If the Book Was Slop, Why Did Fourteen Publishers Bid on It?
Using AI to detect AI means we don’t trust our own read.
A debut thriller went to auction recently and fourteen publishers bid on it until Minotaur won it as a two-book deal for a reported $2.5 million. Then the rumors started that the manuscript might be AI-generated, because it had flat prose, “off-kilter metaphors,” and negative parallelism. Detection software scored it 97 percent AI, and the book got pulled.
The thing I want to point out is that yeah, suspicions were raised during submission, but they were suspicions of AI rather than complaints about the prose. Those pages, hallmarks and all, made it to a fourteen-way auction among some of the sharpest editorial readers in the business. I didn’t read anything that said they thought the writing sucked. It was about what the writing might indicate.
Does my butt look like AI in this outfit
If the hallmarks of AI writing are a real problem — and I think the flatness, for example, is something you can feel — then they should have been more of a problem prior to the auction while the book was dazzling everyone.
Of course, a deal gets made not just because the writing is good. You can have that and never pen your way to a $2.5 million deal. The agents’ stated reason for nixing it was trust. They author’s story changed, they say, on the “did you use AI” question. Were you in the same room as any gluten or AI products during the making of this thriller?
The author says he’s completely innocent, that nobody asked him anything before the rumors took over, and from his perspective, even his own agency didn’t have his back. I have no idea which account is true. But does the software?
Authors warrant originality in their contracts, and text with no human author may not be copyrightable. A publisher paying seven figures wants to own what it’s buying. But notice that none of those reasons required anyone to re-read the book.
Again, always with the AI.
With LLMs there are lawsuits aplenty, and there are well-reasoned arguments about them being weapons of mass plagiarism, and yet as all this happens there’s a wide range of actual use happening. Some people won’t touch it, some research with it, some people draft with it, some people run their finished pages through it and don’t realize how much it rewrites. All of that is a huge debate, and I’m sure it’s all very interesting. But it’s stupid to react to the label instead of the writing.
Above all, we shouldn’t be trusting AI tools to make this call for us. Instead of editors and agents making the call, that gives us one machine hired to sniff out whether another machine probably touched the thing. The software company claims one false positive in 24,000 documents. Again, I have no idea whether to think that’s good or bad. Because, to me, the bigger point is that if a manuscript needs forensic analysis to reveal that it’s hollow, either it isn’t hollow or editors and agents weren’t trusted to do their job based on the work’s own merits.
LinkedIn burped out a button this week that lets you flag a post as AI slop. The same company has been straining just as hard to persuade you to use its other new feature: improve your writing with AI. One step forward, one step back, packaged together. Substack launched a detector of its own this month, and the EU wants AI-generated material watermarked. But if the solution is just to scarlet letter it all, the result is that people learn to scrub the letter, or make it fashionable, until sentiment eventually shifts, and we still haven’t learned to say whether a piece of writing is any good.
Meanwhile, some working writers are out in the trenches second-guessing em dashes while others are carefully revising and fretting: can they tell? If detection becomes the standard we trust over our own felt experience, arbitrary stylistic choices become punishable, and RIP Dickinson for having too many em dashes.
What reading is for
I’ll sell you a bona fide quality detection tool: assess the work on its merits and on how it leaves you feeling. The instrument for that is called reading. If the pages are inert, say they’re inert. If the pages carried a room full of professionals into a fourteen-way auction, then something in them worked, and that says something about what we want to buy. And, granted, we assess for marketability, and that’s different from whether the metaphors are thoroughly on-kilter. If legal clarity existed around AI-generated prose being copyrightable, would the deal have gone through?
It shouldn’t go unnoticed that an industry that treats self-publishing as the slop tier just got caught prepping for hardcover the seven-figure thriller the classifier calls 97 percent machine.
If you write, you write because you want to, and you get better by spending time in the process. Skip the process and you’ve made a trade.
The process isn’t so fixed that one way is pure and another way damns you. I talk drafts into my phone on walks. I write on a sixty-year-old typewriter. I rewrite until the rat’s nest of language shapes into something more like a ferret’s nest. I’m not going to draw conclusions based on whether you work longhand or dictate or hammer keys or use AI. People can learn to be active and present no matter what their creative process looks like.
I’ve had several editing clients finish books because of AI. Knowing how hard it is to finish a book, and how valuable it is to finish, if the tool helps you lean in, engage, and get ‘er done, then that’s worth something. And when you engage like that, you also come to know, firsthand, how AI flattens things. Some of what could have been best, if you’d put in the years and built the discipline, won’t survive an AI revision.
More completed books means an influx of new editing work for me. But more importantly, it shows that there’s a sizable percentage of people who aren’t engaging with the AI-good-AI-bad shouting. They’re getting work done and getting a firsthand feel where it helps and where it sucks. Editing AI prose can be a special kind of hell. But the same goes for any bad writing. When there’s an empty section, you reach out to the author and have a conversation. It’s fixable by being creative, leaning in, bringing the force of your consciousness to a situation that is not fixed and not concluded.
The hardest thing for me with learning to write has been learning the process. There are a lot of other difficulties along the way but figuring out my process is the thing that is not instinctual and doesn’t come natural. It’s harder than the writing itself because it enables the writing. It opens the doors and clears the space.