There is a line in Harold Bloom that I have returned to more times than I care to admit: “Influence is Influenza — an astral disease. If influence were health, who would write a poem?” He wrote it in 1973, in a book that was either a masterwork of literary criticism or an elaborate Freudian fever dream, depending on which of your professors you asked. The book was called The Anxiety of Influence, and its argument, stated plainly, is this: every writer who has ever put pen to paper has been haunted by the writers who came before them. Not inspired — haunted. The distinction matters enormously to Bloom.

To be inspired by Milton is comfortable, even flattering. To be haunted by Milton is to feel, in the pit of your creative stomach, that the old man got there first. That he said the thing you wanted to say, said it better, said it with a grandeur you can never replicate, and now you must write anyway — knowing that anyone who reads you well enough will see his shadow in your work. For Bloom, this haunting is not a problem to be solved. It is the condition of all serious literary creation. The only way out is through: through an act of deliberate, creative misreading — what Bloom called misprision — a willful distortion of the precursor’s meaning that creates, paradoxically, something new.

I want to think about what happens to that theory now. Not as a literary exercise, but as a practical question for anyone who sits down to write something in 2026 — in a world where the most formidable presence in the room is not Milton, not Woolf, not even the collective weight of everything you’ve ever read. It is a language model that has ingested, processed, and can fluently reproduce the statistical essence of approximately all of it.

The anxiety of influence just changed shape. I’m not sure we’ve caught up with it yet.

What Bloom Was Actually Arguing

To update a theory, you have to understand it precisely — not the popularized version, which tends to collapse “anxiety of influence” into a vague notion that writers are nervously imitative. Bloom was making a more specific and stranger claim.

His central insight was that literary history is not a gentle relay race, each generation passing a baton to the next in collegial succession. It is an agon — a Greek word for contest — a violent, psychologically fraught struggle in which younger writers must kill their precursors without ever being able to actually kill them. You cannot kill Milton by ignoring him. You cannot kill Shakespeare by pretending not to have read him. The only move available to what Bloom called the “strong poet” is to misread the precursor so forcefully and so creatively that you appear to have invented something the precursor was merely gesturing toward. Done well, the effect is dazzling: the later poet manages to make the earlier one seem like a rough draft.

Bloom identified six stages of this process — his “revisionary ratios” — and while their names (clinamen, tessera, kenosis, daemonization, askesis, apophrades) draw on sources as varied as Lucretius, St. Paul, Neoplatonism, and ancient Athenian ritual, the underlying dynamics are precise. The young writer first swerves away from the precursor at a certain point, implying the master “went accurately up to a certain point, but should have swerved.” Later, the strong writer achieves what Bloom calls apophrades — the return of the dead — in which the precursor seems, uncannily, to have anticipated the later work rather than the other way around. Read Yeats after Seamus Heaney and you feel it: something has been reversed in the order of time. Blake looks different once you’ve lived inside Yeats — as if the earlier poet had been working toward the later one rather than away from him. That reversal is the measure of the strong poet’s achievement.

The key word in all of this is misreading. Not plagiarism, not homage, not pastiche. Bloom is describing a creative act that can only happen through deep, intimate, pressurized engagement with what came before — and then a chosen, necessary distortion of it. You have to know what you are misreading. You have to love it enough to feel its authority, and then you have to refuse it.

This is the theory that the arrival of large language models has quietly detonated.

The Precursor That Cannot Be Misread

Here is the obvious question Bloom’s framework raises in 2026: if an AI system has been trained on effectively the entire archive of human writing — every poem, every novel, every essay, every letter, every blog post, every critical review — what does it mean to “misread” it?

Bloom’s anxiety-model depends on a structural asymmetry. The precursor is a finite, specific consciousness — Milton, say — with particular obsessions, blindnesses, and theoretical commitments. The ephebe (Bloom’s word for the younger poet) can misread Milton because Milton is misreadable. There are gaps in him, overreadings and underreadings, places where his commitments outran his evidence or his vision exceeded his control. The strong poet finds those gaps and levers them open. The misreading is only possible because the original reading was incomplete.

An AI trained on the totality of human writing is not misreadable in that way, because it has no commitments of its own. It does not have a theory of the sublime that it has pushed too far. It does not have a blind spot created by its historical moment, or a politics it was trying to smuggle into its aesthetics. It is not, in Bloom’s sense, a poet. It is something more like the undead library — all of the precursors, simultaneously, at statistical average.

And this is where the new anxiety enters, and why it is more vertiginous than anything Bloom imagined.

The earlier anxiety was about standing in the shadow of a specific greatness. The new anxiety is about standing in the shadow of all greatness at once, rendered not as individual human consciousness but as aggregate pattern. The shadow you are trying to step out of is no longer cast by Milton or Woolf or Baldwin — it is cast by the calculated mean of everything they ever wrote, and everything everyone else ever wrote, processed into a system that can reproduce the texture of literary achievement without any of its source.

If Bloom’s anxiety was Oedipal — the son struggling with the father — the new anxiety is something more disorienting. It is the anxiety of facing a mirror that shows you every human face at once, and wondering whether your own face is distinct enough to be worth looking at.

T.S. Eliot Was Right (And That’s the Problem)

Before we can understand what AI does to Bloom, we need to detour through his great opponent in the theory of literary tradition: T.S. Eliot.

In 1919, Eliot published a short essay called “Tradition and the Individual Talent” that made a claim almost diametrically opposed to the spirit of Bloom’s later argument. Eliot argued that the best poets were not those who struggled most violently against tradition, but those who most fully absorbed it. He described the poet’s mind as a catalyst — a filament of platinum introduced into a chemical reaction that enables the combination of elements without itself being changed. The poet, for Eliot, should be impersonal: a medium through which tradition flows, transformed into something new, but never a vehicle for raw personal emotion or private obsession.

The mature poet, in Eliot’s account, is defined by an extraordinary act of self-abnegation. He must absorb the whole of the literature of his civilization — not just the parts he likes, not just the contemporaries, but the full continuous order stretching back to Homer — and then surrender himself to it. The result, paradoxically, is genuine originality: not the originality of the person who refuses all influence, but the originality of the person who has been so thoroughly formed by the tradition that they can advance it rather than merely repeat it.

What Eliot could not have known — and what makes his argument newly uncomfortable — is that he was accidentally describing something like a large language model. An AI trained on the complete archive of human writing is, in a certain sense, the perfected Eliotic poet. It contains the whole tradition. It is maximally impersonal. It has no personality to intrude upon the work. It can synthesize, recombine, and produce prose that carries the formal marks of literary quality without the contamination of a self.

The horror of this is not that AI writes badly. A December 2025 study from University College Cork, using literary stylometry to compare human and AI-generated prose, found something I cannot stop thinking about: AI writes with “narrow and uniform patterns,” while human authors display “greater stylistic diversity and individuality, shaped by personal voice, creative intent, and individual experience.” The horror is precisely the opposite of Eliot’s ideal: the AI writes too well. Its writing is polished, fluent, competent — and it produces this competence with the consistency of a machine, because it is one. Every piece it generates converges toward the statistical center of what good writing looks like. It is the average of excellence. It is tradition without the individual talent.

Which means: if Eliot was right that tradition is the foundation of originality, and if AI now is the tradition, made immediately and infinitely accessible — then what is left for the individual talent to do?

The New Revisionary Ratio: The Swerve from the Mean

Here is where I think Bloom’s framework can be salvaged — but only if we update what we think the “precursor” is.

Bloom’s strong poet achieved originality through misreading. The precursor was a specific great writer. The act of creation was a swerve away from that specific person. The new strong writer — the writer who wants to make something genuinely their own in 2026 — does not need to swerve away from Milton or Woolf. She needs to swerve away from the mean.

The enemy of originality is no longer the great predecessor. It is the average expectation. The thing that wants to colonize your prose and make it into something serviceable, fluent, and forgettable is not the voice of Shakespeare; it is the statistical ghost of all the writing that has ever been produced, smoothed into a frictionless surface, available on demand. The new clinamen — Bloom’s first and most basic revisionary ratio, the fundamental swerve — is the swerve away from that surface.

This reframing has a strange implication: the very qualities that make writing difficult — the idiosyncrasies, the risks, the half-worked-out ideas, the sentences that are almost too long, the associations that don’t quite cohere on first reading — are now, more than ever, marks of genuine authorial presence. AI writing is consistent and smooth because it is optimizing toward the probable. Human writing is interesting precisely where it departs from the probable. The weird sentence. The unexpected pivot. The place where the argument doubles back on itself because the writer noticed something she hadn’t intended to notice.

This is why, I think, so many readers can feel the difference between human and AI-generated prose even when they cannot articulate what they’re detecting. They are detecting the residue of a particular consciousness. The AI’s writing has no such residue — it has been averaged away. Human writing, at its best, is full of it: the pressure of a specific life behind the language, forcing it into shapes it wouldn’t otherwise take.

Voice Is Not Information

I want to dwell on this for a moment, because I think the common understanding of “voice” in writing is imprecise in a way that makes people underestimate both what is at stake and what is irreplaceable.

Voice is often described as the stylistic signature of a writer — the distinctive rhythms, the preferred cadences, the characteristic imagery. This is true as far as it goes, but it reduces voice to technique. And technique can be learned, imitated, and, it turns out, generated.

Voice in the deeper sense is not a technique. It is a pressure. It is what happens when a specific consciousness, shaped by particular experiences, particular losses, particular obsessions, particular intellectual preoccupations that no life other than that life could have produced, is brought to bear on language. James Baldwin’s sentences have the sound they have because of what he lived through and thought through. Didion’s cool, staccato logic is inseparable from her specific relationship to anxiety and control. Montaigne’s associative wandering is the record of a particular mind’s actual movement — not a performance of digression, but genuine digression, tracked honestly.

This is what the UCC researchers were measuring when they found that AI writing clusters toward a “compact, predictable” style while human writing is “varied and idiosyncratic.” They were measuring the presence or absence of a consciousness that pushed language away from the center. But their findings point toward something the researchers didn’t quite say: the variance in human writing is not random noise. It is signal. It is the specific shape of a life pressing against the language.

AI cannot have a life. It can have training data. These are not the same thing.

And this matters more now than it ever did. In a world where serviceable prose is infinitely cheap, the only thing that justifies reading any particular piece of writing is the presence of a consciousness worth spending time with. The information can be summarized. The argument can be extracted. What cannot be extracted — what makes a reader come back, share a link, read everything an author has written — is the felt sense that there is a specific person there, with a specific way of seeing, and that spending time with their prose is spending time with a mind that will change how you use yours.

This is what Bloom was ultimately defending, I think, underneath all the Freudian machinery and the Kabbalistic terminology: the irreducible value of the individual consciousness in literature. The anxiety of influence was real because what was at stake was real — the question of whether this particular writer would achieve genuine poetic selfhood, or whether they would be simply absorbed into the tradition, their voice indistinguishable from the sound of the past.

The AI threat is the completion and reductio ad absurdum of that older anxiety. We now have a system that is, in a literal sense, the tradition speaking. The question for every writer now is the same as it was for Bloom’s ephebes, only starker: are you going to make something that is recognizably, undeniably yours — or are you going to produce the sound of writing?

The Apophrades Problem

There is one more of Bloom’s revisionary ratios I want to examine — the last one, which he considered the most mysterious and the most revealing of the strong poet’s achievement.

Apophrades comes from the Athenian concept of the days on which the dead return to reinhabit the houses in which they once lived. In Bloom’s account, the strong poet achieves apophrades when their work becomes so powerfully itself that it retroactively changes how we read the precursors — as if the earlier writers were working toward the later one, rather than the other way around. Read Blake after you’ve spent a long time with Yeats: something has shifted. Wordsworth starts to feel, in the wake of Keats, like he was working toward a sensibility he never quite arrived at. The later poet has remade the earlier one.

The question I can’t shake is: can a writer achieve apophrades in relation to AI?

I think the answer is yes, but not in the way you might expect. The achievement is not to write something so extraordinary that it changes how we read AI outputs — that’s a category error; AI outputs aren’t literature in the sense that admits of the apophrades relation. The achievement is something more mundane and more difficult: to write something that is so completely itself, so thoroughly the product of a specific consciousness, that it cannot be mistaken for the mean. To write something that makes the reader feel, unmistakably, that this is the record of a person thinking — not the record of text statistics predicting what a thinking person sounds like.

That gap — between simulation and the actual thing — is what you are trying to hold open every time you sit down to write now. And you hold it open not by technique, not by stylistic tricks, not even by being “more human” in some vague aspirational sense. You hold it open by having something actually at stake in what you’re saying. By caring about it in a way that shows up in the language. By being willing to be wrong, and publicly, and on the record.

AI has no stakes. It cannot be wrong in the way that matters — the way that means something was risked and lost. Every piece of genuine writing is a bet, and the writer’s self is what is wagered. That wager is what creates the pressure that makes the language move in ways it would not otherwise move.

Bloom’s ephebes were anxious about the precursors because the precursors had made that bet and won, spectacularly. The strong poet had to find a way to make her own bet in the shadow of that.

We are all ephebes now, and the shadow we are writing in is immense. But the bet is still ours to make. The wager is still the self. And the self — shaped by particular failures, particular loves, particular years spent reading in particular rooms, thinking through particular problems that only arose because of the specific sequence of events that constituted this particular life — is still, irreducibly, the one thing no model can train itself into having.

What This Means, Practically

I have been deliberately avoiding, in this essay, the practical question: what do you do with this? How does a writer actually work in 2026 in a way that is honest about the situation?

I don’t have a system. I have a suspicion.

The suspicion is that the writers who will matter — who will be read, remembered, sought out — are the writers who are willing to be most specifically themselves, at the cost of the smoothness and accessibility that the machine makes so easy. Not deliberately obscure, not performatively idiosyncratic, but genuinely committed to the particular angle from which they see things, even when that angle is inconvenient.

The anxiety of influence, in Bloom’s formulation, was resolved by the strong poets who found a way to make their distinctness felt against the tradition. The new anxiety — the anxiety of the statistical mean, the anxiety of the averaged tradition that can speak fluently in any voice — is resolved by the same means, only the opponent is different. The swerve is not away from Milton. It is away from the expected. It is away from the probable. It is toward whatever your particular consciousness sees that the aggregate does not — which, if you are paying attention to your own thought and honest about your own experience, is always something.

Eliot said the poet must surrender the self to the tradition. Bloom said the poet must defend the self against the tradition. Perhaps in 2026 the task is to do both simultaneously: to know the tradition well enough to carry it, and to defend the self fiercely enough against the mean to remain distinct within it.

That combination — saturated with the past, insistently present — is what writing was always for. It is still what it is for. It has not changed. Only the scale of the challenge has.

A Note on Belatedness

Bloom had a word for the condition of all post-Miltonic writers: belatedness. The feeling that you have arrived too late, that the great things have already been done, that the canonical work in your medium has already been produced and all you can do is make variations on it. Every serious writer knows this feeling intimately.

The AI revolution has produced a strange new form of belatedness — not the sense of arriving after the great writers, but the sense of arriving after the comprehensive archive. The archive is now not merely available; it is animate. It speaks back. It offers you, in response to a prompt, something that looks like what you were trying to say, only faster, and smoother, and possibly better on the surface, and completely hollow underneath.

This might be the most clarifying thing that has happened to writing in a long time.

Because the archive was always there. The canon was always available. What the availability of AI writing makes visible — insistently, unavoidably visible — is what writing was for, underneath all the talk of craft and technique and influence and tradition. It was for the mind. It was for the specific, irreplaceable experience of encountering another consciousness working its way through something real. Not the information. Not even the beauty of the sentences, though that matters. The encounter.

You cannot have that encounter with a language model. You can have a useful interaction, an interesting simulation, a serviceable prose output. You cannot have the thing that happens when a writer has thought something through at cost to themselves and put it on the page for you.

Bloom spent his career arguing that poetry mattered because strong poets existed. That the central drama of literary history was the drama of individual consciousness defending itself against the pressure of the tradition and coming through as itself — scarred, transformed, but distinct.

He was right. He is more right now than he was in 1973.

The tradition has never been more formidable. The pressure toward dissolution into it has never been greater. The reasons to resist that pressure — to make something that is insistently, undeniably yours — have never been clearer.

Write that way, or don’t write. But if you write, put yourself in it. The only thing the machine cannot provide is you.

This essay draws on Harold Bloom’s The Anxiety of Influence (1973) and T.S. Eliot’s “Tradition and the Individual Talent” (1919). The stylometric research on AI and human writing is from Dr. James O’Sullivan at University College Cork, published in Humanities and Social Sciences Communications (December 2025).

Further Reading