This article first appeared in Book Gossip, a newsletter about what the literati are really thinking. Sign up here to get it in your inbox every two weeks.
Last month, two incidents appear to have shifted the conversation around AI and publishing for good: (1) Jamir Nazir’s prize-winning short story “The Serpent in the Grove,” chosen by the Commonwealth Foundation and then hosted on Granta’s website per a preexisting agreement, was suspected of being written using an LLM, and (2) author Steven Rosenbaum said that ChatGPT “fucked up” his book about how AI is reshaping reality by inserting a number of misattributed or totally fake quotes. (You couldn’t prompt Chat to write a more on-the-nose irony than that.) While book publishers, who do not systemically fact-check longform nonfiction works, appear particularly vulnerable to unknowingly propagating AI hallucinations, literary-magazine editors are now faced with the dilemma of whether to screen submissions for prose generated with the so-called help of AI and how.
“The Serpent in the Grove” struck a nerve with readers and writers, but lit-mag editors are not yet panicking about the possibility of AI-inflected stories overwhelming their inboxes. “I think we’d be having a different conversation if the technology could do the things we like and want,” said Samuel Rutter, editor-in-chief of the relatively new mag Kismet. “We’re still working with a lot of writers for whom the ideating and the writing is almost the more exciting part than the publishing,” he said. Joyland co-editor-in-chief Walker Caplan agreed that any AI-authored stories would be weeded out through their usual editorial standards organically, arguing that conversations among Joyland staff about AI dovetail into existing questions about the type of work they want to publish in general. “We don’t want to accept writing simply because it sounds like a plausible or optimized story,” Caplan said. “We want to accept stories that have a singular, even idiosyncratic vision.” Editors who prioritize style are already hunting drafts for the type of muddy turns of phrase and mixed metaphors that LLMs are known to deploy.
Unlike news organizations, lit mags have the benefit of taking their time. Per N+1 co-editor Lisa Borst, lit mags are a “zone of beautiful, countercyclical inefficiency. We publish pretty infrequently. The work isn’t meant to be super-easy to digest or to write.” It’s an intensive process that involves the type of back-and-forth between editor and writer that Borst believes would double as a screen for AI. “The question is, ‘Can this person nimbly respond to your suggestion or defend their own work?’ And if they can’t, that to me is the biggest red flag,” she said.
That doesn’t mean there won’t be unintended negative consequences for new writers especially. A couple of the editors I spoke to were concerned that, as magazines increasingly lean on writers they already know and trust, it may become more difficult for emerging talent to break into publishing. Still, many publications remain committed to discovering new talent, even if it costs them time and energy. Kelley Deane McKinney, managing editor at The Paris Review, said when they find a debut writer they want to work with, they might give feedback on several stories before finally publishing them: “Art isn’t produced via efficiency; good writing isn’t better than bad writing because it’s efficient.”
For the most part, even after the hubbub around Nazir’s story, lit-mag editors aren’t necessarily jumping to adjust the terms of their submissions guidelines or author contracts specifically to safeguard against slop. Some already have certain measures in place; The Yale Review added AI guidelines to its submissions page as of this past fall. While many editors are optimistic that writers would understand the existing anti-plagiarism language in their contracts to already exclude the use of AI, others are actively building in more explicit protections against LLMs or thinking about doing so. For instance, Rutter has added AI-specific language into Kismet’s contracts as well as a stronger stipulation: that Kismet would reserve the right to “exercise final determination over the question of AI use” in cases of doubt. Because it’s essentially impossible — at least for the time being — to determine with certainty where AI was used and where it wasn’t, the editor could terminate a story regardless if something felt fishy.
And what of AI detectors? Of the eight magazine editors I spoke to for this story, none of them were interested in using one to sniff out submissions. Even the most advanced detectors, such as Pangram, can be finicky, and many editors are ethically opposed to submitting writers’ work into AI and thereby feeding the environmentally disastrous beast — a beast that was already trained on published authors’ writing without consent.
As essential as it may now be for lit-mag editors to build familiarity with AI’s “style” or “accent” to act as their own AI detector, requiring them to do so can feel like just another thing to ask an already overworked and largely undercompensated group of people. The Yale Review’s executive editor, Meghan O’Rourke, thinks that while it’s helpful for editors to understand the capacities of LLMs, she doesn’t believe they can catch every single instance of their use. “The work of being a magazine editor is to build a relationship based on trust with your writer. The often very overworked people who are doing this are doing it in good faith. If there does happen to be a bad-faith submission that is largely AI generated, I don’t know that we can ask magazines to never fumble that particular challenge.” Plus placing that onus on editors could put magazines in the unsavory position of policing writers. “I don’t think that is why any of us got in the business,” she said.
It’s also not why most readers read literature. When I asked Dirt Media CEO Daisy Alioto about the effect of that surveillance mind-set on readers, she said that “you don’t have to be an AI booster to think ‘catching people out’ is a poor use of the literary community’s energy. I think a lot of people are scared and disappointed by their literary careers, and the badge of honor around avoiding AI becomes a new kind of balm for that. Just do good work and find ways to fund it that you can live with.”
In time, AI-generated text may become trickier and trickier to identify, morphing and developing new tics and tells along the way. So before writers excise em dashes from their stories and readers smell blood at the sight of a list of three, remember: Literary-magazine editors are convinced that AI isn’t yet the storyteller alarmists might think. And why would it need to be? “We’re sitting inside what perhaps biasedly or optimistically I think of as the furthest-reaching, least-relevant use case for AI,” said Joyland’s Caplan. “There are no deadlines; there’s no one forcing you to submit a story. This is an oasis away from our optimized world. The joy of editing is getting to think slowly alongside and with somebody else, to be with them in this system of meaning they’ve created, to perceive what their story wants to say and to excavate that.”
