Where's the Line? A Working Framework for Writers Thinking About AI

A Moment Worth Paying Attention To

In March 2026, Hachette Book Group discontinued a horror novel called Shy Girl by Mia Ballard following an investigation into suspected AI use. The book had been released in the UK in November 2025 and was scheduled for a North American release this spring. It was the first known instance of a Big Five publisher walking back a book over AI concerns. Ballard has said she did not personally use AI and that a freelance editor may have introduced AI-generated text into the self-published version without her knowledge. She is pursuing legal action.

Whatever the eventual resolution, the case has prompted useful reflection across the writing community—not about Ballard specifically, but about the larger question it raised. And that question isn't really "Is AI good or bad for fiction?" It's something more practical:

Where does "using AI" end and "letting AI write the book" begin?

If you write fiction in 2026, this is worth thinking through before you have to answer it for someone else.

The Landscape, Honestly

A few things are true at once, and holding all of them is part of thinking clearly about this.

First, AI use among working writers is more common than some public conversations suggest—and less uniform than others claim. A 2024 Authors Guild poll found that 13% of its members reported using AI tools. The UK Society of Authors' 2024 survey put the figure at around 22%. A more recent 2025 study by Gotham Ghostwriters, which focused specifically on fiction authors, found that 42% of fiction writers use AI at least sometimes. The numbers vary because definitions of "use" vary—everything from spell-check to full prose generation.

Second, the industry is building infrastructure to respond. The Authors Guild launched a "Human Authored" certification mark in January 2025, which expanded to all US authors in March 2026. The mark can be placed on book covers to indicate the text was written by a human. It permits AI use for research, brainstorming, and outlining, but not for generating the prose itself. The UK's Society of Authors partners on the program, and a separate UK initiative called Books by People offers something similar.

Third, What readers often object to is not tool use in the abstract, but the sense that a book’s making has been obscured. But when they discover AI was used without disclosure, a sense of deception enters the picture that's hard to undo.

These three facts, taken together, describe a field that's still finding its footing. Which is why a framework helps.

A Spectrum, Not a Switch

The most common mistake in AI-and-fiction conversations is treating it as binary: either you used AI or you didn't. In practice, AI use sits on a spectrum, and most of the useful conversation happens in the middle.

Two points about this spectrum worth holding onto.

One: the community has rough consensus at the top and bottom of the table, and essentially zero consensus in the middle rows. A writer using AI to generate character names and a writer using AI to draft scenes are doing very different things, but both can honestly say "I used AI." Arguments online routinely conflate these.

Two: your position on this spectrum is a decision, not a default. Most writers haven't consciously chosen where they want to operate. It's worth the half hour of thought.

What Readers Seem to Notice

One of the more interesting aspects of the recent discussion has been readers articulating, often for the first time, what AI prose tends to feel like. Writing in the Boston Globe, Harvard student and writer Zoe Yu offered a line that's been widely shared: it writes the way a mirror talks—it knows shapes, not souls.

The more concrete patterns readers seem to catch include:

• Metaphor pairing. A tendency to stack two metaphors where a human writer would commit to one.

• Adjective accumulation. Three descriptors where one would do the job.

• A consistent "literary" register. Even mundane moments get treated at the same elevated pitch.

• Emotional over-signaling. Telling, showing, then telling again.

• Neat closing images. Scenes ending on a profound note rather than a messy human beat.

None of these are moral failings of the technology. They're byproducts of how these models were trained—on corpora where literary-sounding writing often got rewarded. A writer who uses AI at the prose level without cutting aggressively can end up with sentences that feel competent but rarely surprise. And surprise is much of why readers reach for fiction in the first place.

A Test That's Been Useful

Here's a practical question we keep coming back to when writers ask where their own line should be:

Am I using AI to help me discover what I think, or to skip the discovery?

Writing is, among other things, a process of finding out what you actually believe about something—what a character really wants, what a story is really about. In Good Prose, Tracy Kidder and Richard Todd describe this directly: the heart of a story is usually a place to arrive at, not a place to begin.

• AI used to accelerate discovery—to generate the first ten obvious options so you can find the eleventh, to pressure-test a plot until its weak points reveal themselves—is doing something writers have always done with notebooks, index cards, and long conversations with patient friends.

• AI used to bypass discovery—to produce the scene so you don't have to find it—is doing something qualitatively different. Not necessarily wrong. But different. And the result is no longer, in the traditional sense, your writing. It's closer to something you commissioned.

Readers can often feel that difference. Not always. But often enough that the distinction matters.

Practices That Tend to Serve the Work

If you do want to work with AI in your fiction—and many thoughtful writers do—here are habits that seem to serve the craft rather than erode it:

• Use AI before the draft, not during it. Brainstorm, pressure-test, outline. Then close the tab and write the scene yourself.

• Resist pasting AI prose directly. If a suggestion is useful, retype it in your own words. This small friction keeps your voice in the sentence.

• Let AI help with the parts you find tedious, not the parts you love. If outlining drains you but dialogue energizes you, use AI for the first and protect the second.

• Keep process notes. Documentation of your creative decisions protects your copyright position and, quietly, makes you a more deliberate writer.

• Decide your disclosure approach before you publish. Whether you choose to disclose AI use or not, know what you'd say if asked. Being caught flat-footed is almost always worse than the answer itself.

A Final Thought

The Shy Girl case won't be the last of its kind. Detection tools will improve, then fail in new ways. Certification marks will spread, then be tested. The community will keep negotiating where the line is.

But the line itself isn't really about AI. It's about the question fiction has always asked: did a human being bring something of themselves to this?

That question predates ChatGPT by several thousand years. The tools are new; the underlying commitment isn't.

If you're working on a novel or a script right now, the most useful thing isn't picking a side in the AI debate. It's deciding where on the spectrum you want to operate, and then working there with intention. Not defensively. With clarity about what you're trying to make.

At Novela, we’ve tried to design AI around the phases of writing where assistance sharpens a writer’s own thinking, rather than replacing the act of writing itself. Our tools are built to help you pressure-test a plot, develop a character's psychology, or break through the "what happens next" wall, without writing the scene for you.

The scene, after all, is yours. That's where the story actually lives.