Permitted Assistance
The case currently being made for AI use by authors
The dominant story about artificial intelligence and books is a story of scandal. A self-published horror novel, Shy Girl, was pulled by Hachette after readers and an AI-detection tool flagged it as largely machine-generated. A New York Times freelancer was let go after a book review was found to have lifted, via an AI tool, the work of a Guardian critic. A Commonwealth short-story prize and a Granta competition both became embroiled in accusation and counter-accusation. Roughly 10,000 authors, including Kazuo Ishiguro and Richard Osman, co-published an “empty” book in protest at AI firms training on their work without consent.
Against this backdrop, it would be easy to assume that no credible support exists for AI use in authorship at all. That isn’t true. A growing body of opinion — from bestselling novelists, working editors, professional bodies, philosophically-minded essayists, and even cautious publishing executives — argues that AI has a legitimate and defensible place in the writing process. The argument is rarely “AI should write books.” It is almost always narrower and more specific than that: AI is being defended as an instrument the author directs, not as a replacement for the author. The crucial fault line in nearly every piece of supportive commentary runs between AI as tool — for research, editing, feedback, structuring, accessibility — and AI as author, generating the prose itself.
This essay sets out the case currently being made for the former, organised by the kind of work AI is being asked to do.
The historical argument: writing was never solitary
One of the more developed defences of AI-assisted writing doesn’t start from technology at all — it starts from literary history. Writing in The Times, the columnist Marc Carter argues that the romantic idea of the solitary author has always been a fiction. His central example is T. S. Eliot’s The Waste Land, whose surviving manuscripts show extensive cutting and rearranging by Ezra Pound, with further input from Vivienne Eliot and other readers. Nobody, Carter points out, argues that this collaboration makes the poem any less Eliot’s. He extends the point to the ordinary infrastructure of publishing — editors, agents, fact-checkers, researchers, trusted early readers — none of whom are thought to compromise an author’s claim to a book.
Carter’s argument is that the real question was never “was anyone else involved?” but “what kind of involvement was it, and who exercised final creative judgment?” On that framing, AI is simply a new entrant into a role — research help, developmental feedback, accessibility support — that other people have occupied for as long as books have been edited and published. He adds an accessibility dimension to this: the article opens with the case of Jamir Nazir, a prize-winning short-story writer who used speech-to-text software because illness made typing difficult, and whose entirely human-authored work was nonetheless flagged by detection software as 100% AI-generated. Carter also discusses his own daughter, who has learning differences and uses AI tools to participate in writing at all. His broader claim is that AI assistance may do for under-resourced or differently-abled writers what money has long done for well-connected ones: buy editorial attention they could not otherwise access.
It’s worth noting the limits of the Eliot analogy even on its own terms — critics still debate exactly where Pound’s editorial intervention shaded into something closer to co-authorship — but the comparison has done real work in reframing the debate away from a binary of “tainted” versus “pure” and toward a question of degree and disclosure.
The muse tradition: an older account of where creativity comes from
A different kind of defence — less historical precedent, more first principles — is one I make myself, in an essay called “Daughters of Memory: On Muses, AI, and the Creative Life,” published at my newsletter, The Contemplative Writer. My argument there is that the discomfort many writers feel about AI collaboration is a recent and culturally specific inheritance, not a timeless truth about what creativity is.
My starting point there is the Greek tradition of the Muses, who were not, in my reading, decorative figures but the believed source of poetic inspiration: when Hesiod opened the Theogony by invoking the Heliconian Muses, he was making, as I put it, “a sincere metaphysical claim: that what flows through the poet does not originate in the poet.” I trace the Muses’ parentage — daughters of Zeus and Mnemosyne, the personification of Memory — and argue that a large language model occupies structurally the same role: a distillation of an immense breadth of human language and thought that “does not generate from nothing,” but channels what has already been said and written, offering it back “transformed and particular, in response to the specific need of the specific moment.” If the Muses were daughters of Memory, I write there, “then AI is, in this sense, Memory’s most recent child.”
The more pointed move in that essay is what I set against this lineage: the Romantic figure of the solitary genius, whose authenticity is measured by how little he owes to anything outside himself. I locate the modern discomfort with AI collaboration in that specific eighteenth- and nineteenth-century inheritance rather than in anything intrinsic to the technology, arguing it has hardened into “an unexamined moral assumption” — that originality equals autonomy and that “to receive is to diminish.”
Like every other voice in this piece, however, I’m careful in that essay to bound the analogy rather than let it run loose: “The Muses did not write the poems. They whispered; the poet still had to listen, discern, shape, and labour.” My description there of my own collaboration lands in almost exactly the same place as the working novelists discussed above, just reached by a different route: “What comes back from that collaboration is not a finished text… The formed vision… is mine. The contemplative grounding from which the writing proceeds is mine. The judgment about what to keep, what to discard, what rings true and what rings hollow: all mine.”
What that essay adds to the case here isn’t a new category of use so much as a different order of justification for the ones already established. Horowitz and Chowdhury defend their practice on practical grounds — it’s useful, it saves time, it’s what a good editor does anyway. My own defence rests on the same basic posture — AI as interlocutor, never as author — but on the grounds that receiving has always been part of how creative work happens, and that the burden of proof sits with the Romantic assumption that it shouldn’t be, not with the writers who collaborate.
AI as a research tool
The least contested use of AI among authors who discuss it openly is research — and it is here that the supportive case is strongest.
Anthony Horowitz, the bestselling author of the Alex Rider series, has said he uses ChatGPT “all the time,” but is careful to specify what that means in practice: “Strictly research. I would never put two words from AI next to one another.” For a forthcoming historical novel set partly at Croydon Aerodrome in 1933, he describes asking AI for period detail — what the airport was called, what a passenger arriving from abroad would have encountered — facts he says he cross-checks against other sources before use, much as he would information from a library book. Horowitz is explicit that this isn’t a small caveat: “It would be crazy to give up an incredible resource.”
A more developed example comes from the journalist Katie Prescott, who used Google’s NotebookLM while researching her biography of the tech entrepreneur Mike Lynch. Facing roughly 2,000 pages of accounting judgments and trial transcripts, Prescott loaded her source material into the tool and used it to query and cross-reference documents she could not realistically have read end to end under deadline. It surfaced material she says she would otherwise have missed entirely — passing references, buried across several transcripts, to an odd internal presentation that became a small but telling detail in her book. Reviewers subsequently described her research as “meticulous,” and lawyers involved in the original case were reportedly struck by how quickly she synthesised material that had taken them years to organise. Prescott is careful to draw a boundary around this use: “It’s a tool. It’s not going to do my job for me. You can only [write a book like this] by speaking to people… the stuff that AI can’t do.”
This division — AI for retrieval and synthesis, human judgment for everything that depends on relationships, trust, and interpretation — mirrors how research-heavy fields outside trade publishing have begun to formalise AI use. Academic and scientific publishers, including Elsevier and the American Psychological Association, now permit AI use for literature review and background research (often requiring disclosure) while explicitly barring AI from being credited as an author, on the grounds that authorship implies responsibilities only a human can hold. The same structural distinction — AI as a research instrument, never as a byline — recurs across very different corners of the writing world.
AI as editor and sounding board
A second, slightly more contested category of support concerns AI’s use in shaping and refining work that the author has written.
The crime novelist Ajay Chowdhury, who began experimenting with AI as early as 2023, describes ChatGPT as functioning like “an editor on demand.” He uses it the way he might use a real editor — to talk through a plot problem, propose alternative directions for a scene, or help restructure a chapter for pacing — while estimating that he discards the great majority of what it suggests. He is careful to phrase requests as genuine asks for critique, aware that chatbots have a tendency toward flattering responses, and notes that the feedback he gets is often in line with what his editor at Penguin tells him — “though my real editor is obviously better.” Chowdhury is unambiguous that he does not use AI to write his prose: “I don’t think it writes very well, and… what’s the point?”
This “sounding board” framing has institutional backing. The Authors Guild’s updated AI Best Practices for Writers, published in 2026, explicitly distinguishes “background uses” of AI — research, acting as a sounding board, or fine-tuning a model on an author’s own previous work — from uses that risk compromising a writer’s creative voice or raise authorship and copyright concerns. Wiley, one of the largest academic and professional publishers, has issued similar guidance encouraging authors to use AI tools in manuscript preparation in ways that preserve their authentic voice rather than substitute for it. The throughline in both is the same one Chowdhury describes in practice: AI as interlocutor, not as ghostwriter.
A more unusual example of this category comes from a working novelist’s internal style guide, “Keeping Our Writing Human,” built directly from a research paper that trained a classifier to distinguish human from AI fiction with over 93% accuracy using narrative structure alone — entirely independent of surface-level tics like em-dashes. Rather than treating that research as a threat, the document repurposes it as a craft tool: a checklist of tendencies AI defaults to (tidy plots, over-explained themes, emotion conveyed through bodily sensation rather than named feeling, vague rather than specific cultural references) that a writer collaborating with AI can deliberately watch for and resist during revision. Used this way, AI’s own statistical fingerprints become diagnostic material an author can turn back on the work to protect what is most distinctly theirs — arguably one of the more sophisticated versions of “AI as editor” currently in circulation, since it uses AI-derived insight specifically to safeguard human authorship rather than to imitate or replace it.
AI as an accessibility tool
Running through several of these accounts is a quieter, less headline-grabbing argument: that AI lowers the cost of entry to functions writers have always needed — research assistance, developmental feedback, basic drafting support — that have historically been available mainly to those with money, connections, or institutional backing. Carter’s article frames this most directly, citing both the speech-to-text writer wrongly accused of AI use because illness prevented him from typing, and his own daughter’s use of AI tools to work around learning differences. The argument here isn’t that AI improves on what a well-resourced author could already get from a paid editor or research assistant, but that it extends a version of that support to writers who could not otherwise afford it.
Market and institutional acceptance
Even among gatekeepers with obvious commercial reasons for caution, outright rejection of AI is not the dominant position. James Daunt, CEO of Barnes & Noble, has said he would be willing to stock AI-written books on the same shelves as everything else, on one condition: that they are clearly labelled as such and don’t “masquerade” as conventionally human-authored work. His objection is to deception, not to AI involvement per se, and he has noted that some of the 300,000 titles already on Barnes & Noble’s shelves are probably AI-assisted to some degree without anyone being especially conscious of it.
The Authors Guild’s Human Authored Certification programme — which lets authors apply a registered mark verifying their book was written without more than incidental AI use (permitting basic grammar and spell-check tools) — might seem, at first glance, like evidence of pure resistance. But its existence actually demonstrates the opposite of a blanket ban: an industry body building voluntary, disclosure-based infrastructure that allows AI-free and AI-assisted books to coexist on the same shelves, distinguished by transparent labelling rather than prohibition. That is the same logic Daunt describes from the retail side — the issue is concealment, not assistance.
Where the supportive case draws its own line
What’s notable across nearly all of this material is how consistently its own advocates police the boundary themselves. Horowitz, who calls AI “a wonderful friend… albeit a dangerous one,” still insists he would never let its phrasing reach the page. Prescott calls it a tool that “won’t do my job for me.” Chowdhury discards most of what it offers. I draw the same line myself, for all the theological framing: “The Muses did not write the poems.” The Authors Guild’s guidance protects “background” uses while flagging the legal and ethical risks of anything closer to generation. Even Olga Tokarczuk, who described asking AI questions about her characters’ tastes in music, issued a clarifying statement after backlash insisting the use was confined to research.
The pattern suggests that the support currently available for AI in authorship is real, but narrower and more conditional than either AI’s harshest critics or its more breathless promoters tend to suggest. It is support for AI as research assistant, editor-on-demand, accessibility aid, and structural sounding board — roles long occupied by humans, now partly supplemented by software. It is markedly thinner, even among the same advocates, for AI as the source of the actual sentences a reader will encounter. The Anthony Horowitz who calls AI “an incredible resource” is the same Anthony Horowitz who will not put two of its words next to each other. That distinction, rather than any blanket endorsement, is where the genuine and growing body of support for AI-assisted writing currently sits.
The essay referenced above, “Daughters of Memory: On Muses, AI, and the Creative Life,” is available to read at The Contemplative Writer.