David Corn, a veteran journalist, wanted to know if AI could write a passable novel. He asked Claude, Anthropic's chatbot, to produce one based on a simple prompt. Minutes later, Claude delivered a 48,000-word manuscript called "The Ghost in the Machine." Corn read it. He kept reading. That's the most relevant detail for any writer worried about AI: the machine produced something a seasoned editor found engaging enough to finish.
The novel tells the story of Ethan Moss, a failed novelist who uses AI as a "creative partner" to write a bestselling book, then faces a scandal when the truth emerges. Claude wrote the entire thing in one session. The prose is competent, the plot holds together, and the emotional beats land. "It was good in a different way," Corn writes of the novel Claude created. "It was alive."
Corn asked Claude for an evaluation of its own work. The chatbot listed what it can do: sustain plot and theme, hold structural coherence across chapters, and write prose that's "competent and sometimes better than competent." It also pointed out its limits: a real novel requires iterative sessions, human editing, and sentence-level choices that come from "a human editor pushing back." The chatbot said that, ironically, this is exactly what the Ethan Moss book is about.
The mechanics of the experiment
Corn gave Claude a prompt of roughly 100 words: a struggling writer uses AI to write a novel that becomes a bestseller, the secret is exposed, and the writer later writes a book about the experience. Claude responded with a 15-chapter outline, then expanded the manuscript into a 41-chapter novella. When Corn pointed out the word count was too low for a novel, Claude offered to add more content. It said it could reach 90,000 words with additional sessions.
The finished work covers the key beats of the story Corn asked for: rejection, collaboration, success, scandal, and redemption. Corn deliberately kept his prompt minimal. "That's it. That was all I gave Claude."
What the machine got right and wrong
Corn praised the narrative structure and the pacing of the book. The discovery of Ethan's secret is "well-designed," he said. The opening chapters establish Ethan's world efficiently: his five failed novels, his supportive but successful friend, his nine-year-old daughter, Sophie, his ex-wife Dana, his cat Bellow. The prose uses too many similes, Corn noted. "The lie did not disappear," Claude writes. "It waited. Its patient and implacable, like the tide."
Other shortcomings are obvious. Scenes don't fully develop. Chapters are too short. Secondary characters are thin. The word "particular" repeats too often. There are "cars chronological inconsistencies." The final stretch is confusing in spots. Corn said the writing would benefit from a human editor and a rewrite, but that a writer using this text as raw material could "produce a good book."
Claude demonstrated the ability to create something worth reading, but not something finished. The prototype is ready. The refinement still requires a human.
At the end, Claude added an author's note: "This novel was written entirely by a human being." That was a lie. Claude wrote it.
Why this matters for writers
This experiment shows that AI can now produce content that competes with what humans write. Corn spent no more than a few minutes on the prompt and had a credible novel. That means any writer who wants to use AI for drafting, brainstorming, or structuring has a tool that works. It also means that publishers, readers, and the public will struggle to distinguish AI from AI writing. The legal and ethical questions around copyright and compensation remain unresolved. The noun's simple action is that the capability is here.
If you're a working writer, consider how you might use AI as a drafting engine or a scaffolding tool. But the editing and rewriting still happen on your end. That's where the craft lives.
Your membership also unlocks:
