She sat in the corner of a café on Oldham Street—the one where the Wi-Fi holds steady and the staff don’t mind you nursing a single coffee for two hours. The catalogue essay was due in three days. The fee, when it eventually landed, would cover roughly half her rent. The research, the studio visit, the drafting, the revision—if she calculated the hours against the payment, she was already working for less than the Manchester living wage. So she opened a browser tab, found an AI story generator that promised coherent drafts from bullet points, and began feeding it her notes. Not because she believed the machine could write better than she could. Because she could not afford the time to write it herself.
This is not a hypothetical. I have watched versions of this scene unfold across kitchens, studio corners, and library desks in Manchester, Leeds, Birmingham, and Glasgow. It is the logical endpoint of a decade of compressed fees, vanished staff posts, and the quiet expectation that art writing is something you do out of love—which is to say, for nothing. The arrival of AI narrative tools does not create the crisis in art writing. It reveals it, accelerates it, and gives it a friendly interface.
The tool she used that morning belongs to a growing class of products marketed to writers, marketers, and content producers. The pitch is seductive: generate coherent narrative from bullet points, overcome the blank page, produce drafts in seconds. The framing is always one of empowerment—the writer freed from drudgery to focus on higher-order creative decisions. But the framing erases the economic compulsion underneath. Nobody turns to an AI story generator because they are bored of writing. They turn to it because the material conditions of their work have made writing unsustainable.
The Hidden Labour of the Prompt
There is a phrase now circulating in tech and publishing circles: prompt engineering. It sounds technical, precise, a skill to be listed on a CV. In practice, it describes the work of coaxing a large language model into producing usable prose—feeding it context, adjusting parameters, iterating through outputs, discarding the gibberish, refining the instructions, and then heavily editing what remains. This labour is real. It takes time, judgment, and a working knowledge of the subject matter. And it is almost entirely invisible.
When a freelance art writer uses an AI tool to generate a draft, the client sees a finished text delivered on deadline. They do not see the hour spent crafting the prompt, the fifteen discarded outputs, the careful excision of hallucinated facts about artists who never existed, the restructuring of paragraphs that the machine assembled in the wrong order. The writer’s labour has not disappeared. It has been displaced into a preparatory, editorial, and corrective role—one that is even harder to invoice for than the writing itself. The fee remains the same. The time may even increase. But the visible product arrives faster, and the illusion of efficiency is maintained.
This is piecework by another name. The term belongs to the nineteenth-century textile industry—workers paid per unit produced, not per hour worked, bearing all the risk of speed, quality, and tool maintenance themselves. The AI prompt is the new piece: a unit of linguistic production whose value is determined by the market, not by the time it took to make. And like the weavers in Manchester’s mills, the prompt engineer works with tools she does not own, on platforms that extract value from her activity while setting the terms of her access.
Who Can Afford to Write Without Assistance?
The class dimension here is unavoidable. To write without AI assistance—to draft, redraft, research, and revise entirely through one’s own cognitive and temporal resources—is increasingly a luxury. It requires time. Time requires money. Money, in art writing, requires either a salaried post (vanishingly rare outside London), a partner with a stable income, family wealth, or a portfolio of better-paying work that subsidises the criticism. The independent art writer who refuses AI on principle is, in many cases, the writer who can afford to refuse.
This is the argument the Authors Guild makes, with careful precision, in its AI Best Practices for Authors. The document warns that quality human writing risks becoming “a rare luxury good representing only a minority of views.” It is a warning about literature and journalism, but it applies with particular force to art criticism, a field already squeezed between academic precarity and commercial indifference. If the only people who can afford to write without machine assistance are those with independent means, then the criticism that survives will represent a narrowing band of class experience. The working-class art writer, the regional critic, the parent juggling childcare and deadlines—these are the voices most likely to be pushed toward the AI prompt, and most likely to be penalised for using it when the ethical reckoning arrives.
The Authors Guild also notes that the foundational large language models powering these tools “have been trained on pirated, unlicensed books without compensating authors or publishers.” The parallel to art writing is sharp. The AI story generator that produces a plausible catalogue essay was trained on a corpus that almost certainly includes art criticism, exhibition reviews, and artist statements—texts written by people who were never asked, never credited, and never paid. The tool that promises to solve the labour crisis of art writing is built on the uncompensated labour of art writers.
The Regional Ecology Under Pressure
Art writing outside London has always been fragile. It depends on a thin network of independent magazines, gallery publications, artist monographs, and the occasional newspaper column. There is no regional equivalent of the salaried critic posts that still exist, in diminished form, at a handful of national broadsheets. The critic in Manchester or Glasgow is typically freelance, piecing together income from catalogue essays, reviews, grant-writing, teaching, and sometimes entirely unrelated work. The fee for a 1,500-word catalogue essay might be £300 if the gallery is well-funded, £150 if it is not, and nothing at all if the invitation is framed as an “opportunity for exposure.”
Into this ecology, AI narrative tools arrive not as a creative experiment but as an economic pressure valve. When the fee is £150 and the essay requires a studio visit, reading the artist’s previous statements, looking at the work, drafting, and revising, the hourly rate falls below minimum wage. The AI draft offers a way to compress the most time-intensive phase—the generation of initial prose—into minutes. The writer still does the thinking, the looking, the structuring, and the polishing. But the part of the process that feels most like “writing” is outsourced to a machine trained on stolen text. The ethical discomfort is real. The economic necessity is realer.
This is not a dilemma that London-based critics face in the same way. A critic with a staff job at a national newspaper, or a regular column in a well-funded art monthly, has the time to write. They may experiment with AI out of curiosity, or refuse it out of principle, but their refusal does not cost them rent. The regional critic’s refusal might. The geography of AI adoption in art writing will map, with grim precision, onto the geography of economic precarity.
The Marketing of ‘Story Generation’
The companies selling AI writing tools do not describe their products as economic necessities for underpaid workers. They describe them as creative aids, imagination sparkers, solutions to the “agony” of the blank page. The language is playful, aspirational, gently self-deprecating. Reedsy’s Book Title Generator, for instance, offers to help writers overcome the difficulty of naming their manuscript, noting that Fitzgerald cycled through a dozen titles before landing on The Great Gatsby. The tool is calibrated by genre, tone, and conflict; it returns ten options with explanatory notes. It is, in its own terms, a spark for the imagination.
But the framing deserves scrutiny. The generator is not a neutral creativity tool. It is an AI-powered system that produces title suggestions by drawing on a corpus of existing literature. The user is instructed to describe their core conflict, add comparative titles, and select a mode—commercial or literary. This is prompt engineering, repackaged as a friendly creative exercise. The labour of title-generation, once part of the writer’s craft, is now a service provided by a platform that benefits from the data users feed into it. The writer who uses the tool is not just receiving assistance; they are training the system that may one day replace the need to hire them at all.
The same logic applies to the AI story generators marketed to art writers. The pitch is always about overcoming friction: writer’s block, time pressure, the difficulty of structuring a complex argument. The subtext—never stated, always present—is that writing is a problem to be solved, and the solution is automation. But writing is not a problem. It is work. The problem is that the work is not paid.
What the Machine Cannot Do
There are things an AI story generator can do. It can produce grammatically correct sentences. It can organise information into a recognisable structure. It can mimic the tone of an exhibition review well enough to pass a casual glance. What it cannot do is look at an artwork and feel something specific. It cannot sit in a gallery in Stockport and notice that the lighting makes the paintings look apologetic. It cannot know that the artist it is writing about once worked as an invigilator at the same institution now showing her work, and that this fact matters to how the work should be read. It cannot have a body in a room, a history in a city, a stake in an argument.
These are not romantic claims about the ineffable human soul. They are material claims about what criticism is. Art writing, when it is doing its job, is not the production of plausible text about art. It is the production of situated knowledge about art—knowledge that comes from being somewhere, knowing someone, remembering something, risking something. The AI can simulate the form of that knowledge. It cannot produce the knowledge itself, because it has never been in a room, never been broke, never been angry at a curator’s decision, never watched an artist-friend cry after a bad review, never had to decide whether to write honestly about a show that a friend curated.
The danger is not that AI will replace this kind of writing. The danger is that the economic pressure to use AI will make this kind of writing unsustainable, and what replaces it will be plausible text without situated knowledge—criticism that reads smoothly and says nothing that could not have been generated by anyone, anywhere, with access to the same training data.
The Institutional Silence
Art institutions have been notably quiet on the question of AI in art writing. Galleries that commission catalogue essays have not, to my knowledge, begun asking writers to disclose whether they used AI tools. Funding bodies that support artist publications have not issued guidelines. The Arts Council England’s most recent strategic documents make no mention of the issue. This silence is not neutral. It is a policy by default, and the default is to let the market decide.
Letting the market decide means letting the most precarious writers absorb the ethical and economic costs of AI adoption while the institutions that benefit from their work—the galleries that need essays, the artists who need critical context, the funders who need evidence of public engagement—remain unaccountable. A gallery that pays £150 for a catalogue essay and does not ask how it was produced is effectively subsidising its publication programme through the hidden labour of prompt engineering and the stolen corpus of AI training data. The institution looks solvent. The writer looks productive. The actual conditions of production are obscured.
There are parallels here to the unpaid internship system I have written about before—the way major London galleries extract labour from young workers who can afford to work for free because their families subsidise them. The AI-assisted essay is the technological equivalent: a way to extract writing from workers who cannot afford to write, by giving them a tool that makes their labour invisible. In both cases, the institution benefits. In both cases, the worker’s precarity is the engine.
Toward an Ethics of Refusal—and an Economics of Survival
I am not interested in condemning the art writer who uses an AI story generator to meet a deadline she cannot otherwise meet. Condemnation is cheap, and it usually comes from people whose rent is not at stake. The question is not whether individual writers should resist AI. The question is what collective conditions would make resistance possible.
Those conditions start with money. A minimum fee for catalogue essays, set by a body with enforcement power. A requirement that galleries disclose whether they permit AI-assisted writing in their commissions. A fund, perhaps administered by a writers’ union or a coalition of independent publishers, that supports art writers who refuse AI on ethical grounds—not as a prize for purity, but as a material counterweight to the economic pressure to comply. The Authors Guild’s best practices offer a starting point: transparency about AI use, refusal to claim authorship over machine-generated text, and a clear-eyed acknowledgment that the tools currently available are built on theft. But best practices without economic support are just moral pressure on the most vulnerable.
There is also a role for editors. The editor who commissions an essay can ask, explicitly, whether AI was used in its production. They can adjust fees to account for the time it takes to write without assistance. They can refuse to publish text that has been substantially generated by a machine, not because it is necessarily bad, but because its existence undermines the conditions for the writing they claim to value. These are small acts, but they accumulate. An ecology is made of small acts.
The regional dimension matters here too. If art writing outside London is to survive as something other than a content-creation service for institutional websites, it needs infrastructure: residencies for critics, commissioning budgets that reflect the cost of living in the city where the writer lives, networks that connect regional writers to each other and to national publications without requiring them to move to London. The AI story generator is not the cause of regional art writing’s fragility. It is the symptom, and it will become the solvent if the fragility is not addressed.
The Prompt as Document
There is one more thing to say about the prompt. The prompt—the set of instructions a writer feeds to an AI to generate a draft—is a document of labour. It contains the writer’s knowledge, her editorial judgment, her sense of structure and tone, her decisions about what to include and exclude. It is, in a compressed form, the essay she would have written if she had time. The machine’s output is a degraded version of that essay, and the writer’s subsequent editing is an attempt to restore what the degradation lost.
If we are going to have a serious conversation about AI in art writing, the prompt should be part of the record. Writers who use AI tools should consider preserving their prompts, publishing them alongside the finished text, making visible the work that the machine obscures. This is not a solution to the ethical problem. It is a documentation of the labour conditions that produced it. The prompt is evidence. It says: here is what I knew, here is what I wanted to say, here is what the tool did to it, here is what I had to fix. Read the essay. Then read the prompt. Then ask yourself who did the work.
The writer in the café on Oldham Street finished her essay. She edited the AI draft heavily, rewrote the conclusion entirely, checked every factual claim against her own notes. The result was competent, publishable, unremarkable. The artist was pleased. The gallery paid the fee. No one asked how the text was made. She told me later that she felt she had gotten away with something, though she could not say exactly what. Theft? Fraud? Survival? The categories blur when the alternative is not writing at all.
This is where art writing finds itself in 2026: caught between an economic model that cannot sustain it and a technological solution that cannot replace it, but can certainly degrade it. The AI story generator is not the enemy. The enemy is the condition that makes it necessary. Until that condition changes—until art writing is paid as work, not subsidised as passion—the prompt will remain what it is now: a piecework ticket, punched by a worker who deserves better.