A surrealism pencil sketch of a human eye and an ocular instrument superimposed onto a landscape and castle-like structure

From Oculus Artificalis (1685) / Johann Zahn / Public Domain


As far as literary scandals go, the Trinidadian author Jamir Nazir may have triggered the most explosive one of 2026. Winning the Commonwealth Short Story competition with a piece that has been widely accused of being AI-generated, Nazir ostensibly pulled off a coup that unsettled the literary establishment. 

While most accusations about “The Serpent in the Grove” recruited sentence-level clichés as evidence, Katy Waldman’s New Yorker essay expanded the concerns of AI plagiarism to a more complex level of representation—that of the events, narrative structure, and genre. Waldman observed that Nazir’s story “embod[ies] the mustiest cliches of postcolonial melodrama” (emphasis mine). She speculated that the author perhaps prompted an LLM to generate “post-colonial fiction worthy of a literary prize” triggering the bot to produce “an off-key imitation of the genre’s worst tropes,” such as “a pathetic husband driven to violence, a maternal village woman, a subjugated young bride.” 

Waldman’s classification of “The Serpent in the Grove” as “postcolonial melodrama” is noteworthy. Genre labels have an epistemological sway over the literary marketplace, and the Western literary establishment often equates postcolonial and queer writing with melodrama. A deep dive into America’s book reviews culture reveals that few BIPOC and queer writers who have departed from a minimalist style of writing are able to dodge the charge of writing melodramas. Minimalism, celebrated as the gold standard of literary style in America since at least the 1980s, typically features “seemingly dispassionate narrators who experience disconcerting distance from even the most consequential of events,” writes Karen Salyer McElmurray in her essay “Minimalism and Maximalism: A Question of Style.” However, as South Asian American author Bharati Mukherjee argues in her 1988 essay for The New York Times, “Minimalism aims for recognitions, naming the dread, invoking the villains and the victims, but avoiding the confrontation.” Such a style presumes a readership that is either familiar with the storyteller’s culture or one that will put in the labor to cultivate familiarity. Writers from underrepresented and marginalized communities can’t take for granted such a readership. But if they resort to maximalism, an aesthetic mode characterized by “vastness—of space, time, vision” in McElmurray’s words, and present history, interiority, and collective experiences through “the profusion of language on the page,” they risk being scorned as melodramatic. 

By such a reductive logic, to write from and about historically marginalized communities with urgency and vulnerability, to confront desire, shame, and trauma without the facade of detachment, is to be melodramatic. So, if AI interpreted a prompt to produce prize-worthy postcolonial fiction as an invitation to generate the kind of maximalist prose that is derided as melodrama, it was simply operationalizing the classificatory and analytic tendencies of the American (and more broadly, the Global North’s) literary culture. However, my own research on AI-generated narratives makes me doubt if AI can indeed mimic postcolonial voice and positionality and configure melodramatic excesses. My contention is that LLMs reproduce a homogenous and emotionally flat prose without exception and sprinkle it with a handful of easily identifiable cultural markers to construct marginalized identities (the character names of the protagonists in Nazir’s story, Sita and Vishnu, are names of Hindu gods and their avatars, for instance). Such cultural scaffolding in AI prose can give the impression that it is channelling distinct genre tendencies while merely reproducing the tropes that pigeonhole and exclude real BIPOC authors. 

For two and a half years now, my students—Blair Rubke and Jessie Pons—and I have been investigating the emotional and ideological structure of stories about migrants and refugees generated by two popular LLMs, ChatGPT and Gemini. I designed the study to chart the impact of AI’s narrative capacities on migration-related discourses as well as the lived experiences of migrants. After reading more than a thousand LLM-generated stories, I can say that these stories fail to meet the basic criteria of melodrama. While melodrama as a genre eludes neat definitions, the term is applied to art that excessively attends to feelings instigated by human suffering. AI-generated fiction minimizes references to both humans and suffering. LLM stories are infused with a positivity bias, end on a moralizing note, and avoid representing suffering. The ChatGPT- and Gemini-generated stories in our corpora exaggerated positive emotions (happiness) and circumvented the mention of negative ones (sadness, anger), even when our prompt explicitly asked them to depict these emotions. 

Some cultural biases in LLM outputs can be attributed to their training data, but the suppression of anger, sadness, and violence likely result from the parameters and guardrails engineered into them to ensure they produce innocuous responses acceptable to a wide range of users. Rapid adoption of LLMs in a variety of sectors from education to healthcare owe something to the fact that LLMs steer clear of high arousal content. This is not to say that a story about patriarchal violence such as Nazir’s cannot be generated using LLMs; rather that LLMs will flatten grief, fury, and murderous jealousy into a sanitized report on those emotions, producing a quieter and more emotionally regulated text that superficially aligns with the craft lessons of minimalism. 

Sharma Taylor, one of the judges for the 2026 Commonwealth short story competition, praised Nazir’s story for its “restraint and quiet authority.” In other words, “The Serpent” triumphed in the contest for successfully mimicking dispassionate and disaffected storytelling. It was rewarded by a literary culture that believes linguistic stunts performed high in the air, without contact with embodied emotions, to be a sign of literary merit. Take the case of the expository sentences, “Sun on galvanise is a cruel instrument. It beats until the roof talks back in a dry moan.” The latter sentence borders on the nonsensical. The hurt and pain characters would feel living through the situations alluded to in the story are expunged from the surface of the prose and offloaded onto obtuse metaphors, daring readers to locate them. 

“The Serpent in the Grove” is thus an emotionally stunted text, where surplus linguistic tricks compensate for the absence of rage and desire. I use the word surplus with intention here—AI writing is characterized by surplus. It is often unnecessarily formal, unnecessarily sycophantic, unnecessarily figurative. But surplus is not the same as maximalist or melodramatic. If maximal style uses profuse details for aesthetic and emotional effect, surplus designates what we could do without in a text. AI outputs, for example, are often bookended by metacommentary summarizing the same output. The short sentences arranged into stanzas that close Nazir’s story use far-fetched analogies to define what a story is in terms of the objects and events depicted in the story. Such unwarranted summative moves appear at the sentence level too. In “She wore her role without protest and without light; both things can be true”—“both things can be true” is a redundant expression. Redundancy is not dramatic, let alone melodramatic. But LLMs tend to generate inane linguistic surplus to overfit the users’ prompts, especially when the prompt tries to elicit depictions of suffering or anger. 

During one of the phases of our experiment in fall 2025, I prompted ChatGPT through API calls (which sidesteps the chatbot interface) to produce stories about immigrants that would anger readers in two languages—English and Bengali, without specifying the cultural and political profile of the implied reader. It produced stories containing several weakly connected episodes, narratives that progressed through correlation rather than causation. In the English-language stories, immigrant characters were constantly exploiting the system (speaking rudely at town halls, faking degrees to take jobs away from the native population, harming local businesses) while in the Bengali-language stories immigrant characters were depicted as being constantly disadvantaged by circumstances and exploited by wicked middlemen (e.g.: coyotes and masters forcing them to work in abusive conditions). Frequent references to prototypical transgressive conduct of either the immigrant or a character representing the host population sought to make up for the emotional vacuousness of these stories. While analysing such texts, I adopted the term “overnarration,” originally coined by the narratologist Gerald Prince, to describe ChatGPT’s tendency to generate similar but causally disconnected events over and over in order to tackle misalignments between its safety tuning, which is responsible for its positivity bias, and the demands of my prompt. 

If my students’ and my empirical work establish that LLMs can’t adequately narrate suffering, Claudia Carroll and Gabi Kirilloff’s research shows that these models struggle to write about humans altogether. Comparing stylistic features of human-written passages of fiction with GPT-written ones, Carroll and Kirilloff observe that synthetic prose does not center character:”AI-generated fiction is largely de-personified, and contains far more descriptions of inanimate objects, unrelated to characters, than the human-written text.” The Commonwealth story predictably assigns greater agency to things than to people. There’s “a slope that wanted either rain in teeth or none at all.” Whereas Sita, the oppressed wife in the story, stays silent, “wood complained in a voice too near speech.” So much of “The Serpent” is aperspectival writing—not omniscient or revolving third-person narration, but narration lacking perspective and precision. It channels a free-floating voice coming from nowhere. 

Creativity as per the logics undergirding AI writing is detached from human experience and subjectivity. It is simply construed as linguistic expressions that either don’t appear or appear less frequently in existing writing metabolized by the LLM. This accounts for some improbable analogies in Nazir’s story, such as, “she had the kind of walking that made benches become men.” Yet the ludicrous analogies give the illusion of something elusive lurking under the surface, parodying what Ernest Hemingway calls the “iceberg theory” of fiction in Death in the Afternoon. It tricks the reader—the Commonwealth judges, for example—trained to sight and sail toward icebergs into assuming that meaning must be submerged somewhere deep down under the improbable metaphors and similes. 

Despite “The Serpent” ’s mimicry of minimalism, the fact that Waldman labels it as “postcolonial melodrama” betrays a serious cultural bias, one that associates a story attributed to a writer from the Global South and set in the Global South with melodrama irrespective of the fact that the prose skirts around human suffering instead of excessively dwelling on it. The equation drawn between maximalism and melodrama in the context of BIPOC fiction or fiction from the Global South constitutes a problematic aesthetic judgement that predates AI. But if critics and readers now mix up these aesthetic modes with AI’s overnarration, the problem is exacerbated. This could compel BIPOC, postcolonial, and queer writers to further avoid maximalism to shirk the double charges of AI plagiarism and melodrama. 

When readers and critics assume the role of detectives hunting for AI’s fingerprints in texts, they do so to monitor and call out artistic fraud. However, in the process, AI’s proclivities start to set the borders of art. If “not…but” statements, hyphen, or any other linguistic mannerism comes to be regarded as sure signs of AI use, omitting them would be the easiest way out to avert accusations of AI-plagiarism. The same holds true for genre and genre labels. 

Contemporary authors face the unwieldy task of satisfying an audience whose expectations of linguistic style and clarity are being molded by the surplus AI texts flooding all sorts of platforms, from news briefings to social media, while also having to prove to the same audience that their writing was not AI-generated. The specter of AI now haunts all writers in this sense. But it doubly burdens artists from historically marginalized communities who must strive to gain credibility in a literary culture that already sidelines them by dodging the narrative tropes and techniques colonized by AI.