actually, AI is making you a bad writer

and also kind of a fascist

My latest novel, Carrion Crow, was released in the US and Canada this week. The TLS called it 'a brutal, unflinching medi­tation on the phenomenology of womanhood’. The Guardian called it a ‘claustrophobic gothic grotesque’ and a ‘stomach-clenching contender for awards’. The best interview I’ve done about it is here. If you like social critique dressed up in abject Victorian viscerality, do grab a copy from your favourite indie bookstore, or from Powell’s.

---

Like almost all other novelists working today, I am, to some degree, engaged in the business of teaching other people to write. I’ve been at it for close to twenty years in some fashion or another, first as a nonfiction editorial intern and proofreader, then copyeditor, then senior nonfiction editor, then a copy writer at a North American tech startup, then freelance ghostwriter, developmental editor, fiction editor and now editorial director of a tiny literary magazine, as well as fiction writing mentor, workshop facilitator, etc. Mine is a path that has meandered more through editorial than some others, perhaps, but this is where we all end up, because teaching other people the skills of writing ends up more valuable, in most cases, or at least more reliable, than royalties from our actual books. As the publishing industry descends further into chasing tedious algorithmically-decided trends and editors are disempowered from actually editing, because fewer people care if a book is terribly written and the whole business operates more on hype than quality, and as humanities degrees become either completely unaffordable or are scrapped, the job of the published writer becomes less about actually writing and more about feeding more fodder into this machine. Hence: we all end up being asked to run workshops, to mentor, to teach.

For the most part, I enjoy this work, especially mentoring. The joy of teaching comes when you are allowed to do it well; when you can either teach something you really care about, and understand, to a group of people who are genuinely keen to learn, or when you are gifted enough time with a single writer to really dig into their weeds of their writing: their process, their motivations, their goals, their prose. Both of these require commitment on the part not just of the teacher but of the student, and it is worth remembering that all such teachers were once on the other side of this equation, going through the hard stuff of learning about their own weaknesses, their crutches, their lack of confidence, their technical errors, their sentences—and, more importantly, how to find their authentic, unique voices, and how and when to break the ‘rules’.

This is the key to finding your voice as a writer: learning when NOT to do things the common way. Knowing how and when to push against conventions, forms, grammar and popular style. How to break through cliché, and form a new way of saying something that may have been expressed, in some way or another, a hundred thousand times before you. It requires constant self-critique, reassessment and discipline. None of us were born knowing how to write well, and the further you get into your career, the less you’re convinced that you ever have learned. Those of us who have published books have been through that particular ringer many times. We know it’s hard. We also know that putting your own work through the process of criticism is unavoidable if you want to become the best writer you can be, and that ultimately, when you produce work you’re intensely proud of, all the effort becomes worth it.

There are, however, times when this work is utterly miserable, and the circumstances in which it’s miserable are this: the person you’re teaching isn’t at all interested in examining their own output, in hearing constructive critique, in changing how they’re doing things in any way. They have signed onto your course, or enrolled at an institution, or have engaged your services because they want to be told that they’re a singular genius, and what they are really hoping is that you’ll say their unpublished manuscript is unimprovable, and hand them over immediately to your agent, as if you have any control over who she signs, or to your editor, as if you don’t exist in a perpetual state of terror that your editor will no longer be interested in your work, let alone anyone you say they should sign. They are convinced, or have been convinced, that what they’re working with is already perfect.

There’s a brand new version of this, of course, and it’s the above, mixed with generative AI. That is: a person has been ‘writing’ with AI, or ‘editing’ with AI. They consider AI to be an expert in writing, because they have been told it is, a thousand times over, and because it is programmed to pretend it is, just as it is programmed to pretend it is an expert in everything, even though the most basic research reveals it not to be. The best that AI can do, as a writing or editing tool, is to make your writing functional, and probably boring. Yet those creating the relentless and inescapable hype around AI position it as a tool to be used to become skilled at anything, by anyone—and they target people with acknowledgement of their real-world grievances, cynically using this to funnel endless amounts of money into the pockets of the already mega-rich, who both benefit from and uphold the very system that creates these grievances.

Here is an example of what I mean. There’s been a notable uptick, this year, in celebrities and CEOs shilling for AI with a particular angle: that women, especially, are being ‘left behind’ by these technologies, not because they’re perhaps smart enough to be critical of hyper-capitalist, resource-draining industries that care devastatingly little for the effects of their profit-seeking, but because they are simply too timid, too unconfident to embrace AI in the way that men apparently are. This isn’t exclusive to this year, however. Take this, from a Time magazine op-ed, from January 2025, entitled Why We Need More Women in the AI Revolution:

Imagine having a mentor available at any hour, offering guidance without judgment. Master new skills at your own pace, free from the weight of imposter syndrome that haunts so many women in male-dominated spaces.

The sexism is written all over this, of course, but the real issue with it is slightly more hidden: imagine a mentor offering guidance without judgment. Think about this for a second. How is a person supposed to get better at anything if they are learning from a person who has no capacity to judge what they are already creating? Judgement is a key part of being a teacher; you must assess what has been written, analyse it for quality or originality, measure it against what the writer is trying to produce and help to steer them towards the better version of that work. This is what editors do, and it is why writers love good editors. The idea of a writing coach or editor who never tells you you’re wrong or that what you’re writing could be better is a strange manifestation of the supposedly ‘woke’, supposedly leftist tendency to hand out participation trophies for everything so as not to make anyone feel bad—except it’s coming from the very people who critique this idea. Learn to be better, it promises, without ever being told you’re not a special perfect genius! You don’t ever need to feel less than!

When you remove the judgement of an editor from the writing process, the work becomes noticeably worse. But the notion of quality—of an objective or even subjective measure of something’s success as a work of art—is notably absent from all the conversations around AI and the creative industries. The idea that some creative work is bad and some is good is anathema to exactly what AI companies are selling. The USP of generative AI is that it can instantly turn anyone into a professional at anything, and introducing a measure of quality control into this makes the sales pitch fall apart instantly. If anything, the concept of quality hovers above the discussion, circling, like one of those malevolent ghosts from Mario Party trying to steal your hard-won stars. You are left desperately batting it away, inevitable though the theft is.

The other day, the algorithm decided I needed to see a glossy advert pointing me towards a glossy website for a ‘free AI book writer and editor’. I clicked through, out of curiosity. All the sample text was awful, the worst sort of rote, clichéd, personality-free and barely functional prose. Even the text from the ad was terrible:

Rachel never considered herself a writer.

Yet here she was, writing a book.

But despite bounds of enthusiasm, Rachel would be the first to admit:

Writing a book is hard.

Keeping track of storylines and characters.

Editing thousands of words.

Trying to get honest feedback.

At times, she thought about quitting.

Thankfully, she didn’t.

To put my tedious hat on for a moment: this isn’t even grammatically or linguistically correct. The phrase is ‘boundless enthusiasm’, not ‘bounds of enthusiasm’. It should be ‘her boundless enthusiasm’, if anything. There are four sentence fragments, not sentences, following the the semi colon. Either this has been generated by the AI service advertised, or it has been edited by it, or it has been written by a marketing person in order to sell the writing and editing AI service. Whichever it is, it’s not a great advert for what it’s trying to sell. The above is very basic stuff, but it is also the stuff that you increasingly need to teach, as an editor or mentor, because people know very little about sentence structure these days. I spend a lot of my time explaining subject/object issues in sentences that leave you unable to understand who is doing what, or inconsistent lists, or subordinate clauses. Understanding and playing with grammar and syntax are key parts of expressing yourself through writing, but you have to understand the rules of them first, and when we reduce everything down to the simplest possible sentence fragments that aren’t even grammatically logical we give up both clarity and artistry. There’s a reason so many readers complain that prose these days is just bad: it is bad. And AI is making it a lot, lot worse.

As others elsewhere have noted, it is almost impossible to edit AI-generated work, because there is no intellectual or emotional framework undergirding it, and no unachieved intention behind its creation. On a line level, it is just extremely poor, full of platitudes and unoriginal, LinkedIn-level phrases and sentence fragments, no interest in syntactic flow. But worse is that there’s just nothing deeper than that, no natural voice trying to get out, nothing that really needs to be expressed. You can’t pare back the unnecessary parts to get to what the author was really trying to do, because the author—the large language model putting words together in whatever string it guesses you might want—isn’t trying to do or say anything. There is quite literally no reason for the work to exist, so you can’t even try to get at that intention, and neither can you strip away all the cliché, because it is practically all cliché.

Cory Doctorrow—sci fi writer, technologist, and user of AI—explained this extremely neatly in his new book The Reverse Centaur’s Guide to Life After AI: How to Think about Artificial Intelligence Before It’s Too Late. Discussing why AI will never be able to write anything decent, he says:

When an artist prompts an image generator (or a creative writer prompts a chatbot), the only artistic communication the resulting piece can yield is that prompt. The generative AI doesn’t know anything about the big, numinous, irreducibly complex feeling in the artist’s head. The AI can’t know that. It’s a next-pixel-guessing statistical machine, not a mind reader.

Using AI as a writing teacher is like using the room in Searle’s famous ‘Chinese Room’ thought-experiment to teach you Mandarin. Searle describes a room in which a non-Chinese-speaker sits with a set of instructions: when certain Chinese characters are slipped under the door, respond by slipping out the corresponding characters, and thus the person outside the room will believe you can speak the language. The person inside performs as requested. All well and good. But can this person understand Chinese? Of course they cannot. All they are doing is responding to prompts by rote. This is not how we understand what knowledge is, and neither is it how we use language. There is no fluency here, no possibility of expression or inventiveness, no judgement. The room has almost no real engagement with what it has ‘produced’.

You might believe that AI is teaching you to put the ‘right’ things on the page, but the fact is that there are no ‘right’ things when it comes to expressive or imaginative writing. Writing requires you to tap into your own voice, your own preoccupations, your own linguistic nuances and tics, your own feelings about long sentences, and auto-cancellations, and tautologies, and semi-colons, and all the other devices we use to convince and argue and describe and evoke, even—in fact, especially—the ones that are technically ‘wrong’. Even if you think that you, in fact, are using it differently, and that you use the AI ‘advice’ to prompt you to write the opposite of what it says, you are still being innately led by the AI; if you write the exact opposite word, you are still tied to its suggestion. You are being dragged along by a blind dog, thinking that you are cutting your own path.

---

Last week I saw a post on Substack from a man defending his use of generative AI. He claimed that ‘the craft of writing has always involved collaboration’ and that ‘using [AI] as a collaborator is akin to working with a gifted editor’. Only one of these things is true. The latter is false because a) AI is not ‘gifted’ at anything and b) applying an algorithmic tool can never be akin to working with a human being. If this was true, the advent of spellcheck and grammar check on Microsoft Word would have wiped out the need for editors and proofreaders in traditional publishing.

This position was also taken by Marc Carter, for an incredibly poorly argued piece in The Times, in which he claims that T.S. Eliot and Ezra Pound invited and utilised feedback from other people in their processes, and so using AI for writing or editing is exactly the same. His own argument undermines his position, however, because he equates AI with human collaboration. These things are fundamentally not equal. A human has experiences, embodiment, judgement, intelligence, expressiveness, a working knowledge of language as a part of their consciousness; if a sentence is intending to express or describe a feeling, a person can measure this against the real-world experience, and feedback whether or not it succeeds. All of this they bring to their appraisal of the work. AI does not have this. Carter cannot even fathom that a human expert might be better than AI, despite recent studies showing that even when doing very basic summarising tasks, AI is wrong 45% of the time. Discussing Hilary Mantel’s collaboration with Cromwell scholar Mary Robertson, and the two decades of research and thinking the author did before she published Wolf Hall, Carter says (in very AI-coded sentences) that ‘Mantel’s greatness resided not in the travel expenses but in 20 years of thinking about Cromwell before she wrote a word. In knowing what to leave out.’ And yet, immediately after he says:

Google arrived and suddenly much of what had required a flight to California became accessible in minutes from a desk in London. And now AI goes further still.

AI, by this very argument, necessarily removes the 20 years of thinking and researching that Carter has literally just said was key to Mantel’s genius. The entire argument here is that using AI instead of doing all this is just as good. This is, in fact, exactly the problem: that we are leading people to think they don’t need to develop relationships with the humans who are experts in their field, don’t need to create a dialogue with people whose judgement has been honed by decades of thinking and doubting and looking at primarily sources and rejecting easy explanations. By Carter’s own argument, neither Robertson’s research or Mantel’s supreme skill in storytelling or prose or character creation matter, because now AI can just do either one, despite all evidence showing that it can’t do either to any degree of accuracy or originality. The argument negates itself.

In Carter’s piece there is not a single mention of quality, of something being good or inventive or original or even basically effective in achieving its own goals, except for those which actually undermine his point. He mentions that AI helps people realise the stories they want to tell, but not whether or not this makes them good writers, by any standard or measure. All that matters is that everyone can be a writer now; everyone can say they have written something. Discussing his daughter’s use of AI tools, he says this: ‘The industry’s current approach would call my daughter’s process cheating, and shut the door on her before she had even reached it.’ Except this is not what is happening. If anything, the industry is currently too in thrall to extremely badly written books that simply appeal to some algorithmically created trend, with no interest in their actual quality or longevity or uniqueness of voice or impact on readers. The industry as it stands loves the rapid creation of derivative works with barely functional prose that will sell millions because they have a strong hook and nothing else. The truth is that the industry would be—and notably is—fine with AI generated work if it turns a profit; it is readers that don’t like it. People are picking up on AI-created work because it is bad. Readers are noticing nonsensical metaphors, and thin characterisation, and overuse of the rule of three, and unmoored sentence fragments, and the extremely clichéd tells of AI precisely because they are bad, and overused, and clichéd, and because there has been no intelligent critique applied to try to remove some of these things from the whole because these books are not being edited, because editors are overworked and underpaid and no one really cares if these books are good or bad. The reason it is so hard to pin down exactly if a work is AI generated or if it is just bad human writing is that AI-generated work is just an amalgamation of all the worst, most obvious, most overused and nonsensical parts of much human writing and the experts—the editors—aren’t being allowed to do their jobs. The lack of judgement is entirely the issue.

But we are told that this doesn’t matter. There is an inherent position taken in Carter’s piece, and over and over again in others, that actual talent or skill should not be a necessary condition for publishing a book. That you should not have to work hard at the craft of writing, as you have previously had to for many centuries, to hone a unique voice, to express yourself in new and inventive ways, to put the words on the page in a beautiful and surprisingly way that is entirely your own. To insist on this is to be exclusionary, to create barriers, to elitist. To be ‘accessible’, by this view, is not to remove the very real obstructions that stop talented marginalised people from enjoying equal opportunities to the non-marginalised; instead, it means that the publishing industry must publish any old shit written (or indeed AI-generated) by anyone who wants to be a writer, and if it has nothing to say and only exists to be published, then that’s fine, and to suggest anything else is sheer snobbery. The truth is that there are real, material barriers to people who are trying to make a career in the arts, right now, and especially those who are doing this from a marginalised position, but the technology that Carter and others posit as a solution is in fact exacerbating those very conditions. If you actually gave a fuck about barriers in the creative arts, you would swear off generative AI forever.

I think it does an enormous disservice to writers, including Carter’s daughter, to tell them that using AI to generate or edit work will make them good writers. It will not. Improvement as a writer requires human intention, human critique, and extreme humility. And when their writing is exposed to thoughtful readers, the reader response will not be positive. When I work with new writers, I tell them that part of our job is to stress-test their work before it goes out into the world, to find all the holes and the inconsistencies and the clichés and the confusing sentences and the lack of clarity before readers find them and dismiss the work through frustration. We cut away what is not needed, or what isn’t working, to get to the real, strong, unique stuff underneath; to let the prose and the ideas shine. We expand on what is promising, and build the writer’s confidence in the things they do that are truly brilliant and truly their own. AI cannot do any of this because it cannot use judgement, only comparison against what’s most common. We cannot pretend that using poor tools to create makes someone a good writer; it doesn’t. And of course, it is perfectly fine to be a bad writer; I am a bad painter, a bad piano player, a bad weightlifter. We all must be bad for a long time before we get good. But I am not telling the world that these industries are wrong for excluding me on the basis of lack of skill, nor writing numerous op-eds insisting that my bad art is should be considered as legitimate as that of Jenny Saville or Jasleen Kaur. I am not demanding to be taken seriously as a painter.

The obsession with artistic skill being somehow inherently bourgeois (rather than, for instance, the means of production of art being controlled by a certain class), and the related belief that technology will free the masses from the yoke of this bourgeois oppression is, of course, a fascist idea. I do not mean that it is vaguely right wing; I mean that the idea of human art being decadent was clarified in the writings of the Italian Futurists, specifically those authored by Filippo Tommaso Marinetti, which I’ve written about here. Marinetti, who went on to co-author the Fascist Manifesto, wanted to remove the ‘I’, the human, in writing; he wanted to introduce the violence of technology to the creation of art, in order to break it free of the supposedly decadent and elitist artists who produced it. Then as now, instead of focusing on real material barriers to creative expression—instead of tackling illiteracy, for instance, and its related concern, poverty—the insincere solution was the machine. And this purported (and false) concern for the proletariat was convincing, at least in the early days, to even young, thoughtful figures on the left. As Mark Ford outlined in the London Review of Books twenty-two years ago:

Gramsci, too, was initially exhilarated by the ‘impetuosity of their youthful energies’, their conviction that an alliance of artistic and technological innovation would sweep away bourgeois culture once and for all.

So often are fascist intentions coddled in language of anti-intellectualism, anti-elitism, the idea that automation can somehow free creativity from its ludicrous human constraints. The slide from this into real violence is not theoretical; the Nazis, too, decried ‘degenerate’ art, created by the wrong humans (Jewish, Communist, or ‘un-German’), as dangerous to a functioning state, promoting in its place art that glorified the militaristic, the racially pure, the idea of obedience to a paternalist power. So Mayer wrote about this brilliantly in their book A Nazi Word for a Nazi Thing, and philosopher Judith Butler beautifully summed up how AI functions as an extension of this in a 2023 conversation with Elizabeth Weil:

There’s a narcissism that reemerges in the AI dream that we are going to prove that everything we thought was distinctively human can actually be accomplished by machines and accomplished better… Or that human potential — that’s the fascist idea — human potential is more fully actualized with AI than without it.

Butler calls this the ‘perfectibility thesis’, the idea that there is a version of the perfect human that is achieved by introduction of technology, and they are right to call it fascist, for where there is perfection there is also imperfection: the degenerate, the expendable, the impure. Language, as Butler intimates, is a key part of resisting this process. What gets eclipsed by the uncritical belief in generative AI is, according to them, ‘what’s living in my speech, what’s living in my emotion, in my love, in my language’. This is exactly what another human looks for, when they engage with your work, and what a human will help you to hone in on, bring out, and clarify: what’s living, what’s unique, what’s yours.