Op-Ed: Should We Care If African Writers Are Using AI?

I spoke recently with the Nigerian writer and pianist Echezonachukwu Nduka about his latest poetry collection, “Jazz Negotiations.” As the title suggests, it’s about blue notes, Louis Armstrong, B.B. King, and all that jazz. He wrote it by hand, as he does his compositions, and keeps the autographs stowed away in his apartment in leafy Pennsylvania. He tells me it’s a hedge against the future. “You never know in this era of AI allegations,” he says, alluding to the collective paranoia that has defined the African literary scene in recent months. That paranoiac hypervigilance found a new peak in October, after an Atlantic article alleged that “The Blue Fighter,” the Caine Prize-winning short story by Zimbabwean writer Simbarashe Steyn Kundizeza, was likely written with generative AI. 

Kundizeza is one of several Black and African writers caught up recently in a whirl of accusations that has thrown literary reputations into question, scuttled multimillion-dollar book deals, challenged the integrity of African literary institutions and called attention to the danger that the AI panic poses to Black and African literature. Writers Mia Ballard, H.M. Wolfe, three of the five regional winners of the 2026 Commonwealth Short Story Prize, Thélyson Orélien and Nigerian novelist Jerry Falade have all bled through allegations this year. Falade perhaps bled the most in terms of U.S. dollars; after being suspected of writing his debut crime novel with AI, he would lose a publishing deal worth about $2 million.

Inquiries were launched, such as by the Caine Prize and the Commonwealth Foundation, even as Kundizeza and the other writers denied the allegations. Against this backdrop of investigations, debates attempted to parse the implications for AI and African literature. For some, like Gresham College professor Daniel Susskind, the turn of events shows AI can now produce prize-winning fiction. Others, like Nigerian poet Romeo Oriogun and Ugandan writer Ben Anena, saw a racist agenda at play. After American novelist Joyce Carol Oates threw her National Book Award-winning weight behind the Atlantic piece, Anena decried the “systematic attempt at unraveling the African literary infrastructure.” But for Vauhini Vara, who wrote the Atlantic piece, the controversy comes with a teachable moment for Luddite literary judges, whom she implores to acquaint themselves with “AI output” in order to excel at sniffing out AI-generated writings.

What these conversations do not address, however, is the question of whether or not it is even necessary to try to determine if literary works were produced with the help of AI. Barring the rare case of a writer forgetting to remove AI prompts from published drafts — as was the case of romantasy authors K.C. Crowne and Lena McDonald earlier this year — it is almost impossible to definitively prove AI use. Given the potential harm in trying to do so amid this air of uncertainty, it seems best to do away all together with the need to sleuth out writings for AI clues. 

AI writing is often identified by its reliance on certain quirks. Much has been said about the bias that LLMs have for em dashes, tricolons, negative parallelism (“It’s not X, it’s Y”), and clichés (the Atlantic piece cites Kundizeza’s use of “stereotypes about African life” as evidence of AI use). However, these so-called AI shibboleths are equally present in human writings — human writings with which AI was trained — and so do not definitively prove AI use. Not every errant metaphor or contrastive phrasing is evidence of AI-assisted prose; they could very well be symptoms of poor writing habits. Also, consider that Shakespeare used negative parallelism in “Julius Caesar,” as has anyone who has called it quits with a beau: “Baby, it’s not you, it’s me.” Em dashes serve as musical notations in Emily Dickinson’s poems. And Joseph Conrad’s “Heart of Darkness” traffics in dehumanizing stereotypes about the dwellers of 19th-century Congo, and yet we know Conrad had access to neither Grok nor the Great Automatic Grammatizator, a fiction-spitting machine from a 1953 short story by British author Roald Dahl.

Human-centered approaches to proving AI use can also be derailed by cultural bias. In 2024, during the early days of the AI witchhunts, British-American author Paul Graham famously branded the word “delve” as a reliable AI tell, the rationale being that it is frequently used by ChatGPT while uncommon in American English. However, as many rightly pointed out, Graham failed to account for linguistic environments where the word is commonly used, like in the Nigerian corporate world of blazers and starched shirts, where line managers routinely ask colleagues to “delve deeply” into an issue. To go by one Guardian report, convictions like Graham’s may result from the cold-blooded capitalist extractions happening across the techscape. Not only are Africans exploited for cheap labor by tech companies in training LLMs, they also end up shaping AI English, thus creating an unfortunate situation where various African Englishes become synonymous with AI-speak. 

Similarly, AI detectors are fallible. They can be biased against non-native English speakers, while their use can violate authorial consent (without the expressed permission of the respective writers, Vara ran all Caine Prize-winning stories through the AI detector Pangram). In July, the company behind Pangram claimed that the program generates false positives only once in about 24,000 cases, and though the numbers are impressive, there remains the sliver of chance that a writer can be wrongly accused, can have her life and career turned inside out over a computer glitch.

The possibility of ruinous false allegations, as well as the general climate of paranoia, can be removed entirely if the literary world, if publishers, literary judges and readers focused instead on what is in front of them, what can be more definitively proven: the quality of a piece of writing. It may be hard to know for sure whether or not a short story was written with AI, but it can be more definitively ascertained — and without the risk of causing undue distress to writers and literary institutions — whether or not a story’s use of negative parallelism or tricolons is effective.

This may seem like a call to downplay literary fraud or to ignore the implications that AI has for literary ethics, but the very opposite is true. Because a significant use of AI tends to result only in serviceable writing, prioritizing literary quality will also have the effect of filtering out a great deal of AI-generated texts, all without subjecting writers to harm or perpetuating the kind of anxiety-stricken culture where autographs become necessary in proving the humanity behind literary outputs. The Sudanese writer Lina Aboshouk makes a similar case in a compelling essay for Africa Is a Country, arguing that had the Commonwealth Foundation screened exclusively for quality rather than fed into racialized expectations of literature from the Global South, the controversy may have been altogether avoided.

On the surface, the inquiries into the AI allegations are an earnest effort to administer literary justice. But there is little suggesting that these investigations are made on the back of a robust evaluation of the role of AI in literary practice. Prior to the allegations, a search for AI policy on the websites of many African institutions — from the Caine Prize to the Evaristo Prize — yielded nothing. The investigations thus appear partly driven by unexamined time-old assumptions about literary authenticity. The assumption that every kind of AI use constitutes a form of plagiarism, for instance, is complicated by the fact that literary tradition has long allowed for creative collaborations, nowhere more profound than in the relationship between writers and editors. Sometimes, editors leave such palpable imprints that it is hard to tell where a writer’s ideas end and where an editor’s begin. Flustered over his editor Gordon Lish’s significant rewriting of his short story collection, Raymond Carver would write a protest letter in the summer of 1980. Does it amount to plagiarism when Lish put his “masterly gifts,” as the American writer Cynthia Ozick described it, in the service of Carver?

Would it amount to a literary scam if Carver, weary of Lish’s influence, sought editorial counsel instead from ChatGPT? Would it be inauthentic if African writers asked Claude for suitable synonyms rather than sift through a thesaurus for the mot juste? To what degree is AI use permissible in the creative process? If these questions meet puzzling looks and head scratches, it is because the role of AI in literary originality is yet to be robustly articulated in the African literaryscape. 

That’s likely to change, though, as the recent string of controversies has made it harder to ignore the AI elephant in the room. Just days ago, the Caine Prize published its AI use policy, which states that “content generated by AI, even if later edited or rewritten, is not eligible for submission.” More African literary institutions are expected to follow suit. One can only hope that as they do, as they confront the issue of AI use, it is always with respect to the potential harm facing writers, and by thoughtfully engaging with long-standing literary assumptions. As one of Nduka’s poems goes, one can only hope that the African literary world negotiates the “change in temperature, reworking roofs and walls, ears plugged with new top songs.”