Who Is Holding the Pen?

It was the boxes of notes, the lost files, the half-remembered names

I have been carrying some of my stories around for decades. Not one story, either. A horror trilogy. A science-fiction novel. A supernatural thriller. Whole worlds that kept returning in fragments while life, work, and time did what they do. I had paper notes. I had computer files. I had stray ideas scattered through a half-dozen apps, some of which I no longer use and a few I cannot even open anymore. Characters changed names. Rules changed shape. Locations drifted. Somewhere in there were the stories I had always meant to write, buried under my own archaeology.

When I started using ChatGPT, the first real change was not that it wrote for me. It was that it helped me gather the mess. I could feed it scraps, old summaries, character sketches, half-finished scene ideas, and ask it to organize them into living world bibles. Not static documents that die the moment you close them. Working references. Single places where plot rules, setting details, timelines, character histories, and open questions could stay current as the projects evolved.

That may sound administrative, because it is. But fiction is full of administration. A horror trilogy needs rules. A science-fiction world needs political and technological logic. A supernatural thriller needs its own physics of dread. If a character learned something in book one, I need to know whether he still knows it in book three. AI has been strong at that kind of work: not inventing the soul of the story, but keeping the floor from shifting underneath it.

And that is where the public argument about AI and fiction often starts too late. People jump straight to the image of a machine spitting out a novel while a lazy person waits for a cover file. That exists. We have already seen the cheap flood: books that read assembled instead of written, manuscripts with prompt debris left inside them, genre packaging wrapped around prose that has never met a human pulse. But plenty of writers are using AI in quieter, more serious ways: as a research desk, a continuity reader, a sparring partner, a map of what they have already said and what they still need to solve.

The question is where it belongs, who is in charge when it arrives, and what part of the work we are willing to give away without admitting it.

Research, resistance, and remembering what you already built

The strongest case for AI-assisted fiction is not speed. Speed is seductive, and the least interesting part. The better case is access to friction that used to stop a writer cold. If I am writing a police officer, a trauma surgeon, a soldier, a boxer, or a person raised in a place I know only from research, I still have to make the character mine. But I do not need to spend three days wandering search results before I understand the basic professional pressures that would shape how that person enters a room.

That matters in fiction because bad research does not merely create factual errors. It creates false behavior. A doctor notices different things than a teacher. A person trained in Krav Maga does not move through a fight the same way a movie brawler does. Someone raised inside a military hierarchy may interpret silence, command, fatigue, and fear differently than someone who has never lived under that kind of pressure. AI can help me ask better questions before I write the scene. It can tell me where my assumptions are thin. It can suggest what I need to verify from better sources.

This is also where AI helps with roadblocks. Not the romantic version of writer's block, the blank page made dramatic. The ordinary version: I know what the scene needs to accomplish, but I cannot see the next three feet of hallway. In those moments I might ask for possible scene shapes, competing motivations, ways a character with a specific history might resist what the plot is asking of him. I am looking for the click that lets me keep going.

Other writers describe similar uses. K.M. Weiland, writing at Helping Writers Become Authors, frames the current moment as one of cautious practicality: AI can help with research, marketing, ideation, and other supporting work, while raising serious questions about originality and storytelling integrity.² Greenleaf Book Group makes a related distinction, granting that AI can help with brainstorming or getting unstuck while warning that AI-generated text becomes predictable and generic when it replaces the authorial act.³

That matches my experience. AI is much better at widening a doorway than deciding which room I should enter. It can give me ten possible reasons a character lies. I still have to know which lie hurts.

It can produce plausibility. Fiction needs pressure.

The best argument against using AI to write a novel is not that the prose is always terrible. Sometimes it is perfectly readable. That is part of the problem. Readable prose can disguise a dead center. It can move a character from grief to acceptance in the correct number of beats. It can end scenes with a little turn of insight. It can write dialogue that sounds like dialogue until you ask whether anyone in the room is truly risking anything by speaking.

Richard Lowe, writing for the Association of Ghostwriters, gets close to the heart of it: a novel does not fully exist before the writer goes through the work of discovering it.⁴ That has been true in every long project I have carried. The outline is not the book. The world bible is not the book. The book is what happens when all of that preparation hits a sentence that refuses to behave.

This is especially true in horror, which is where much of my own fiction lives. Horror is not a collection of dark props. It is duration. It is knowing when to keep the reader in the room after politeness says to leave. AI often retreats toward summary when a scene needs to remain physically, morally, or emotionally uncomfortable. It explains when it should observe. It tidies when it should let the stain spread.

That tidiness is not a small flaw. A lot of fiction works because of the uneven parts: the subplot that seems minor until it becomes the emotional spine, the chapter that runs longer because time has changed inside the story, the detail that returns eighty pages later with teeth. AI can imitate narrative order. It struggles with the private accumulation that lets a writer recognize an accident as a gift.

It also has no lived stake in the work. I know what my stories look like in my head. Only I know where the image is wrong on the page. Only I know when a character's answer feels borrowed, when a line sounds impressive but false, when a scene has all the required components and still has no blood in it. No model can be embarrassed by a bad draft. A writer can.

Assistance changes the conditions. Generation changes the origin.

Writers need a cleaner vocabulary than the one we are using. "AI writing" now covers everything from spell-checking a sentence to generating a 90,000-word manuscript from a chapter outline. That range is too wide to be honest. A writer asking for a list of 1920s slang terms is not doing the same thing as a publisher releasing an AI-written romance series under a fake name. A novelist asking whether a character's military background would alter his response to an ambush is not doing the same thing as asking a model to produce the ambush scene.

Amazon's Kindle Direct Publishing policy is useful because it is blunt. KDP says authors must disclose AI-generated text, images, or translations. It defines AI-generated content as material created by an AI-based tool, even if the author later edits it. But it does not require disclosure for AI-assisted work, including brainstorming, idea generation, editing, refinement, or error checking when the human created the content.¹ That is not a theory of art, but it is a practical line with consequences.

The U.S. Copyright Office draws the legal line in a related place. Its 2025 report says copyright protection can apply when a human author determines sufficient expressive elements, but not when prompts alone substitute for human authorship. AI assistance does not automatically poison a human-authored work; control over expression is what matters.⁵

That phrase, expressive choices, is the whole argument. The expressive choice is not only the adjective. It is what the scene includes, what it refuses, what it delays, whose perception governs the paragraph, how much of the violence is shown, whether a death happens on the page or behind a closed door, whether a character says the thing or fails to. If AI supplies those decisions and I merely sand them down, I am no longer using assistance in any meaningful creative sense. I am supervising output.

Disclosure, suspicion, and the coming fight over labels

The publishing ecosystem is not waiting for writers to finish their philosophical debate. Platforms are drawing administrative lines. Readers are developing instincts. Editors and agents are learning the shapes of AI prose. And AI companies are beginning to mark their own outputs, sometimes in ways that make writers nervous.

That last point is no longer theoretical. Recent reporting on Anthropic's Claude describes invisible text watermarking tied to EU transparency rules, designed to mark AI-generated content through subtle statistical patterns in word choice.⁶ The controversy is obvious. Transparency matters, especially in schools, journalism, publishing, and any setting where authorship is part of the promise. But writers who use AI for light editing or idea work worry about being branded as if they outsourced the whole act. The label may not understand the workflow.

This is why the divide between AI-assisted and AI-generated writing cannot remain vague. If a model corrects a typo, that is not authorship. If it rewrites a paragraph, maybe it depends how much survives. If it drafts the chapter and the human revises it, Amazon still calls the underlying text AI-generated. Readers may or may not care. Publishers certainly might. Copyright offices will care in their own way. The writer should care most of all, because the wrong workflow can leave you defending a book you did not quite write.

There is also an ethical question beneath the policy question. Some critics focus on training data and whether authors' work was absorbed without consent or compensation. Others focus on the future labor market, where publishers can use synthetic prose to drive down the value of human writers. Still others focus on the writer's own development. If a person never learns to solve a broken scene, but learns only to request a replacement, what skill is being built?

The software is moving toward the manuscript itself

The next stage of AI writing will not look like a blank chatbot window. It will live inside the places writers already organize their manuscripts. I use LivingWriter, which now markets itself as an AI writing app for authors, screenwriters, manuscripts, and documents. Its feature set is telling: outlines, manuscript chat, summaries, analysis, rewrites, element generation, boards for planning and research.⁷ Other fiction-centered platforms, including Sudowrite and Novelcrafter, aim directly at story bibles, character context, scene expansion, and long-form fiction workflows.⁸

This integration is powerful because it reduces the distance between the manuscript and the assistant. It is also risky for the same reason. When the system understands your characters, your outline, your tone, and your plot arcs, the invitation to let it continue the story becomes constant.

For me, the best version of this future is not a machine that writes my trilogy. It is a workspace that remembers every rule I have made, catches contradictions before they metastasize, helps me research the professional and historical details I need, lets me interrogate a character's logic, and then gets out of the sentence. I want help with the scaffolding. I do not want a substitute for the private, maddening, necessary act of making the thing sound like it came from me.

That distinction may sound precious until you have carried a story for twenty years. Then it becomes practical. I am not writing because I need another revenue stream. I have a fulfilling job and a life I value. I am writing because these stories keep living in my head, and I want to reproduce them on the page as accurately as I can. That is a strange compulsion, but it is the one I trust. AI helps me move faster through the parts that used to bury me. It does not know what I am trying to rescue.

A novel is not a content problem

The phrase "content" has done real damage to how we talk about writing. Content fills slots. Fiction does something more particular and less obedient. It creates an experience of consciousness under pressure. It asks a reader to inhabit a pattern of attention that belongs to someone else. Whatever that pattern is, it has to be chosen.

AI can help a writer reach the choosing faster. I believe that because it has helped me. It has made old projects feel newly possible. It has turned decades of scattered notes into usable systems. It has helped me think through world-building, background research, plot logic, character histories, and the small practical details that make invented lives feel less generic. I am cautiously pro-AI because I know what it has opened for me.

But my optimism has a boundary. I do not want to produce books that merely satisfy the outward requirements of books. I do not want a horror trilogy that has horror imagery but no real nerve. I do not want a science-fiction novel that understands the machinery but not the ache. I do not want a supernatural thriller whose plot fits together because the software dislikes loose ends. I want the story I have been carrying, including the inconvenient parts, the wrong turns, the obsessions I cannot fully justify until I write them.

The divide between writing with AI and letting AI write with your supervision is not one bright line. It is a series of smaller decisions made over and over. Who decides what the scene is really about? Who decides where the paragraph turns? Who decides what to leave unresolved?

If the answer is still the writer, then AI can be a powerful addition to the desk. If the answer is the machine, even politely supervised, then the byline starts to mean something else.

I am not interested in purity. I am interested in ownership. Not the legal kind, though that matters. The creative kind. The private knowledge that the thing on the page is the closest I could get to the thing I saw, feared, remembered, invented, and chased. AI can help me carry the maps, sharpen the questions, and keep the names straight. But when the scene finally opens, I still want to be the one walking in first.