AI Story Flood Forces a Science Fiction Magazine to Close Submissions

Clarkesworld closed its submission portal after receiving more than 500 submissions it flagged as spam in February. Editor Neil Clarke says the surge is crowding out time for emerging writers, while unreliable detection tools risk mislabeling some authors.

WTF Index IDIOCRACY
◄ Terminator 1 Idiocracy 4 ►

AI-generated submission floods waste editors’ time and crowd out developing writers, though unreliable detection also risks mislabeling authors.

AI Story Flood Forces a Science Fiction Magazine to Close Submissions

Clarkesworld, a science fiction and fantasy magazine that had kept its submissions open for most of its 17-year history, closed its portal after a sharp rise in spam. Editor and publisher Neil Clarke said more than 500 submissions were flagged as spam in February alone, making it harder for staff to focus on writers seeking to improve their work.

A sudden change in the submission flow

For Clarkesworld, the closure was unusual. The magazine had previously shut submissions only briefly for website or software upgrades, and Clarke described this decision as a spur-of-the-moment response to the volume arriving.

The scale of the change stands out against the magazine’s earlier experience. From June 2019 through the beginning of this year, its monthly spam count never exceeded 25, and many months had none. In 2022, Clarkesworld received around 11,500 submissions in total.

Clarke said generative tools can produce far more material than a person could write unaided in the same span of time. That capacity changes the editorial workload: the challenge is no longer simply reading a large number of stories, but finding serious submissions among a flood of low-effort entries.

Why the volume matters to writers

Each submission takes attention to review. When many entries are considered spam, staff have less time to read and respond to writers who are genuinely trying to develop their craft. Clarke said the deluge was not coming from bots, in his view, but from real people hoping to make quick money with minimal effort.

He connected some of the traffic to side-hustle blogs, YouTube channels and TikToks that promote sending AI-generated work to publishers. Clarke said Clarkesworld and another magazine could trace submissions to shared lists promoted by chatbot gurus. The publications were receiving entries from the same person, with the second magazine getting them a few minutes after Clarkesworld.

Other science fiction publications have faced similar problems, particularly those that accept submissions continuously or pay higher rates. The source article notes that a publication must pay at least eight cents per word to qualify as a “professional” short fiction market by the Science Fiction & Fantasy Writers Association. That may make such markets attractive to people seeking quick returns; other literary fiction and poetry markets can pay less or nothing.

Detection tools create another problem

Publishers might hope software can separate AI-generated writing from human work, but Clarke said costly detection programs he had encountered were not accurate enough. The source article also describes a test of seven AI text detectors, none of which was particularly effective, and says OpenAI estimates its own synthetic-text detector misses 74% of AI-generated text.

False flags could hurt writers as well as fail to catch generated submissions. Clarke has a particular interest in publishing international authors, and he noticed that writers who are not native English speakers were sometimes flagged as spam by detectors. A language pattern that looks unusual to a tool may reflect how someone learned English, rather than the use of AI.

That makes automated screening a difficult shortcut. A publisher needs to reduce the burden of spam without filtering out the writers it wants to discover. Clarkesworld’s experience shows how a tool intended to solve one problem can introduce another when decisions depend on imperfect signals.

An unsettled issue for literary publishing

The stories themselves raise questions about originality and plagiarism because systems such as ChatGPT and GPT-3 were trained on text from the internet, including copyrighted stories from publications like Clarkesworld. If an AI-generated story were selected for publication, the publisher would face ethical questions about that work. Clarke said the submissions reaching his magazine were not good enough to create that scenario so far.

More broadly, the magazine’s closure has drawn attention beyond the science fiction community and into debate over whether AI can make good art. Clarke said he was less interested in taking a side in that argument than in figuring out whether a submission came from someone serious about writing.

Clarkesworld had not announced when submissions would reopen. Clarke was exploring ways to deter spam while the industry worked through a new and unclear problem. For authors, his advice was straightforward: “if you think it might be suspicious, don’t do it.”