Open the book generator and you are not looking at a tool. You are looking at a curriculum. Chapter templates, genre presets, stylistic suggestions—all of it promising to turn a handful of prompts into a finished manuscript. The interface is clean. The options are legible. The friction is minimal. That is the problem. What this generator actually delivers is not neutral infrastructure. It’s a value system with a user interface. A curatorial hand nobody hired, automating a specific vision of what a book should be, what a chapter should do, what an argument should sound like. For art writing—criticism, monographs, exhibition catalogs—this matters enormously. The defaults baked into these tools are already training the next generation of texts. And the defaults are not neutral.
The Unsloppy tool makes a useful case study because it makes explicit what most AI writing tools obscure. It offers structure. It offers genre. It offers tone. Click through its options and you are selecting from a menu of pre-approved literary forms. The chapter templates don’t emerge from the ether. They’re trained on a corpus of existing books—predominantly Anglophone, predominantly commercially successful, predominantly adhering to the narrative arcs that Western publishing markets reward. Ask the tool to generate a monograph on the post-war Polish avant-garde. It won’t consult the archives of Warsaw’s Foksal Gallery. It consults a statistical model of what monographs look like. And what monographs look like, in that model, is a problem.
Consider the genre presets. The tool offers categories like ‘non-fiction,’ ‘memoir,’ ‘how-to,’ ‘academic.’ Each comes with an implicit architecture. A non-fiction preset assumes a problem-solution structure: introduce a crisis, examine its dimensions, propose a resolution. An academic preset assumes a literature review, a methodology section, a findings chapter, a discussion. These are not natural forms. They are conventions. And conventions are arguments about what knowledge looks like. When an AI tool automates them, it doesn’t just save labor. It naturalizes those conventions. It makes them feel inevitable. The young curator using a book generator to draft an exhibition catalog may not realize she’s importing the epistemological assumptions of a mid-tier American university press. But she is.
The stylistic suggestions are worse. Ask for a ‘scholarly’ tone. You get sentences thick with passive voice, nominalizations, the hedging that tenure-track academics mistake for rigor. Ask for an ‘accessible’ tone. Complexity gets flattened into short declarative sentences that sound like a TED talk transcript. Neither mode serves art criticism. The scholarly mode erases the critic’s voice—the specific sensibility that makes a T.J. Clark or a Rosalind Krauss worth reading. The accessible mode erases the difficulty of the art itself, pretending a Laura Owens painting or a Hito Steyerl video essay can be digested like a business book. What’s missing from both is friction. Unresolved tension. The sense that the writer is thinking through the work rather than summarizing it.
This isn’t a Luddite complaint. The technology is impressive. The problem isn’t that AI can generate text. It’s that the text it generates carries a hidden curriculum. And that curriculum is aggressively Anglophone. The narrative arcs embedded in these tools—three-act structures, hero’s journeys, problem-solution frameworks—are not universal. They are culturally specific. They emerged from Aristotle via Hollywood via the American creative writing workshop. When a book generator applies them to an exhibition catalog for a Southeast Asian contemporary art survey, it does violence to the material. It forces a polycentric, non-linear, often deliberately fragmented artistic tradition into a narrative shape designed for linear, individualist, resolution-driven stories. The result is a synthetic ‘international style’ of art-historical reasoning. It sounds plausible. It reads as professional. It erases the very regional specificity that makes the art worth examining.
Be concrete. The chapter templates in the Unsloppy tool include options like ‘Introduction: Setting the Stage,’ ‘The Problem,’ ‘The Solution,’ ‘Conclusion: Looking Forward.’ These are not benign labels. They encode a teleological view of history. They assume art movements progress toward solutions. That exhibitions resolve problems. That curatorial projects have tidy endings. But the most important art writing of the last century—Walter Benjamin’s Arcades Project, Okwui Enwezor’s Documenta 11 catalog—refuses this structure. It accumulates. It juxtaposes. It leaves gaps. It trusts the reader to navigate complexity without a tour guide. The AI template cannot do this because its training data rewards coherence. And coherence, in art writing, is often the enemy of truth.
The market assumptions are equally troubling. The genre presets are designed to produce books that sell—or at least books that resemble books that have sold. This means privileging certain subjects, certain artists, certain historical narratives. A monograph on Gerhard Richter will generate smoothly. The corpus is vast. A monograph on the Ethiopian modernist Gebre Kristos Desta will struggle. The source material is sparse. The existing English-language literature doesn’t conform to the template. The tool doesn’t announce this bias. It simply underperforms, producing a text that reads as thin or generic. The user, unaware of the mechanism, may conclude that Desta is not ‘book-worthy.’ The bias is laundered through the interface.
The implications for art criticism are dire. We’re already living through a crisis of art writing. Museum publications have become luxury objects—heavy, glossy, unread. Exhibition catalogs are increasingly written by the same handful of approved authors trading in the same references (Deleuze, Haraway, Benjamin—always Benjamin) and producing the same cadences. Art magazines have abandoned the long-form negative review because it threatens advertising revenue. Into this vacuum steps the AI book generator, offering speed, consistency, an end to writer’s block. The result will be a flood of texts that are competent, readable, and utterly interchangeable. The machine will not produce the next Ways of Seeing. It will produce the next 500 museum shop paperbacks that no one reads but everyone cites.
What can be done? The Authors Guild has published AI Best Practices for Authors that emphasize transparency and the preservation of authorial voice. Their guidance is a start. They recognize—rightly—that AI isn’t going away and that the fight is over the terms of its use, not its existence. But transparency alone is insufficient. We need a critical literacy about these tools. Curators, critics, art historians must learn to read the defaults. They must ask: What chapter structure is this tool offering me, and what does that structure assume about my subject? What stylistic register is it defaulting to, and whose voice does that register erase? What corpus was this model trained on, and what traditions does that corpus exclude?
The Purdue OWL’s Creative Writing Introduction reminds us that creative writing instruction has long debated the value of formal constraints. The workshop model, with its emphasis on conflict, character arc, and resolution, has been criticized for producing homogenous fiction. The same critique applies, magnified, to AI-generated structures. The difference: a workshop is a social space where norms can be challenged. An AI tool is a black box that presents its norms as features. When you select ‘Academic Tone’ from a dropdown menu, you’re not making a stylistic choice. You’re accepting a definition of academic writing that has been statistically derived from a narrow slice of English-language scholarship. That definition excludes the essayistic tradition of John Berger, the aphoristic density of Anne Carson, the polemical fury of Dave Hickey. It excludes most of what makes art writing worth reading.
The regional dimension deserves special attention. Art scenes outside the North Atlantic are already forced to translate their practices into a discursive framework not their own. Artists from Jakarta to Lagos to Tbilisi learn to write artist statements that sound like they were drafted in a Chelsea gallery office. Curators from these regions learn to pitch exhibitions using the language of ‘intervention,’ ‘discourse,’ ‘negotiation’ because that’s what international funders and biennale directors expect. The AI book generator accelerates this homogenization. It provides a shortcut to the approved language. It makes it easier than ever to produce a catalog that sounds exactly like every other catalog from the last decade. What it cannot do—what no current AI can do—is capture the specific texture of a local art scene: the gossip, the informal economies, the way an artist’s work responds to a particular building or a particular political moment. The machine trades in generalities. Art lives in specifics.
There’s a deeper irony here. The art world prides itself on challenging norms, subverting expectations, ‘problematizing’ everything. Yet the tools it is adopting—and it will adopt them, because the economics of publishing are brutal—are norm-enforcement machines. They are conservative by design. They predict the most likely next word, the most probable chapter structure, the most common stylistic register. They are, in a precise sense, anti-experimental. An artist who spends a career resisting easy legibility will find their work processed by a tool that insists on it. The catalog will be clear. It will be structured. It will be dead wrong.
What would a better tool look like? It would expose its own assumptions. It would offer templates that include non-Western narrative structures: the spiral, the fragment, the call-and-response. It would include stylistic registers beyond ‘scholarly’ and ‘accessible’—perhaps ‘polemical,’ ‘elliptical,’ ‘lyrical,’ ‘forensic.’ It would tag its training data so users could see what voices are amplified and which are absent. It would treat its own architecture as a political artifact rather than a neutral convenience. This isn’t a utopian demand. It’s a design challenge. And it’s one that no current AI book generator—Unsloppy included—has even begun to address.
The art world has a choice. It can treat these tools as harmless productivity aids, using them to generate catalog drafts and wall texts without examining their embedded values. Or it can treat them as a new front in the long struggle over who gets to shape art-historical discourse. The first path leads to a monoculture of art writing—smooth, professional, forgettable. The second path demands a critical vigilance that the art world has rarely mustered for technology. It demands that we read the defaults as closely as we read the text they produce. It demands that we ask, every time we open a book generator: What is being automated here, and whose taste is doing the automating?
The answer, right now, is a taste for the middle. A taste for resolution. A taste for the Anglophone narrative arc that flattens the world into a problem and a solution. That taste is not neutral. It is not universal. It’s a cultural artifact as specific and contingent as a Gothic cathedral or a Mughal miniature. And it’s being built into the infrastructure of the next generation of art writing with no public debate, no critical scrutiny, no acknowledgment that anything of value is being lost. The machine in the monograph is not coming. It’s already here, humming quietly in the cloud, waiting for a prompt. The question is whether we’ll notice what it’s doing before it’s too late to do anything about it.