A disputed AI flag highlights how familiar literary conventions can invite suspicion—and why detection results alone can’t establish who wrote a story.
The dispute over whether a prizewinning story was written with artificial intelligence points to a problem larger than any single detection result: African writers have long been rewarded for stories that fit familiar expectations, and those same conventions can now make their work seem machine-made.
Zimbabwean writer Simbarashe Steyn Kundizeza won this year’s Caine Prize for African Writing for “The Blue Fighter.” Five days later, novelist and journalist Vauhini Vara reported that the AI detector Pangram had flagged almost the entire story as generated by AI. Kundizeza has denied using AI and said he drew on people and settings from his own life. A detector’s result raises a question; by itself, it doesn’t settle who wrote the story.
When familiar stories invite suspicion
I found “The Blue Fighter” moving and effective in places, yet its plot felt familiar. A boy narrates the story of his father, who leaves a career as a civil engineer to build aquariums and breed exotic fish. As the father spends the family’s dwindling resources on tanks and equipment, the mother works to keep the household afloat. The family moves to a poorer neighborhood; the mother confronts her husband and sells most of the tanks to pay the bills.
Those elements overlap with conventions often found in prizewinning African fiction: a young narrator, an ambitious but impractical father, a capable mother carrying the household, and mounting economic hardship. My own Caine Prize-winning story, “Miracle,” also features a young narrator. Familiarity alone doesn’t show that a writer used a chatbot. But it may help explain why some readers thought Kundizeza’s story sounded artificial: the plot assembles several recognizable features with unusual neatness.
That resemblance matters in part because chatbots learn from large collections of human writing. Those collections include portrayals of Africa shaped by what publishers, magazines, critics and prize committees have chosen to recognize. Ask a chatbot for an “African” story, and it may reproduce the conventions that have circulated and earned attention for years. A machine can imitate literary patterns precisely because people have written and rewarded them.
A market with narrow expectations
Ten years ago, I joined a 12-day writing retreat in Lusaka, Zambia, with writers who included four former Caine Prize winners: NoViolet Bulawayo, Okwiri Oduor, Namwali Serpell and me. We wrote, workshopped stories, visited local schools and read our work to the public. We also talked about the distance between the literature we wanted to make and the stories the market seemed ready to accept.
The prize, first awarded in 2000, was created to bring African writing to a wider international audience. As African fiction gained visibility, many celebrated works repeatedly returned to war and its aftermath, poverty, violence against queer people, migration to the West, or familiar American and European stories retold in African settings. At the retreat, we wondered whether stories that followed less predictable paths—and challenged narrow ideas about what “Africa” means—could receive the same recognition.
Writers can pursue original work while recognizing that some accessible stories are more readily accepted. That creates an uncomfortable bind. The literary world has often rewarded recognizable conventions; now, those conventions can be read as evidence that a writer may not have produced the work independently.
Detection is not a verdict
Debate over the Caine Prize story has touched on AI’s writing ability, the state of contemporary literature and racial bias. Earlier this year, three regional winners of the Commonwealth Short Story Prize were accused of using AI. Those allegations led to wider claims that chatbots might fool prize judges. Others worried that if machine-generated stories could win, contemporary fiction had become predictable. In parallel, some readers questioned whether AI detectors were being used unfairly against Black and brown authors.
There is reason to be cautious about drawing conclusions from a flag. After investigating allegations against several Commonwealth Prize regional winners, the Commonwealth Foundation reviewed drafts, notes and time-stamped documents and concluded that AI had not been used to write their stories. In a recent essay, Cambridge professor Doyle Calhoun observed that the five major literary AI controversies he examined that year all involved Black and brown writers. The Kundizeza accusation adds another case to that pattern.
Questions about AI assistance also lack an agreed answer. Word-processing tools now prompt users to rewrite sentences. A writer might accept a suggested edit, ask a chatbot to polish a completed story, or use other kinds of help. Established writers often publish work shaped by several human editors, while a single author receives the byline. Where assistance ends and authorship begins remains unsettled, and accusations of cheating are difficult to assess without clear standards and evidence about how a work was made.
What the controversy leaves writers facing
Black and African writers are being asked to meet the expectations of the literary market and to prove that their work is distinct from machines trained on that market’s own conventions. That pressure can make a familiar story suspicious, even when it draws on a writer’s lived experience. The problem isn’t solved by assuming either that a detector is right or that every accusation is racially motivated. A fair assessment needs more than a software score, and the writing world needs clearer ways to discuss different forms of assistance.
At the Lusaka retreat’s close, many of us reached a conclusion that felt impractical then and feels urgent now: writers had to set aside market signals and write what they wanted. That doesn’t resolve how contests should assess AI use. It does offer a useful reminder that literary originality can’t be measured by how far a story departs from a machine’s imitation of familiar forms.
