AI for writing books stopped being a thought experiment the moment someone put a real number on it. Coral Hart — a pen name she uses to protect her real identity — says she's published more than 370 titles since February 2025, and that she's cleared $1 million in sales this year alone. She's also received death threats over it.
What happened
Hart isn't a newcomer chasing a trend. She spent 14 years working in publishing and wrote several dozen books the old way before she touched AI. Then she spent three months learning the different models and building a repeatable system that could turn a brief into a finished manuscript. That investment is the part most people skip when they hear "370 books."
Her setup is a relay race between three models, each doing what it's best at. In her words, ChatGPT is "really good at dialogue, terrible at prose." Claude has "really good story structure, terrible at sex drives." Grok is "completely unhinged" but handles the "spicy scenes" well. She doesn't ask any single tool to write a book from scratch — she says that gets you "a flaming dumpster fire of trash" because the model has nothing to work with.
The actual workflow
So what does she feed it? The concept, the style, the character details, and a chapter-by-chapter outline. Only then does the system start generating. When she began, she could produce about one paragraph per go, usually across two or three prompts. Today she can draft an entire book in one sequence.
After the draft, it goes back through AI for "developmental edits" — the big structural pass a human editor would normally do. Then a second pass catches continuity problems: flat middles, incomplete character arcs, dropped plot threads, people whose eye colour mysteriously changes. Finally, a human proofreader reads the whole thing, fixes the "AI tells that slipped through," and does some rewriting. "And it's usually ready to go," Hart says.
The numbers behind it
The output is 371 titles, many of them romance, most of them aimed at specific readers identified through market research rather than gut feeling. One novel cracked Amazon's top 100 bestsellers. Most didn't — but she estimates she's sold more than a million books across digital, print and audio, and earned over $1 million so far this year. She publishes under more than 10 other pen names, heads an operation that has trained several hundred people in her method, and is preparing to launch software for AI authors called Authorythm.
The backlash has been brutal. Social media calls her a "grifter" and a "villain." She says she's received tens of thousands of hate messages and several death threats. She only discloses her AI use when a sales platform requires it, and points out that "for the last 16 years, nobody's asked me what my creative process was." Her answer to the "AI slop" accusation is blunt: "Books are selling. Readers don't read rubbish books; they return them."
The debate isn't going away. A novel by Haitian-Canadian author Thelyson Orelien was pulled from the shortlist for France's Goncourt prize after accusations that he used AI — accusations his publisher confirmed.
What it means for you
You probably don't want 370 novels. But the underlying method — brief first, generate second, human polish last — works for almost any writing task you've got. Here's what it looks like in real life.
At home
Think of AI as a fast, tireless first-drafter for anything you dread starting. A family Christmas letter, a eulogy you can't face, the memoir your mum keeps asking you to write. You supply the memories and the tone; the tool supplies structure and a rough draft you then rewrite in your own voice. The emotional work stays yours. The blank page doesn't.
At work
That weekly report nobody reads, the proposal due Friday, the process doc that's three years out of date. Write a one-page brief — audience, purpose, key points, tone — and let the model produce the skeleton. You'll spend your time editing instead of staring at a cursor. Hart's point about a good brief is the whole game here: garbage in, dumpster fire out.
In business
Small teams are using this pattern for product descriptions, email sequences, onboarding docs and support macros. The trick isn't "write me a landing page" — it's giving the model your customer research, your tone guide and a rough outline first. Then run a second pass for consistency, exactly like Hart's continuity check. A human still signs off before anything ships.
For study
This is where the brief-first rule really pays off. Instead of asking for an essay, ask the model to explain a concept at three levels of difficulty, then quiz you on it. Or paste your own notes and have it find the gaps and contradictions. You're using it as a study partner, not a ghostwriter — and the understanding still has to be yours.
For creativity
Writer's block is often a briefing problem. If you can describe your character's secret, the setting, and the mess they're in, most models will happily throw out ten directions the story could go. You pick one. That's the same job Hart does before her system drafts a chapter — define the concept, the style, the people, the outline.
For income
Hart's niche-romance operation shows the commercial version: pick a category with hungry readers, study what's already selling, then publish to that demand rather than to your mood. It's publishing as a business, not a lottery ticket. Note that only one of her 371 titles hit Amazon's top 100 — the money came from volume, market research and multiple pen names, not from a single breakout.
How to try it right now
You don't need custom software to test this. Start with the free tier of ChatGPT and follow the same sequence Hart uses.
1. Pick one narrow target. A specific reader, a specific topic, or one work document you genuinely need to finish this week.
2. Write a one-page brief. Concept, style, audience, 3–5 characters or key points, and a chapter-by-chapter (or section-by-section) outline. This is the step people skip, and it's the step that decides whether the output is usable.
3. Generate one section at a time. Don't ask for the whole thing at once. Feed it the brief, then ask for chapter one only.
4. Run a development pass. Ask the model to critique its own draft: what's flat, what's confusing, where does the momentum die?
5. Run a continuity pass. Ask it to check names, timelines, facts and repeated phrases. This mirrors Hart's fix for changing eye colours and dropped plot threads.
6. Rewrite by hand. Read it out loud and fix the robotic bits. Hart pays a human proofreader for exactly this — the "AI tells" are real and readers notice them.
7. Keep a swipe file of free tools. If you want extra helpers alongside ChatGPT — summarisers, prompt builders, formatting utilities — there's a useful free collection at MyKreaTool worth bookmarking.
If you later find the free limits are slowing you down, that's the signal to look at paid tiers or a second model like Claude for structure. Don't upgrade before you've finished one project on the free version.
Upsides and what changes
The obvious win is speed. Hart went from one paragraph per round of prompting to a full book draft in a single sequence, and she credits most of that to better models, not just practice. Her summary: "The prose has improved. The syntax has improved. The sentence structure has improved."
The less obvious win is consistency. A structured system doesn't get bored on chapter 19 or forget a character's eye colour — and when it does, you can catch it with a pass instead of a re-read. For anyone who writes for a living, the boring parts (outlines, first drafts, continuity, formatting) get cheaper, and the judgment calls stay yours.
It also lowers the price of trying. Ten years ago, testing a niche meant months of unpaid writing. Now you can draft the brief, generate a sample, and find out whether the idea has legs before you commit real time.
Limitations
Be honest about what this isn't. Hart's model works because a 14-year publishing veteran with a market-research habit sits on top of it — she defines the story, judges the edits, and pays a human to catch what the machine misses. Strip that out and you get exactly what she warns about: a flaming dumpster fire. Expect three months of learning before anything is smooth. Expect the "AI slop" label from readers and other writers, sometimes fairly. Expect platform disclosure rules that vary and change, so check them before you upload anything. And expect the economics to look like hers — one book in the Amazon top 100, and 370 others doing quieter work. The volume isn't a flex; it's a distribution strategy that only pays off if the research and the proofreading are real.
Conclusion
Coral Hart didn't prove that AI writes good books. She proved that a writer with a system, a market and a proofreader can out-produce almost anyone — and she's taken death threats to say it out loud. Whether you find that inspiring or grim, the method underneath it is mundane and copyable: brief first, generate section by section, edit twice, polish by hand.
One action for today: open ChatGPT's free tier and spend 20 minutes writing a one-page brief for something you actually need to finish this week — a chapter, a proposal, a report. Just the brief. No drafting yet. You'll learn more from that page than from a month of reading about AI.



Comments 0