Clara Voss and I are tucked into the back corner of a bookstore café in Southeast Portland, the kind of spot where the espresso machine hisses louder than the customers speak. She’s ordered a pot of chamomile tea and holds the cup with both hands, as if warming herself against something. We’re here to talk about her fourth novel—a book that, by her own account, almost didn’t exist. Not because of writer’s block, but because of a more unsettling problem: she had become too good at being Clara Voss.
“I could feel myself writing the same sentence shapes,” she says. “The same emotional arcs. The same kind of character—intelligent, wounded, a little ironic. I’d open a draft and know, within three paragraphs, exactly where it was going. That’s not mastery. That’s a cage.”
Voss is the author of three literary novels, each praised for its psychological acuity and prose that critics have called “luminous” and “restrained.” But after her third book, she stalled. The problem wasn’t a lack of ideas; it was a surplus of habit. She had developed what she calls “aesthetic autopilot”—a set of deeply ingrained stylistic reflexes that made every new page feel like a variation on a theme she had already exhausted.
So she did something that surprised her agent, her editor, and eventually herself: she fed her completed draft into an AI story generator. Not to write for her. Not to generate prose. But to sabotage her own patterns.
The Draft as Patient, the Machine as Diagnosis
Let me be precise about what Voss did, because the details matter. She had a full draft of her fourth novel—380 pages, polished to the point where her workshop readers were calling it “finished.” But she knew it wasn’t. The book felt competent, even beautiful in places, but it lacked the one quality she couldn’t manufacture: surprise. The characters were behaving exactly as she had designed them to behave. The plot was resolving along lines she had drawn months earlier. Nothing in the manuscript seemed to have a life of its own.
“I remembered reading about the Oulipo group,” she says, referring to the French collective of writers and mathematicians who used constraints—writing entire novels without the letter ‘e,’ for instance—to force language into new shapes. “The idea was that freedom doesn’t produce originality. Constraint does. But I’d already tried every constraint I could think of. I’d written scenes in second person. I’d imposed word counts. I’d reversed chronology. None of it worked, because I was still the one designing the constraint. I needed a constraint I couldn’t anticipate.”
This is where the AI entered. Voss began using an AI story generator that fits the draft workflow—a tool she describes as “Unsloppy AI” in her notes—not to generate new text, but to generate alternative possibilities. She would take a scene she had already written, strip it down to its structural skeleton (who is present, what is the conflict, what is the setting), and ask the tool to produce three alternative versions of what could happen next. Not better versions. Just different ones.
“The first time I did it, I felt almost physically ill,” she says. “The machine gave me a plot branch where my protagonist—a woman I’d spent two years making sympathetic—did something genuinely cruel. Not villainous, just… casually unkind. And I realized I had been protecting her. I’d been keeping her likable because I was afraid readers wouldn’t stay with her otherwise. The AI didn’t care about likability. It just followed the logic of the situation.”
She pauses, rotating the teacup in her palms. “That was the moment I understood what I was actually doing. I wasn’t outsourcing creativity. I was outsourcing my own blind spots.”
The Uncomfortable Gift of the Unwanted Suggestion
Over the course of six months, Voss developed a method. She would feed the AI a scene, a character description, or a fragment of dialogue, and ask for alternatives she would never have considered. The machine’s outputs were, by her estimate, about 80 percent unusable—clichéd, tonally wrong, or simply nonsensical. But the remaining 20 percent contained something she couldn’t generate on her own: genuine otherness.
“There’s a scene in the novel where my protagonist, Elena, is confronting her mother about a childhood betrayal,” Voss says. “In my original draft, the mother deflects, then breaks down, then asks for forgiveness. It was a good scene. It was also the same scene I’d written in my second novel, just with different furniture. The AI suggested something else: the mother doesn’t deflect. She agrees. She says, ‘Yes, I did that. I’d do it again.’ And then she leaves the room. No tears. No reconciliation. Just a door closing.”
Voss kept that version. Not because the AI wrote it well—she rewrote every sentence—but because the AI proposed a psychological possibility she had been unwilling to imagine. Her own mother, she tells me, was a complicated woman, and Voss had spent years in therapy learning to forgive her. That personal work had bled into her fiction, making every maternal conflict resolve toward empathy. The machine, having no biography, no therapy, no need to forgive anyone, offered a colder logic.
“I had to ask myself: is this true to the character? And the answer was yes. It was truer than what I’d written. I just hadn’t been able to see it because my own life was in the way.”
This is the core of Voss’s argument, and it’s worth sitting with. The AI didn’t teach her anything about craft. It didn’t improve her sentences. What it did was expose the places where her own psychology had become a stylistic tic. The machine, by having no self, revealed the self she had been hiding behind.
The Line Between Inspiration and Authorship
At this point in the conversation, I ask the question that has been hovering since we sat down: where is the line? If she is using AI-generated suggestions to reshape her novel’s emotional architecture, who is the author of the resulting book?
Voss doesn’t flinch. “I am. Unequivocally. The AI didn’t write a single sentence that appears in the final manuscript. It proposed possibilities. I evaluated them. I rejected most. I rewrote the ones I kept. The machine was a collaborator in the way a random number generator is a collaborator for a composer like John Cage—it introduced chance, but the artistic decisions were all mine.”
She points me toward the Authors Guild’s AI Best Practices for Authors, which she consulted before beginning the experiment. “They make a distinction that I think is crucial: using AI as a tool within your creative process is different from using AI to generate the work itself. I never crossed that line. The machine never held the pen.”
Still, the discomfort is real. Voss describes a period of several weeks where she couldn’t look at her manuscript without feeling a kind of vertigo. “I’d read a scene and not remember whether the emotional beat came from my original draft or from an AI suggestion I’d rewritten. The provenance got blurry. That was terrifying. But it was also productive. It forced me to re-examine every choice, to ask whether it was mine or just… available.”
This blurriness, she argues, is not a bug but a feature of the method. The goal was to disrupt her own aesthetic patterns, and disruption is inherently disorienting. “If you’re not uncomfortable, you’re not breaking anything,” she says.
The Oulipo Connection: Constraint as Liberation
To understand what Voss was attempting, it helps to place her experiment in a longer tradition of writers using external constraints to bypass their own habits. The Oulipo group—founded in 1960 by Raymond Queneau and François Le Lionnais—treated constraint not as limitation but as generative engine. Georges Perec wrote La Disparition, a 300-page novel, without using the letter ‘e.’ Italo Calvino used tarot cards to generate narrative structures. These weren’t gimmicks; they were methods for accessing material the conscious mind would never produce on its own.
Voss’s AI experiment belongs to this lineage, though with a crucial difference. Traditional Oulipo constraints are designed by the writer. The writer chooses the rule and then submits to it. Voss’s constraint was designed by an algorithm she didn’t control and couldn’t fully predict. “It’s Oulipo with a blindfold,” she says. “You don’t know what the constraint will be until it arrives. That’s the part that scared me, and the part that worked.”
Purdue University’s Online Writing Lab, in its introduction to creative writing resources, notes that constraint-based exercises are a standard pedagogical tool for helping writers break out of familiar patterns. But Voss’s method pushes the concept further: the constraint isn’t a fixed rule but a dynamic, unpredictable source of alternatives. The writer’s job becomes curation rather than generation—a shift that, Voss admits, felt like a loss of identity at first.
“I’ve spent twenty years thinking of myself as a generator. The person who makes things up. To become a curator—someone who selects among possibilities she didn’t invent—that felt like a demotion. Until I realized that curation is just as creative. Choosing is making.”
What the Machine Saw That She Couldn’t
I ask Voss to give me a specific example of a character backstory the AI proposed that she would never have imagined. She thinks for a moment, then leans forward.
“There’s a secondary character in the novel, a neighbor named Marcus. In my draft, he was a retired academic, widowed, lonely in a dignified way. He served a function: he was the person Elena talked to when she needed to voice her thoughts aloud. He was a sounding board. The AI, when I fed it his description, suggested something else: what if Marcus wasn’t widowed? What if his wife had left him, not died, and what if the reason she left was that Marcus had once, decades ago, done something unforgivable—something he’d never told anyone? The AI didn’t specify what the thing was. It just proposed the structure: a secret, a shame, a man whose dignity was a performance.”
Voss took the suggestion. She invented the secret herself—a financial betrayal that had ruined another family—and wove it into the novel. Marcus became not a sounding board but a mirror: his hidden guilt reflected Elena’s own unacknowledged cruelties. The novel deepened. A functional character became a thematic engine.
“That’s the pattern,” Voss says. “The AI didn’t give me content. It gave me structure. It said: ‘What if this person had a secret?’ And I, left to my own devices, had made him secretless because that was easier. I was writing efficiently, not deeply.”
The Ethics of the Experiment
No conversation about AI and authorship can avoid the ethical questions, and Voss doesn’t try to. She is aware that some readers and fellow writers will see her method as a betrayal of the craft. She has already received emails accusing her of “cheating” and “selling out.” She reads them, she says, and takes them seriously.
“I understand the objection,” she says. “If I were using AI to generate prose, to avoid the labor of writing sentences, that would be a different thing. But I’m not. I’m using it to generate questions. ‘What if the mother doesn’t cry?’ ‘What if the neighbor has a secret?’ ‘What if the ending isn’t redemptive?’ Those are questions a workshop reader could ask. A good editor could ask. The difference is that the AI has no taste, no agenda, no desire to please me. It’s a stranger. And sometimes you need a stranger to tell you what your friends won’t.”
She pauses, then adds: “I disclosed all of this to my publisher. I wrote an author’s note explaining the method. I’m not hiding anything. If people want to dismiss the book because of how I revised it, they can. But I’d rather they read it first.”
The Authors Guild’s best practices document emphasizes transparency as a key principle: authors should disclose when and how they’ve used AI tools. Voss has done that. But transparency doesn’t resolve the deeper unease—the sense that something essential about authorship is being eroded. I push her on this.
“Do you worry that this method, if it becomes common, will change what it means to be a novelist?”
“Yes,” she says. “But everything changes what it means to be a novelist. The typewriter changed it. The word processor changed it. The internet changed it. The question isn’t whether the tool changes the practice—it’s whether the practice remains an act of human judgment. As long as I’m the one deciding what stays and what goes, I’m the author. The machine is just a very strange mirror.”
The Novel That Emerged
The book that came out of this process is called The Unforgiving Room, and it is, by Voss’s own assessment, the best work she has done. It’s also the most unsettling. The prose is still recognizably hers—precise, rhythmic, attentive to the weight of silence—but the emotional terrain is rougher. Characters do things that don’t resolve neatly. The ending offers not redemption but a kind of exhausted clarity. Several early readers have told her it’s her “darkest” book. She takes that as a compliment.
“I didn’t set out to write a dark book,” she says. “I set out to write an honest one. The AI helped me see where my honesty had been edited by my own fear of being unlikable, or unresolved, or too cold. It didn’t make the book darker. It made it less defended.”
I ask her whether she plans to use the method again for her next novel. She laughs—a short, almost surprised sound.
“I don’t know. The experiment worked because I needed it. I was stuck in a specific way. The next book might be stuck differently. The danger would be turning this into a new habit—relying on the machine to disrupt me so I don’t have to learn to disrupt myself. That would be the real failure: outsourcing not just my blind spots, but my growth.”
She sets down her empty cup. “The goal isn’t to keep using the tool. The goal is to internalize what the tool taught me—to learn to ask myself the questions the machine asked, without needing the machine to ask them. That’s the long game.”
What This Means for Writers (and Readers)
As our conversation winds down, I find myself thinking about the broader implications. Voss’s experiment is not a template. It’s not a recommendation. It’s a case study in what happens when a skilled writer, confronting the limits of her own aesthetic, uses an external intelligence—not to replace her judgment, but to challenge it.
For writers who feel trapped in their own patterns, the lesson is not “go use AI.” The lesson is that patterns can be broken, but breaking them often requires an outside force. That force could be a workshop reader, an editor, a book that shakes your sense of what’s possible—or, yes, an algorithm that doesn’t care about your feelings. The tool matters less than the willingness to be disrupted.
For readers, the lesson is about what we value in fiction. If we value surprise, psychological depth, the sense that a book is smarter than its author—then we should be interested in any method that produces those qualities, provided the method is transparent and the author’s hand remains visible. Voss’s novel is unmistakably hers. The AI didn’t smooth it out; it roughened it up. The result is a book that feels less polished and more alive.
Before we leave, I ask Voss one last question: “What did the machine teach you that you couldn’t have learned from another person?”
She thinks for a long moment. The café has grown quiet; the espresso machine has stopped its hissing. Outside, the Portland rain has begun, a soft gray curtain against the window.
“Another person,” she says, “would have been kind. A good editor, a good friend—they would have seen what I was doing and tried to guide me gently toward something new. The machine wasn’t kind. It wasn’t unkind either. It was indifferent. And indifference, it turns out, is a very useful editor. It doesn’t care about your intentions. It only cares about the logic of the situation. I needed that. I needed someone—something—that wouldn’t protect me from my own book.”
She gathers her bag, stands, and offers a small, tired smile. “The machine was my unreliable muse. It gave me bad ideas and good ideas and ideas I still don’t know what to do with. But it never once asked me what I meant. It only asked what was possible. And that, I think, is the question I’d stopped asking myself.”
