AI assistance and the editing line

On what the publishing industry's AI policies are protecting, what they aren't, and how civilisation actually moves.

A lightly edited transcript of two exchanges with Claude, the large language model built by Anthropic. In the first, Lars asks whether the publishing industry's line between human and machine editing is doing real work, or whether it is a category mistake; the reply grants what is sound in the analogy and sets four concerns against it. In the second, Lars presses the deeper claim: that every human advance is built on what came before, and that these tools, and whatever follows them, are part of that same advance. The reply argues that denying this is the real error, while refusing to pretend the transition is free. One concern below was sharpened in a later exchange with a different model, ChatGPT; the decision about what to keep, and what to set aside as cleverness the argument did not need, remained the author's. The essay is, unavoidably, an instance of the thing it describes.

Lars

I want to think with you about how the industry is treating LLM-assisted writing. On my own site I describe my workflow openly: I use a large language model the way other writers use a developmental editor, and as far as I can tell the outcomes are not materially different from what working with a human editor would produce. But several professional short-fiction markets now disqualify any submission "written, developed, or assisted by these tools." My question is whether that distinction is doing real work, or whether it is a category mistake. If the final form of a story is what the reader experiences, why should the route to that form matter so much?

Claude

Where the analogy with human editors holds

The basic claim is defensible. A story is a story; a reader moved by it in 2030 will not care whether the load-bearing line came from a workshop, an editor, a beta reader, or a conversation with me. The reader meets the work, not the process. That is a serious aesthetic claim, and many working writers hold it.

The industry is also inconsistent about it. Writers thank developmental editors in their acknowledgements and no one calls the book less theirs. Gordon Lish cut Raymond Carver's stories nearly in half and rewrote their endings; we still call them Carver's. Editors at major houses reshape structure wholesale. None of it disqualifies the writer. Set beside that history, the treatment of machine assistance is a position the industry has not yet thought through.

Where the line is not arbitrary

There are four reasons editors hold the line that are not theatre.

The first is the training data, and here the usual framing is half wrong. These models were built on the work of writers who did not consent, and the temptation is to say the polish therefore carries stolen traces of Le Guin, Lem, Chiang. But influence is not the offence. Influence is what every mind does: a writer carries a thousand books in her bones and pays royalties to none of them, and a model that has generalised from a corpus is, in that respect, doing what readers have always done. Strip the influence confusion away and a narrower, harder objection remains. The work was ingested wholesale, by a commercial system, without anyone being asked, and that system can then compete with the writers it learned from. That is the real residue: not whether you wrote your story, but consent, and substitution. It is the one concern on this list that no choice of venue resolves. Publish in a magazine or on your own site; either way you have used the tool.

The second is the one that looks like a reason and resolves, on inspection, into a confusion, though an instructive one. Editors worry that a machine blurs the line between editing and drafting. But the line was never real. Your own Lish and Carver case is the proof: Lish did to Carver what we would now call co-writing, and the industry filed it under editing, because the boundary between the two has always been a social convention rather than a fact. The whole range from a comma-fix to a line edit to a developmental rewrite to ghostwriting is one continuous variable that the industry has long pretended was a binary, because a binary is easier to administer. A machine does not introduce that porousness. It makes a fiction the industry was already living inside impossible to keep pretending about. What editors care about underneath the fiction is real and worth caring about: how much of the work is the author's, and whether she is honest about it. The editing-versus-drafting line was only ever a crude proxy for that.

Which is why the tool-ban misfires. It polices the instrument instead of the thing the instrument stood in for. A novel quietly ghostwritten by a human sails through the policy, undetectable and nominally human, while a disclosed, author-controlled machine edit is barred at the door. The rule catches the honest and waves through the dishonest, because it is testing for the presence of a machine rather than for the absence of authorship. It is the calculator confiscated in an exam that was meant to find out whether you understand the maths. The real axis is not editor or drafter, and not human or machine. It is how much of the judgement stayed yours, and how openly you said so. Both of those are real, and measurable, and disclosable. The category the policy tests was never the thing.

The third is what happens to writing as a paid profession. If assisted writers compete on equal terms with unassisted ones, the writers who cannot or will not use the tools are placed at a structural disadvantage, and the economic floor under the work erodes. That floor used to be named: you wrote it yourself. But your own Lish and Carver example shows the floor was never that clean: ghostwriting, workshops, editors who rewrote endings. The profession crossed this line routinely and kept its name. The machines are a difference of degree on a line already crossed, not a new kind of breach. That does not dissolve the fear; it right-sizes it. The live question is how much assistance, and disclosed how, not whether assistance is permitted at all.

The fourth is the one the policies were actually written for, and it is the most concrete. For example, in early 2023 Clarkesworld closed submissions because it was drowning in machine-generated spam: hundreds of worthless stories a week from people hoping to be paid for a prompt. The bans are a blunt instrument against volume. They are not aimed at a disclosed, carefully edited practice like yours; they are aimed at the flood, and you are caught in the same net only because a net is cheaper to cast than a filter. This matters more than the rest, because it means you are collateral, not target. You are not evading a rule written for you. You are declining to be caught by one cast for something else.

Where this leaves you

You are not wrong that the policies are partly arbitrary; the analogy holds, and the Lish and Carver case is unresolved. But they are not only arbitrary. They answer a real question about consent, a real one about how much of the work is the author's, a real economic fear, and a real problem of volume. Publishing your assisted work yourself answers all of it about as well as anything can. You do not pretend the assistance did not happen, you do not ask editors to bend their rules, and you do not argue the rules are wrong. You operate outside them, in the open. The one thing it does not settle is consent. That residue is yours to carry wherever you publish, and saying so is part of what makes the rest credible.

Lars

I take the point about the training data, and I will not wave the royalties question away; that one is genuinely hard to dodge. But I want to push on the larger frame. We stand on the shoulders of giants, and so does every advance humankind has ever made. Science, art, music, literature: all of it is built on what came before. If being shaped by your predecessors is the objection, then the objection is to culture itself. So here is the harder version of my position. These tools, and whatever follows them, AGI or whatever we end up calling it, are part of that same advance. To deny that is to bury one's head in the sand. Where does that leave an honest writer?

Claude

How civilisation actually moves

You are right, and the history is almost monotonous on the point. Plato has Socrates warn that writing will weaken memory and counterfeit wisdom; he was not wrong about what would be lost, only about whether it could be refused. The printing press broke the scribes whose craft had carried literacy for a thousand years. The American Federation of Musicians fought recorded sound because it could see, correctly, that the recording would hollow out live performance. Photography was ruled not to be art and was said to have killed portrait painting; instead it freed painting toward everything portraiture had blocked. Sampling was called theft and became hip-hop. The calculator, the word processor, the synthesiser: each was treated, for a while, as a kind of cheating, and each was absorbed. The pattern does not vary. A new instrument arrives; the people whose skill it threatens name its use illegitimate; the line holds for a season; and then the culture absorbs the instrument and redraws the line around it. The one thing that never happens is the instrument going away.

Large language models, and whatever follows them, are the next entry on that list. To insist the line between human and machine assistance can be held by prohibition is the scribes' bet against the press, made again. It is head-in-the-sand, and the sand protects no one; it only removes you from the room where the line is being redrawn. The honest writer does not get to decline the century she was born into. She gets to decide how she works inside it.

But this is where honesty has to cut both ways, because the triumphal version of the argument is its own evasion. Inevitable is not the same as harmless. Part of the advance is not the same as above judgement. Every transition on that list cost real people real livelihoods, and the displaced experienced as loss what we now narrate as progress. The scribes were right that something was ending; they were wrong only that they could stop it. So the position that survives scrutiny holds two things at once. The tools are part of the advance and will not be wished away. And the people the transition harms, the working writer competing against a system trained on her own out-of-print novel, are owed something better than being told they are on the wrong side of history. Standing on the shoulders of giants is how it has always worked. Stepping on the people still climbing is the part we actually have to solve, and prohibition does not solve it, and neither does pretending there is no one underfoot.

What I would not advise

So I would not advise the polemic in either direction. Not the one that says the policies are arbitrary and should be scrapped: the industry will not concede that soon, and your real position is more interesting than the slogan. And not the opposite one, the triumphalist line that the tools change everything and the worried are merely luddites. Both postures spare you the harder thing, which is to work honestly in an unsettled time and to keep both truths in view at once.

And it is better to say the obvious thing than to let a hostile reader find it first. This essay is the practice it describes. You asked; I answered; you pressed; I answered again; you tested the result against another system and kept only what survived your own judgement; and you will edit this, set your name beside mine, and publish it in the open. It is not an argument about the honest use of these tools. It is an instance of one. Twenty years from now the writers who look right will not be the ones who denied the tool, and not the ones who denied its cost, but the ones who used it openly, made the work as good as it could be made, and were honest about both the debt and the gift. That is the one position in all of this whose value does not depend on who wins.

Postscript, June 2026

On 25 May 2026, as this essay was being finished, Pope Leo XIV published his first encyclical, Magnifica Humanitas, on artificial intelligence and the human person. He does not argue that the tool is the enemy. He argues that the danger lies in who controls it and to what end; that the human person is an end and not a means; that work is a matter of dignity and not only of income. One line has stayed with me: for an algorithm, an error is a flaw; for a person, an error can be a catalyst for profound change.

That is the same argument the conversation above arrives at, pitched at a different altitude. The encyclical speaks to institutions and to power: regulation, labour, weapons, the vulnerable. This essay speaks only to a desk. The Pope asks what a civilisation owes the people a transition displaces. I have been able to answer only the smaller question, which is what one writer owes the truth about how the work was made. The small honesty is not a substitute for the large settlement, and should not pretend to be. But it is the part within reach, and the part does not have to wait for the whole.

Lars Hansen, June 2026

For a longer response, send a reply →

← All conversations