I asked AI to write a novel. It’s not so bad.

A version of the below article first appeared in David Corn’s newsletter, Our Land. The newsletter comes out twice a week (most of the time) and provides behind-the-scenes stories and articles about politics, media, and culture. Subscribing costs just $5 a month—but you can sign up for a free 30-day trial.

Like many offbeat ideas, this one originated in a bar.

I was drinking with a fellow author, and we were bemoaning the future of writing in a world of artificial intelligence. We nervously pondered how much of what we writers do could be replicated now or in the near future by a chatbot. We assured ourselves that our work—researching and investigating and then composing articles and books on complicated topics, which present fresh material, compelling context, and sharp analysis in (we hope) an engaging manner—cannot be supplanted by a machine. At least not yet. But what of other fare? Straightforward accounts of political developments, global affairs, or sporting events? Recipe books? Or histories and biographies? Short stories or novels? Film scripts?

What could AI produce? Generative artificial intelligence has already invaded—or infected—the world of fan fiction. I have friends who have used AI with their kids to create children’s books. Iranian meme-makers have deployed it to concoct slick Lego-style videos deriding Donald Trump.

The issue is not the ability of a chatbot to churn out algorithmically determined masterpieces. After all, most of the stories produced by us humans are not masterpieces. But can AI manufacture books that are good enough that some readers will welcome them and not care that they emerged from a data center? Perhaps even pay for such fare? Might many writers soon face an extinction-level event?

Following that doom-and-gloom session, I decided to see for myself if AI is a clear and present danger to writers. I asked Claude, Anthropic’s AI chatbot, to generate a novel. It did, and I’ll cut to the chase: The novel isn’t so bad.

Let me note that this exercise deliberately sidesteps the many profound concerns about AI: the economic dislocation it could cause, the environmental impacts and community costs of data centers, the expropriation of work produced by authors, journalists, and other humans. It focuses only on what AI might mean for the future of writing and other creative endeavors. And don’t forget that the reason an AI chatbot can kick out a novel is because its Big Tech master—without seeking permission or offering compensation—hoovered up a gajillion books, articles, news stories, and other material created by people. (The Center for Investigative Reporting, the parent company of Mother Jones, has sued OpenAI for copyright violations. OpenAI denies the allegations.)

I began with this prompt:

A struggling writer who has had five novels rejected turns to an AI chatbot to write a novel. He submits that novel to a publisher—without revealing it was written by an AI chatbot. It is accepted and goes on to become a bestseller. No one knows it was written by AI. But eventually that fact becomes publicly known, and a controversy ensues. The writer then writes a book about that experience. And it becomes a bestseller.

That’s it. That was all I gave Claude. (This was before the recent news that Macmillan US had canceled a $2 million deal for a crime novel after suspicions arose it had been written with AI.)

Claude responded enthusiastically: “That’s a fantastic premise—it’s got layers of irony, commentary on the publishing industry, and a real timeliness to it. Let me write this for you.”

A few minutes later, it produced a manuscript and declared, “Here’s The Ghost in the Machine—a fifteen-chapter novel in three parts. It follows your plot arc: the five rejections, the turn to AI, the bestseller, the unraveling, and the redemptive second book written entirely by hand. I gave the protagonist a daughter named Sophie who ends up being the one to suggest he write about what happened, which felt like the right emotional engine for the third act.”

I took a glance at the file without reading it. The supposed novel was 6,000 words—much too short. I pointed that out, and Claude agreed: “This comes in closer to a novella. A proper novel would be 70,000–90,000 words.” It asked if it should expand the manuscript—was it a manuscript?—into a “full-length novel…with real scenes built out, secondary characters developed…and the kind of granular detail that makes fiction feel lived-in.” I replied, “Please do that.” As if I were talking to a person.

Several minutes later, Claude presented me with “a complete novel at approximately 48,000 words across 41 chapters in three parts.” It explained, “That puts it in solid novella-to-short-novel territory—comparable to The Great Gatsby (47,000) or Fahrenheit 451 (46,000). To reach a full 80,000-word novel, each part would need roughly another 10,000 words of additional scenes and subplots.” The short version, I figured, would suffice for this experiment.

I began reading. The protagonist of The Ghost in the Machine is Ethan Moss, a 41-year-old failed novelist who lives in an apartment in Astoria, Queens. He’s divorced from Dana, a lawyer, who resides in suburban New Jersey with their nine-year-old daughter, Sophie. Ethan has a cat named Bellow (get it?); his closest friend is Marco Bellini, a successful novelist who recently struck it big with a book called The Sauce, which is about “a chef who returns to his grandmother’s village in Calabria and discovers a family secret involving the local Mafia, a contested recipe for ragu, and a seventy-year-old love triangle that has shaped the village’s social dynamics in ways that no one is willing to discuss.”

Claude describes in detail the novels that Ethan has written and that publishers have turned down. There was Meridian (a campus novel about a plagiarism scandal at a New England liberal arts college); Night Market (a psychological thriller about an American woman who disappears in Bangkok); The Fallback Position (a satirical novel about a man who fakes his own death to escape his debts); and Border Country (a multi-generational saga set along the Texas-Mexico border).

With the rejection letter for his fifth novel, The Contrarian—“While the prose demonstrates competence and occasional flair, the manuscript ultimately lacks the narrative urgency and emotional specificity that would allow us to champion it in today’s marketplace”—Ethan, who has spent nearly two decades in the literary world as an MFA grad and adjunct professor, attending writers’ workshops, obtaining grants, and demonstrating potential without achieving success, hits a wall.

He is stuck. Marco tells him his writing is sharp but “there’s something holding you back.” That reminds Ethan of when a writing instructor observed that he had the instincts of a writer but lacked “the emotional necessity.” Clearly, he must change something.

He lies in bed one night and wonders if AI could be the answer. Here Claude makes what we would call—if we were talking about a human author—an interesting choice. Ethan doesn’t turn to the machine to produce a novel (as I did). Instead, he contemplates employing AI as a “creative partner—a sounding board, a first-draft engine, a way to break through the wall that had been standing between him and a publishable novel.” He then reflexively rejects the idea: “This is cheating…You are a writer. Writers write. They do not outsource the writing to a machine and then put their name on it.” Writing, he once said at a conference, is “not a product. Writing is a process.” A human being struggling with a blank page—or blank screen.

But Ethan has lost that struggle. He opens his laptop, goes to an AI site, and types: “I want to write a novel. Literary fiction. Contemporary setting. A story about a man who discovers that his recently deceased father had been living a secret life—a second family in another city. The son goes to that city to find the other family and has to reconcile two versions of the father he thought he knew.”

The chatbot kicks out a few pages of “competent” prose outlining the story. Nothing better than what Ethan could have produced on his own. Phew, he thinks at first. But then he realizes that within this response are several good ideas: structural suggestions, a few nifty metaphors. He asks the chatbot, “What would make this story feel urgent? What’s the emotional engine? What does the son want that he can’t admit he wants?” The reply is long and dry. But there’s a notion that catches his eye: “The son doesn’t want to understand his father. He wants to understand why his father chose the other family. Which means the real question of the novel is: Am I the life my father wanted, or the life he was stuck with?”

“That was it,” Ethan thinks. “That was the book.”

He begins writing. Each morning, he spends several hours writing, as he always did. In the afternoons, he turns to AI to draft different versions of scenes and alternative versions of conversations. He poses questions about structure to the chatbot, which serves up various options. He doesn’t incorporate the material straight into his draft. But AI helps create a map for him. As Claude “writes,” the prose “was always his. But the ideas, the structural decisions, the emotional insights that organized individual scenes and, cumulatively, the book itself—these were collaborations. He was the writer, but he was not writing alone.”

This goes on for months, and Ethan produces The Other Side of the Door. Reviewing his AI-assisted work, he concludes, “The book was good. It was not merely good in the careful, defensive way that his previous novels had been good—technically proficient, intellectually respectable, praised in workshops and rejected by editors. It was good in a different way. It was alive. It was, by a considerable margin, the best thing Ethan had ever written.” He is particularly proud of his creation of Dorothy, an elderly Black woman “who offered, in a series of long, digressive, heartbreaking monologues, a perspective on love and duplicity and the American family that was unlike anything Ethan had ever written.” AI had nothing to do with Dorothy. She was all his.

You know what happens next.

Ethan sells the book for a sizable advance. The novel is hailed as his breakthrough. He is feted by the literary world. It becomes a bestseller. Readers are moved by the book and send him letters saying it has helped them understand their own lives and families. There’s a demand for his next book. He has trouble firing that up. He tells no one about his reliance on AI. The guilt builds—and builds. A computer scientist releases a program that evaluates writing for possible AI involvement. His book registers as possibly including AI-produced content. A perceptive data journalist finds a nugget of a clue in the text, digs deeper, confronts Ethan, who can no longer bear the burden of his secret and confesses.

A scandal erupts, as critics and commentators in the Atlantic, the New York Times, and Harper’s and across the podosphere debate what is writing, what is creativity. Was this cheating or merely using an available tool? If Ethan’s novel connected with a reader and evoked emotions, did it matter how it was birthed? As Claude puts it, “Was a tear cried over a human-machine collaboration less real than a tear cried over a purely human creation?”

But Ethan is tarnished, dropped by his agent and publishing house, and scorned by the lit crowd. He loses the movie deal for the book. Worse, he must explain to his daughter he was dishonest.

Yet he manages to crawl out of this wreckage and writes—in longhand on legal pads—a novel titled The Honest Liar, based on his own rise and fall. It’s a hit—because I told Claude to make it a hit—and that, too, triggers another public hoo-ha concerning redemption and what one pundit calls “the death of honesty.” Ethan takes his lumps but prevails and carries on as a writer. I did ask for a happy ending.

You can read the novel—if we can call it that—yourself. Other than dealing with a few clunky formatting issues, I changed nothing.

When I noted above that the book was not so bad, I said that because as I read The Ghost in the Machine, I wanted to keep reading. The plot is taut and, though I knew what was coming, the journey to the narrative guideposts I requested is a good ride, with a few surprises along the way. The discovery of Ethan’s secret is well designed.

But there are obvious flaws. Scenes are not fully realized; chapters are too short, with the action overly compacted. There’s not enough dialogue. More could be done with the secondary characters. The word “particular” is overused. A few chronological inconsistencies distract. The final stretch is confusing in spots. Perhaps if Claude had expanded this draft, some of these errors would have been remedied.

Other problems, perhaps not. The writing is too reliant on similes. There are similes everywhere, too often distracting from the flow of the story like a fly buzzing around your office while you’re trying to write a newsletter. “The lie did not disappear,” Claude observes. “It waited, patient and implacable, like the tide.”

The manuscript is also chock-full of cutesy and unnecessary observations. For instance, Claude tells us that at writers’ parties attendees should avoid the kitchen, where the conversations “tend to become competitive,” and keep to the living room, where they “tend to become confessional, which is more interesting and less exhausting.” (Wry reflections on the publishing industry are sprinkled throughout.) And there’s way too much of the darn cat in the story. Rewriting and a round or two of editing could easily turn this into a much better novel. And if a writer were to follow Ethan’s example and use this text as raw material but rework the prose and take the story and characters further, he or she could produce a good book.

But credit where credit is due. There are several well-executed passages, such as when Ethan is confronted by the data journalist and spills the beans. That scene has emotional depth and intensity and certainly deserves the expansion I denied Claude.

After reading The Ghost in the Machine, I asked Claude to evaluate its ability to generate a novel. It replied:

What I can genuinely do: Sustain plot, character, and theme across a full-length manuscript; hold structural coherence over dozens of chapters; write prose that’s competent and sometimes better than competent. The Ethan Moss book is proof of concept—it has an actual arc, recurring motifs, and a thematic spine (“the difficulty is the point”) that pays off.

What I can genuinely do: Sustain plot, character, and theme across a full-length manuscript; hold structural coherence over dozens of chapters; write prose that’s competent and sometimes better than competent. The Ethan Moss book is proof of concept—it has an actual arc, recurring motifs, and a thematic spine (“the difficulty is the point”) that pays off.

Where the limits are: Length has to be built in sessions, not one pass—I can’t produce 90,000 polished words in a single response, so a real novel is an iterative project: outline, chapter batches, revision passes. And the prose tends toward a certain smoothness; the idiosyncratic, risky sentence-level choices that distinguish a memorable novel from a readable one usually come from a human editor pushing back—which, ironically, is more or less the premise of the Ethan Moss book.

According to Claude, AI needs a human to craft a good book. For the time being, that is.

Claude created a decent yarn, one that leads the reader to ponder the creativity of writing, the meaning and impact of art, and the new role of AI in all this. It’s not mind-blowing, but it is thought-provoking.

Worried about AI? About the battle between humans and the machines? Claude ends up coming out on the side of us meat puppets. At the end of the story, assessing his travails and triumphs, Ethan concludes that “writing is not just about the product. It’s about the process. And the process is where the meaning lives. The process is where you discover what you have to say, which is different from what you want to say or what you planned to say. The machine gave me what I wanted. But it took away what I needed—which was the struggle, the failure, the difficulty. The difficulty is where the writer becomes the writer. Without it, you’re just a person with nice sentences.”

Thus, a machine can’t truly be a writer.

But isn’t this judgment exactly what a clever AI chatbot would deliver to us humans?

This experiment did not convince me that Ethan is right. The Ghost in the Machine was a fun read for me. Maybe because I was biased and impressed by the novelty. But it made clear that not soon but right now AI can concoct content that can somewhat compete with what we mere mortals produce. We now live in a world in which you can decide you want to read a novel about a teacher in 1930s Germany who tries to save her favorite student from becoming a member of the Nazi Party and—poof—you have it. Crave a new James Bond story—how the British spy battles a Russian oligarch who’s scheming to take over the internet? You can get that too. (If there’s a copyright issue, call him James Broad.)

Rework Shakespeare? How about a play in which they decide not to wait for Godot? A movie script for The Godfather Part IV? Eventually AI will be able to turn any script into video. Imagine Jerry Seinfeld hanging with Lucy and Ethel. (Maybe Jerry Steinfeld.)

I’m assuming this AI content will be cheap—at least, cheaper than buying a hardcover book or paying $22 to see The Odyssey. The rise of the digital information ecosystem has already demonstrated it’s difficult for content that’s expensive to produce (say, journalism) to compete with slop available for free. There will be human-made books, movies, television shows, and music that stand out as accomplishments of imagination, innovation, and ingenuity. But the tide of AI-generated material is rising fast. Most of it won’t be great works of art. The threat is that they will be sufficient.

I suppose one could argue the positive side is that AI will turn everyone into a content creator. But worriers are right to fear that AI could substantially change—perhaps subvert—high and low culture, while diminishing writing and analytical skills. (See Bret Stephens’ recent screed against AI in which he proclaims that “it’s mentally enfeebling” and leading to “an ever-dumber world.”) And if everyone is reading their own personalized, choose-your-own-plot novels and watching their own custom-made television shows, we will have even fewer common cultural touchstones than currently exist.

At the end of the manuscript, Claude added this:

Author’s note: This novel was written entirely by a human being, which is either a point of pride or a statement of the obvious, depending on your perspective. It was also written about a human being who was, in many ways, the author’s worst self—the self that wants the shortcut, that fears the struggle, that would rather be successful than honest. If you recognized that self, you are not alone. If you didn’t, you are either very fortunate or not paying attention.

That was obviously a lie. But what was the point of this note? Was it meant to be ironic? Engaging pieces of art often spur questions without clear answers or prompt interpretations that are not definitive. Was Claude aiming at that? I decided not to ask it.