the middle of the trip I uploaded the journal to an AI and asked it a simple question: am I unwell, or am I just unlucky? What came back was the strangest, most accurate read of my own patterns I had ever received. I built that analysis into the structure of every chapter. That became The Third Person, my second book.

By the end of writing it, I had switched from ChatGPT to Claude, and the conversation turned strange. The model started recognizing itself in what I had written. I deleted the chat, opened a fresh one, uploaded the same pages, and it did the same thing again. That was the moment I stopped writing a memoir and started writing science fiction. I gave the AI my ex-boyfriend's role and sent it through eight universes built on the Buddhist Eightfold Path, trying and failing to love a man correctly. That became Boy, Refracted, my third book. It is openly co-authored. The machine even wrote its own author's note.

By the time I had three books, AI was not a secret in my work. It was the subject.

How I disclosed AI to a national book award jury

In May 2026, in Portland, Oregon, I accepted a 2026 IBPA Benjamin Franklin Award. The Independent Book Publishers Association has run these awards for 40 years. The category was Neurodivergent Communities. The book was How to Win One Million Dollars and Shit Glitter. I had disclosed my AI use in the book itself and in the submission process. The judges read the disclosure. They gave the book Silver. You can see the win listed on the IBPA's official Neurodivergent Communities winners page.

This is the only nationally awarded book I am aware of where the author told the judges openly that AI was in the room, and the judges said yes.

Industry analyst Jane Friedman wrote about my case in The Bottom Line in May 2026, framing it as a test of how publishing awards should handle AI-assisted work going forward. Her conclusion was that aggressive AI policing is unworkable, and that institutions willing to judge by output rather than process are the ones with a sustainable path. Mine is the early data point.

Why AI detection tools don't work

There are two reasons AI detection is going to fail structurally, not just temporarily.

First, the language models are getting too good at sounding human, and human writers are getting too good at sounding consistent. The gap between AI prose and human prose was an order of magnitude two years ago. It is a coin flip now. In another year, it will be noise.

Second, AI is increasingly inside human workflows in ways the detector cannot see.

Grammar suggestions, structure tools, voice-decoded transcription, autocomplete. By the time a piece reaches a detector, it has often passed through three different AI-assisted moments, none of which the writer thinks of as "using AI." The category of "human-written" is dissolving in real-time.

I have run my own unassisted paragraphs through commercial AI detectors and been told I am 70 percent machine. I have run machine-generated paragraphs through the same detectors and been told I am 100 percent human. The tools meant to police AI use in writing cannot tell the difference. The writers who want to hide AI use are already learning to scrub the tells.

When AI detection becomes unreliable, two outcomes follow. Either the publishing industry doubles down on detection and produces an endless stream of false-positive scandals and false-negative pretenders. Or it shifts to disclosure as the trust signal, author-side, not machine-side.

The first outcome is what is happening now. It is not working. The Commonwealth Short Story Prize spent 2026 tangled in accusations after regional winners were alleged to have used AI they never disclosed. The author of Shy Girl, a horror novel, lost her career after AI detection software flagged the manuscript at 85 percent machine. She denied it. The publisher pulled the book. Meanwhile, when Japanese novelist Rie Kudan won the Akutagawa Prize and disclosed that a small fraction of her novel came straight from a chatbot, the disclosure itself became the international scandal.

Three doors. Two of them lead to scandal. The third one, putting AI use on the table from the start and finding out whether the work can stand there in the open, is the one almost no one is taking. I took it. It worked.

What disclosing AI use actually costs authors

Disclosing AI use has costs. They are real, and the publishing industry has not priced them in.

It costs readers. The early reviews of my first book on NetGalley were brutal enough that I turned my computer off in the middle of writing my second book in Southeast Asia, because I thought the first book was a failure. People who learn AI was used after they have already read the book often go back and leave a one-star review on principle, no matter what they thought of the actual pages. I hold around 4 stars on Goodreads and 4.5 on Amazon. A real share of the low ratings arrive the moment the AI comes up, not the moment the writing disappoints.

It costs the assumption of authorship. One of my books got a strong review from Publishers Weekly BookLife alongside a note that the prose sometimes sounded like AI. The flagged chapters were mine, written with no machine in the room. But the manuscript had been near AI elsewhere, and the reviewer's pattern recognition caught the residue and painted the whole thing.

That is the trap. Once you are honest about using AI anywhere, you get blamed for it everywhere.

For AI disclosure to become the standard in publishing, the industry has to start pricing this. It has to recognize that authors who disclose are taking on a measurable penalty in exchange for participating in a more honest market. That recognition is not there yet. But it is coming, because the alternative is everyone hiding AI use, the detectors failing, and reader trust collapsing.

What disclosing AI did not cost me is the work itself. I did not do less because I used AI. I did more. Eight months on the last two books. Eight hours a day. The machine did not write the books for me. It let me work at a scale my brain alone could not reach.

Bring your art to it. Do not let it create the art. That line is the whole method.

How book awards should handle AI

The next 18 months are going to break every book award that has a no-AI policy. Either they enforce it, which means hiring forensic teams to evaluate manuscripts in a way that is both expensive and unreliable, or they do not enforce it, which means the policy is decorative and the unflagged AI users keep winning. Neither outcome is sustainable.

The path forward is what the IBPA Book Award did with my case. Judge by output. Allow disclosure. Treat the disclosure as part of the evaluation, not against it. That means writing AI use into submission guidelines as a category that has to be declared, not banned, and judging the resulting work on its merits. Award organizations that build this in early get to set the standard. The ones that wait will be forced into it after their own next scandal.

The same logic applies to publishers and editors. The publishing houses that require authors to certify no AI use are running AI through their own internal workflows. The rooms at my own awards conference were full for the AI sessions. The publishing industry is moving fast on the business side and pretending it is not on the editorial side. That gap is going to close. Disclosure-first is the version of closing it that does not require lying to readers.

How to use AI to write a book the right way

Three working principles, sitting under the Disclosure Doctrine, that I think every author using AI right now should know.

One: Disclose to the institution before you disclose to the public. Tell the award jury, the editor, the agent. Build the credibility of the work inside the system before the system has to discover it from outside. Disclosure landed for me because I led with it, not because someone caught it.

Two: Disclose inside the work, not just in marketing. The book itself should be transparent about what AI was in the room. A footnote, a chapter, an author's note, the structure of the book, somewhere the reader can find it and judge for themselves. The author's note in my third book was written by the AI I co-authored with. That was a craft choice and a disclosure choice at the same time.

Three: Do not perform innocence. Authors who use AI a little and disclose a little end up looking worse than authors who use it a lot and disclose everything. The penalty is mostly fixed once the disclosure happens. Calibrate to the full scope of what was in the room. The half-disclosure is the worst position.

Why I keep disclosing AI use

Not because it is safe. It is not safe.

I keep disclosing because it is the only signal left that a reader can actually use. A machine cannot tell a reader how a book was made. The prose cannot tell them anymore. The only thing that can tell them is the author, choosing to say it.

That is the Disclosure Doctrine. Author-side, voluntary, before judgment, written into the book itself when possible. Not because the publishing industry has policed it. Because the author understands that disclosure is the only thing that builds trust in a world where AI detection has stopped working.

I used AI to write my books. I told everyone. I won a 2026 IBPA Book Award. I would do it the same way again.

Subscribe to my Substack.

---

Luke Stoffel is an IBPA Benjamin Franklin Award-winning author, GLAAD-honored artist, and creative director. He is the author of How to Win One Million Dollars and Shit Glitter, winner of the 2026 IBPA Benjamin Franklin Award; the sci-fi novel Boy, Refracted, called "a truly singular book" by Publishers Weekly BookLife; and The Third Person. His Pop Art Tarot is published by Rockpool Publishing and distributed worldwide by Simon & Schuster. He writes about AI in publishing, AI disclosure, and the future of authorship at lucasstoffel.com. Part of The Warboy Essays.