AI Can Write. But Can It Be a Writer?

Samo Rensly
1 day ago
8 min read

Generative AI can produce beautiful sentences, compelling scenes and even entire manuscripts. But as more writers hand the act of writing itself to machines, I keep coming back to a harder question: what happens to the writer?

With the growing rise of artificial intelligence, we are seeing more literary work—novels, screenplays, essays and everything in between—created with some degree of AI involvement. That involvement can mean anything from checking grammar to generating entire chapters, and I think that difference matters more than almost anything else in this conversation.

I am a columnist and a novelist, and I work directly with a literary talent agency. When people send queries, AI-heavy writing is often easier to notice than writers think. There are patterns in the language, in the rhythm, in the way a paragraph explains itself, and sometimes in the strange absence of a truly individual voice. We use screening tools as part of the process, but no detector should be treated as an infallible truth machine. The larger point is that people who read manuscripts for a living become familiar with what writing sounds like when a person is actually wrestling with the page.

That has gotten me thinking lately about an earlier technological shift. When computers became common and authors put down the pen and paper—or retired the typewriter—was there a similar period when publishers and agents looked down their noses at would-be authors who composed on a screen? Did people say the computer was making writing less authentic simply because the machine had replaced the ribbon, the ink and the legal pad?

I imagine some did. Every new tool makes the old guard nervous.

But artificial intelligence is different.

A word processor changed the instrument. It did not supply the sentence. Spell-check could underline the mistake, but it did not invent the scene. A thesaurus could offer five alternatives, but the author still had to decide whether the character would whisper, mutter, hiss, answer or say nothing at all. Generative AI can now participate in the language itself. That is a much larger line to cross.

The idea may still begin in a human brain. The character may still be ours. The plot may still be something we dreamed up while driving home, lying awake at two in the morning or staring at the ceiling when we should have been working. But when the machine begins supplying the prose, what we are no longer receiving in the same way is the expansion of literary knowledge that comes from the writer having to solve the sentence.

The publishing world is already drawing distinctions here. Amazon KDP’s current content guidelines distinguish between AI-generated content and AI-assisted content. Amazon requires disclosure of AI-generated text, images or translations, while it does not require disclosure when an author created the material and used AI only to edit, refine, error-check or brainstorm. That distinction is useful because the question is no longer simply whether AI touched a manuscript. The question is what the AI actually did.

The law is drawing a similar boundary around human authorship. In 2025, the U.S. Copyright Office concluded that generative-AI output can receive copyright protection only where a human author has determined sufficient expressive elements. Merely providing prompts is not enough, while using AI as an assistive tool does not automatically prevent copyright protection. In other words, even the legal system is being forced to ask an old artistic question in a new form: who actually made this?

Beautiful Is Not the Same as Authored

Can AI-generated art be beautiful? Of course it can. Can AI-generated words have impact? Absolutely. With each iteration improving dramatically, that ability is only going to increase. There will be AI-written paragraphs that make people laugh. There will be generated images people genuinely want on their walls. There will be scenes that are technically clean, emotionally legible and commercially effective.

I do not think denying that helps anybody.

But there is still a difference between traditional writing and generated writing, and the difference is not merely nostalgia. It is process. It is authorship. It is the mental work that takes place between having an idea and finding the exact words capable of carrying it.

My agency would not publish an AI-generated novel. We have several ways of screening manuscripts and we rely heavily on editorial judgment. But lately the question has become less black-and-white and much more gray: how much AI was actually used in the creation? Was it a spell-checker? A research assistant? A plot organizer? Did it help the author remember that a character had blue eyes in Chapter Three? Or did it write the scene? Did it create the voice? Did it decide how grief sounded coming out of that character’s mouth?

To me, the acceptable use of AI in literary writing is on the organizational side of the desk. Let it help keep track of plot, characters, timelines, structure and continuity. Let it function like a wall full of index cards. But I do not want it writing the novel, and I do not want it establishing the tone. The prose should still belong to the person whose name appears on the cover.

When I cannot find the word I want, I still flip open a thesaurus. When I am trying to make a word sound more intelligent, I do the same thing. That little act matters to me. I have to look at the alternatives. I have to understand the differences between them. I have to choose. The final word may be better, but more importantly, I had to get there.

What Happens to the Writer’s Brain?

This may be the part of the AI conversation that interests me most. We tend to debate whether the finished product is good enough. We spend less time asking what happens to the person when the process that used to require mental effort becomes something we can outsource in seconds.

There is no single part of the brain that “writes novels.” Creative thought appears to emerge from cooperation among several systems. Research in creativity neuroscience repeatedly points to interaction between the brain’s default mode network, associated with internally generated thought, imagination and memory, and executive-control systems involved in focusing, evaluating and refining ideas.

A large multi-center study involving 2,433 participants found that creative ability could be predicted in part by dynamic switching between default-mode and executive-control networks, with the best creative performance associated with a balance between the two. The study is available through the National Library of Medicine.

Other neuroscience research has found a direct role for the default mode network in creative thinking and describes creativity as a collaboration between spontaneous idea generation and deliberate evaluation. Research published in Brain in 2024 adds to that evidence.

That matters because writing is not only an act of transcription. The struggle is part of the work. Remembering what came before, imagining what could happen next, rejecting the obvious sentence, finding the better image, deciding that the clever line is wrong for the character—those are not inconveniences surrounding writing. They are writing.

The Convenience Problem

Many of us already use AI to write emails we do not want to write, to clean up a message, or to make ourselves sound a little smarter than we feel that afternoon. I have done it. And what I have noticed is uncomfortable: after doing it enough, my own writing can start to feel lazy.

AI is doing its job perfectly. The problem is that I am no longer challenging my own brain to do mine.

There is early research that makes this concern worth taking seriously. A 2025 MIT Media Lab study followed people completing essay-writing tasks either with an LLM, with a search engine or without external tools. Among 54 participants in the first three sessions, the brain-only group showed the strongest and most distributed EEG connectivity, the search-engine group was in the middle, and the LLM group showed the weakest connectivity. The LLM group also reported less ownership of its essays and had more difficulty accurately recalling or quoting what it had written.

Those results are provocative, but they should not be exaggerated. The study was relatively small and was released as a preprint, and later researchers published a methodological critique arguing that several findings should be interpreted more conservatively. So I am not saying that using ChatGPT “damages your brain.” The evidence does not justify that claim. What the study does give us is a reason to ask whether repeatedly outsourcing difficult cognitive work changes how deeply we engage with that work.

That question should matter enormously to writers.

Writing is repetition. It is practice. It is the accumulation of thousands of decisions that slowly become instinct. A novelist does not develop a voice by selecting “make this sound literary” from a menu. Voice develops because a person writes a bad sentence, hears why it is bad, fixes it, writes another one, reads better writers, fails again, and eventually develops an internal sense of rhythm that is difficult to explain but easy to recognize.

A Dying Breed—or a More Valuable One?

I worry that truly individual writers could become a dying breed: artists who build completely unique ideas through the strange cooperation of memory, imagination, language and judgment inside a human brain, and who then have the patience to drag those ideas into words themselves.

At the same time, I wonder whether the opposite may also happen.

If machine-generated prose becomes endless, inexpensive and technically competent, then unmistakably human writing may become more valuable rather than less. The flaws may matter. The strange sentence may matter. The regional phrase, the uncomfortable opinion, the memory nobody else could have had, the joke that only works because the writer’s grandfather used to say something peculiar at the dinner table—those things may become signals that a person is actually there.

The writing profession is already responding in that direction. The Authors Guild’s Human Authored certification allows authors to identify books whose text was written by humans, while permitting minimal uses such as spelling and grammar tools or research. The Guild has also published model contract language addressing how much AI-generated text may appear in manuscripts and how publishers may use AI.

I find that encouraging, not because I think AI is going away—it is not—but because it suggests we are beginning to define the difference between using a tool and surrendering a craft.

The Future of Writing Still Needs Writers

What does all of this mean for the future of writing? I do not think it means novels disappear. I do not think screenwriters disappear. I do not think every book published twenty years from now will secretly have been written by a machine.

I think the future will be messier than that.

AI will become woven into the mechanics of creative work. Writers will use it for research, organization, continuity, translation questions, brainstorming and tasks we have not imagined yet. Some will use it to generate prose and call the result their own. Publishers, agencies and readers will continue arguing over where the boundary belongs.

But I am hopeful that great authors will still emerge—people who have something to say badly enough that they are willing to do the difficult part themselves.

Because the real threat is not that artificial intelligence will learn to write beautifully. I think it will.

The threat is that human beings may stop practicing how to.