Recently Bona Books posted a discussion of how they unwittingly bought an AI-generated story for an anthology — more than one, actually, but one of them from the (now infamous) scammer Bella Chacha.

There is understandably a lot of distress right now in publishing over how to catch these things and how to deal with them. In reading Bona Books’ post, though, I found myself thinking about an angle I’ve not seen discussed much at all.

Well before AI became a problem, there were conversations about the “fast fashion” mentality manifesting in publishing: flimsy products with very little staying power, but produced quickly to capitalize on the trend of the moment. Amazon’s algorithms bear a fair bit of responsibility for this; they have always tended to increase visibility for those who put out work on a quick schedule, so if you’re self-published and want to succeed, one of the ways to improve your odds is to write and release very very fast. But of course, if you’re operating on that kind of treadmill, you don’t have a lot of time to develop depth and richness in your writing — you may not even have time to copy-edit! — so this pushes the ecosystem toward authors churning out endless iterations of simple tropes, and never mind whether there’s much of a there there. It’s not all self-published authors, naturally, but the selection pressure is there. And meanwhile, you get books taking the online world by storm where even the people recommending them are laughing about what hot messes they are, but the vibes.

Now, before I thoroughly cement my curmudgeon cred here, I should note that this is hardly a brand-new phenomenon. Dime novels were similarly cheap, quickly produced, and intended to cater to trends; so long as they delivered exciting tropes, it didn’t much matter if their prose was rough, their plots full of holes, their characterization non-existent. But dime novels didn’t have to contend with AI.

With AI, you can mass-produce this type of thing vastly quicker than any human being can hope to do. AI has been trained on all those tropes and can spit out more or less competent prose in industrial quantities. What it can’t do very well is the big-picture stuff, because that requires you to have an actual thinking brain that can make decisions about how all the smaller bits fit together. AI just takes one step after another until it reaches the end of what it’s been told to do.

So now I want to loop back to that Bona Books post. Let me pull out two specific quotes about the Bella Chacha story:

It hit our brief with a perfect intensity that felt difficult to resist. A queer woman, building a machine from scrap metal and rage to destroy her colonial oppressors? Of course we loved it.

and

The closer we examined that first draft, the more the story’s centre seemed absent. The protagonist was active, but lacked interiority. The worldbuilding had colourful details that somehow failed to build consistently. Certain hallmark sentences recurred over and over (“Not X, but Y…”). And the pacing was off, with important story beats given oddly little narrative weight.

Most of the discussions around catching AI-generated stories, and (to the best of my knowledge) all of the detection programs, have focused on prose-level markers: overuse of em-dashes, overuse of the “rule of three,” vague generalities, and so forth. One of those shows up here — the “not X, but Y” example — but most of the red flags the editors found on a re-read are issues of fundamental storycraft. Exactly the kinds of points you might hear raised in a critique group, where readers are primed to look for the flaws.

So why did Bona Books initially think the story was good enough to buy? I can’t speak for the editors, but based on their own statement, it sounds like the concept and vibes carried them away. Those were so shiny, people didn’t see beneath the surface to the gaps and weaknesses below.

And that’s why I’m thinking about “fast fiction” today, and how it relates to AI. Even before ChatGPT hit the scene, Amazon’s algorithms were training a lot of readers to accept fiction that prioritizes concept and vibes over skill in execution. The two aren’t incompatible; in fact, a story that mashes all the id buttons with solid craft to back it up is amazing! But treating the latter as an optional side bonus creates a situation where, sure, why not have AI write your stories? The characterization and worldbuilding and pacing of the plot aren’t important anyway.

I am, of course, not an editor. (I have entertained thoughts of it from time to time. I currently have no time and also, quite frankly, am not sure my sanity could survive the ongoing hurricane.) But so many people out there want to be able to just pop some text into an AI checker — which itself often runs on AI! — and get an answer back on the spot. Yet we have a thousand examples of those programs giving both false positives and false negatives, while often feeding the same industry that’s causing the problem in the first place.

(I’m pretty sure “create a problem and then sell the solution” is the tactic used by any number of villains in thrillers and superhero fiction . . .)

What we don’t have — not that I’ve seen anyway — is more conversation about using your actual human brain to look, not at the prose, but at the narrative craft of the piece. I would absolutely not advocate for using that to accuse people of deploying AI; at this level, a bot-generated piece is indistinguishable from a writer who’s simply not very good yet. But if we ask more of our fiction than just emotional intensity and some shiny bits sprinkled on top, we might do more to weed out AI-generated stories than any amount of running them through POS checkers like Pangram.

Now you may ask, can’t somebody generate an AI story and then revise it to have good craft? In theory, yes. In practice, there are two hurdles in the way of that. The first is that, if you aren’t in the habit of actually writing the fiction yourself, then you’re unlikely to be great at fixing it, either. It’s like letting a robot pile up a bunch of heavy objects for you, and then trying to rearrange them into a nice shape with the muscles you didn’t build by collecting them in the first place. And the second is that — based on what I’ve heard from freelancers hired to “clean up” AI-generated text for companies — the cleanup process is often more obnoxious and more labor-intensive than it would have been to just write the damn thing themselves.

But sure, somebody might do it. Probably already has done it. I have issues with that person, because I have issues with AI that go well beyond the fact that it’s bad at storycraft, but right now I’m focused on a different battle. I’m looking at Bona Books’ statement and wondering why a story with such fundamental flaws seemed good enough to accept in the first place. (Especially since, by their own admission, “Bella Chacha’s” revisions didn’t address those flaws at all.) I’m feeling empathy for the writers in their slush pile whose works maybe weren’t as flashy on a first look, but had more substance under the hood. I would rather have seen them given a chance.

We deserve stories that fire on all cylinders, not just the brightest and loudest.