AI, Write Like an Eighth Grader. Or a Robot. Possibly Both; or,
De l’art de ressembler à ce qui nous ressemble

I can’t decide whether I’m more offended by having my writing mistaken for "AI slop" or by realizing that the alleged evidence consists largely of things I learned in middle school.

I finally received my first DNF review. I knew one was coming. A memoir about a Chamber of Commerce in a town of twenty-five hundred people isn’t exactly a guaranteed crowd-pleaser. For those unfamiliar with the terminology, DNF means Did Not Finish. Mine was the literary equivalent of someone getting halfway through a dinner they enjoyed, putting down their fork, and announcing that while the food was quite good, they have concerns about the chef.

The review was actually rather thoughtful. The reader described my book as interesting, cozy, warm, and easy to connect with. They appreciated its portrayal of small-town life, the struggles of local businesses, and the communities trying to hold themselves together. They even loved the setting. I’ve been considering unironically using that part in my marketing materials. Lemons. Lemonade.

Then came the problem. The reviewer felt the writing sounded "prefabricated," their word. They weren’t sure whether AI had been used to polish the prose, but certain patterns kept pulling them out of the story. In particular, my fondness for the "it’s not this, it’s that" construction. They stopped reading halfway through.

Fair enough. Nobody owes me their continued attention, and I certainly don’t expect everyone to enjoy my writing. But I did find myself staring at the criticism for a while, not because the reviewer was necessarily wrong about the patterns, but because the patterns they identified are precisely the ones I was taught to use.

I like "it’s not this, it’s that." There, I said it. I also use parallelism. I like rhetorical contrasts. I sometimes organize arguments in threes. I construct paragraphs with a recognizable beginning, middle, and end. I learned to do all of those things in school, from actual humans, using chalkboards and textbooks and a seemingly inexhaustible supply of five-paragraph essays. I hated writing those essays at the time. At last I have a legitimate reason for my ire.

Somewhere along the way, the techniques we were taught to make our writing clear, persuasive, and coherent have become the telltale signs of artificial intelligence.

And I realize how we got here. AI writing often leans heavily on familiar rhetorical devices: tidy lists, symmetrical sentences, grand declarations about what something really means—and the persistent insistence that something is not merely one thing but something else entirely. I deliberately included four examples, incidentally, because three of anything now requires a disclosure statement. The em dash is extra credit.

It’s exhausting. I know because I’ve not only read plenty of human writing that does exactly the same thing, but I’ve written it.

That’s the awkward part.

For the record, yes, I used AI during the editorial process. I’ve been quite open about that. I used it to critique, challenge, copyedit, and help refine a manuscript I’d written myself. Sometimes it was useful. Sometimes it was confidently wrong. Occasionally, it was so determined to improve a sentence that it managed to remove everything interesting about it.

I treated its suggestions the way I treated those of my human editor: as suggestions. I accepted some, rejected others, and occasionally had very heated arguments. In AI’s case, those arguments were with a screen. I remain responsible for the finished prose, including its irritating habits. Nobody gets to blame the robots for those.

Anyway, the real AI tell, I believe, isn’t structure. It’s beige prose. That raises a question about my own writing I’m now going to have to ask Claude, and I’m not sure I want the answer. (Beige)

Which brings me back to middle school.

I won a statewide writing award in seventh grade. Presumably that writing employed many of the same rhetorical habits that now get writing flagged as AI-generated. I have no particular reason to believe I was any less fond of parallelism, tidy arguments, or dramatic declarations at twelve than I am today. If anything, I probably took them more seriously. Seventh-grade me was quite convinced of his own literary potential.

Fortunately, I can establish beyond the shadow of AI that ChatGPT didn’t write that essay. It wouldn’t exist for another forty years. Or, if it did, I believe you would’ve needed a university mainframe, a hefty research grant, and several stacks of punch cards to use it. My middle school was considerably less well-funded. I was a precocious kid, but none of my teachers ever suspected I was secretly outsourcing my homework to WOPR.

There’s yet another irony here. AI learned to write by consuming enormous quantities of human writing. It absorbed our conventions, our habits, our clichés, and our occasionally petty enthusiasm for em dashes. For all I know, it even absorbed some of my own adolescent prose. The award is part of the historical record, so the machines can find evidence that I won it. What they can’t tell me is whether the actual essay found its way into their training data. Of course, I’ve asked.

It’s a rather unsettling situation, considering that I spent much of my youth trying to sound like an adult and am now spending much of my adulthood trying not to sound like a machine.

But the larger absurdity remains. We trained machines to imitate human writing, and now our readers are examining our writing for evidence that it resembles the machines we trained to resemble us.

We’ve created a peculiar literary feedback loop in which the better a machine becomes at imitating familiar prose, the more suspicious familiar prose becomes. Somewhere in the middle of all this, the five-paragraph essay has become a scarlet letter. Two of them, actually: A.I. Hester Prynne is going to need a bigger bodice.

There’s a difference, though, between identifying a repetitive stylistic habit and determining who, or what, wrote something. The first is literary criticism. The second is an attribution claim. And they’re not the same thing.

A reader is entirely justified in finding a particular construction irritating. I personally would like to eradicate semicolons from the face of the Earth. But if my own habit of contrasting one idea with another becomes predictable enough to interfere with the reading experience, that’s useful feedback. I may not like hearing it, but it’s fair. And if the reviewer suspected AI-assisted polishing, that wasn’t an entirely unreasonable question.

But recognizing a familiar rhetorical pattern doesn’t tell you whether a human or a machine produced it, or whether an editor of either variety subsequently suggested changing it. What strikes me as peculiar is how readily we’ve begun treating stylistic familiarity as evidence of artificial authorship.

As if humans had never been formulaic before, or entire generations hadn’t been taught to introduce a thesis, present three supporting arguments, and conclude by restating the original point in slightly more sophisticated language. As if the English teachers of America hadn’t spent decades forcing us to manufacture exactly the kind of prose we’re now treating as evidence of artificial manufacture.

I spent considerable time writing a book about the complicated, funny, frustrating business of trying to help a small community—about, among other things, why the same six people keep getting stuck setting up the folding chairs.

The part that’s stayed with me is that the reader seemed to like that story. They connected with the subject, appreciated the setting, and even encouraged other readers to give the book a chance. They just couldn’t get past the writing patterns, and that’s entirely their prerogative.

Because the rhetorical habits are mine. The occasionally overwrought observations are mine. Even the fondness for a dramatic sentence that probably thinks rather highly of itself is mine. I’m perfectly capable of producing formulaic prose without technological assistance, and I have an award to prove it, thankyouverymuch.

But there’s a certain indignity in having your entirely human literary shortcomings attributed to a machine, particularly when you’ve spent as many years as I have developing them.

The book reflects my experiences, my observations, my sense of humor, and, evidently, my inability to resist a well-placed rhetorical contrast. AI helped me examine that manuscript. It also tried, with remarkable persistence, to make my writing sound like something produced by a committee of extremely earnest management consultants. Which is probably a larger audience than the one my own book will eventually find.

We had disagreements, some of them knock-down-drag-out, but the voice is mine, for better or worse. And now, among the other unexpected lessons of becoming an author, I’ve discovered that readers may recognize your voice so clearly that they mistake it for someone else’s. Or something else’s.

My first instinct was to respond to the review with an outraged diatribe, something long, indignant, beautifully structured, and undoubtedly organized into three devastating arguments. I even wrote one. Fortunately, I thought better of sending it. I’m entitled to my opinion, but so is the reviewer. They gave the book a chance, explained why they stopped reading, and even encouraged others to try it. That’s more consideration than many authors receive.

Besides, when I asked my AI overlords to chime in on the harangue, they weren’t particularly enthusiastic about helping me refine it. Apparently, even the robots think I need to work on my tone.