Something felt wrong with this year’s Commonwealth Short Story Prize winners. Since 2012 the British magazine Granta has published these regional picks. One story stood out. Jamir Nazir’s “The Serpent in the Grove” screamed AI.
Not just vibes. Technical markers. Mixed metaphors. Anaphora. Lists of threes.
I know that sounds suspicious. I just listed three things. I promise I typed these words myself. No robots helping.
I used to side-eye these AI paranoia waves. People swear em dashes are dead giveaways. They hate the word “delve.” They distrust short sentences that follow long ones.
Humans do those things too.
LLMs learn from us. They mirror what we feed them. If they use em dashes and lists, it is because we gave them to those patterns.
But AI writing still feels off. Uncanny. Even when you cannot name exactly what is wrong.
Nabeel S. Quresh noticed it first. He is a former visiting scholar at George Mason. For him, the opening lines were enough proof.
They say the grove still hum at noon. Not the bee’s industry or the cutlass on the vibe but a belly sound as if the earth swallowed a shout.
“In general AI writing has a particular rhythm… there’s a spectrum from AI helped me edit to AI wrote this… this case reads like the latter end.”
He suspects full generation. He isn’t sure though. Nobody really is.
Razmi Farook leads the Commonwealth Foundation. His statement relies on trust. Authors swore their work was original. No AI claimed to be the writer.
Farook admits the industry lacks a good detection tool for unpublished fiction. So they trust the writers. Until better technology arrives.
Granta tried to check.
Publisher Sigrid Rausing said they fed Nazir’s story into Claude. A chatbot. They asked it if it was AI.
Claude replied: “Almost certainly not produced unaided.”
Wait.
Claude isn’t a detector. It is an LLM. Asking a generator if something was generated by a generator is like asking a fish if the water is wet. Granta might not understand what it bought.
“It may be that the judges… have awarded a prize to an instance of plagiarism.”
Plagiarism of what? The dataset?
Publications are getting tricked. Sometimes the “authors” are ghosts. Fakes. People even doubted Jamir Nazir existed. Kevin Jared Hosein confirmed Nazir is real. They chatted about it. Nazir published a poetry book in 2018 too. He declined to comment.
Hachette pulled a horror novel by Mia Ballard. She denied AI use. Blamed a hired editor instead.
So what is allowed?
Idea generation? Research help? AI transcription? Where is the line?
Polish Nobel winner Olga Tokarczuk drew fire. She admitted using AI in her process.
“Darling how could we beautifully elaborate this?”
That is her prompt.
She knows the tech hallucinates facts. She admits it has errors. But in fiction it has leverage. Unbelievable leverage.
She also mourns the old ways. The isolation. The months spent alone with thoughts.
“I am heartbroken by the departure of the traditional literature.”
She misses Balzac. Nabokov. She thinks modern chat cannot match their style.
Tokarczuk later clarified. She does not write books with AI. She uses it for fast documentation and fact checks. Then she verifies the facts herself.
She even gets inspiration from dreams. Just human dreams. Her own.
Her comments sparked outrage. People were paranoid. Publishing hates losing control.
James Daunt runs Barnes and Noble. He said he would sell AI books. Only if labeled clearly. As fake.
He faced a boycott. He walked back. Not entirely.
“Book banning is a danger… we will not sell AI books that masquerade as by real authors.”
So fake AI books are bad. Real AI books with labels are okay? The line moves.
It doesn’t explain the weirdness.
I ran Nazir’s story through Pangram software. 100% AI.
Pangram flagged three things:
1. Triads
2. The word “stubborn” (6x more common in AI text)
3. The phrase “as if it had”
Again. Three items.
I ran my own writing through Pangram. Unpublished draft. Weak prose.
It found triads in my paragraph too. But the software called me human.
I tried again with better excerpts. Same result. Human.
Kevin Nguyen wrote a novel for The Verge. His chapter tested 100% human too.
Pangram found the 2024 and 2023 award winners in Granta were likely AI. But it missed others?
Maybe AI prose is like bad manners. You feel it. You cannot measure it.
Human work has an ineffable quality.
So does its inverse.
