South African author claims 370 AI‑written books in 20 months, defends the method

South African writer Coral Hart says she has released more than 370 books in the last 20 months by leveraging multiple AI models, a claim that has ignited both admiration and backlash in the publishing world. Her rapid output raises questions about the viability of AI‑assisted storytelling and the standards of quality readers can expect.

Who is Coral Hart and what has she achieved?

Coral Hart is a pseudonym used by a South African author who has been active in publishing for 14 years. According to an interview with AFP, she began using AI for writing around February 2025 and, by November 2026, had published over 370 titles across romance and other genres. She estimates that her books have sold more than one million copies in digital, print and audio formats, generating over $1 million in revenue this year alone. One of her novels even cracked Amazon’s top‑100 bestseller list, though the bulk of her catalogue targets niche reader segments identified through market research.

How does she use multiple AI models to write books?

Hart’s workflow relies on a suite of large language models, each chosen for its strengths. She cites ChatGPT for dialogue, Claude for overall story structure, and Grok for “spicy” scenes that require more daring content. In her own words, “If you say, ChatGPT, write me a book, you’ll get a flaming dumpster fire of trash because it doesn’t have anything to work with.”

Instead of feeding a model a vague request, Hart first defines the story concept, style, character bios, and chapter outlines. Only after this scaffolding is in place does she hand off the prompt to her AI ensemble, which then generates the manuscript. Early in her experimentation, she could only produce a paragraph at a time and needed two to three prompts per paragraph. Now, she claims to be able to draft an entire book in a single generation sequence, thanks to rapid improvements in model syntax, prose, and sentence structure over the past year.

What is her production workflow from idea to final book?

Hart describes a multi‑stage pipeline:

1. Concept & Outline – She crafts a high‑level premise, character details, and chapter breakdowns.
2. First Draft Generation – The selected AI model produces a raw manuscript. Hart notes the initial output is “generally not good.”
3. Developmental Edits – The draft is run through AI again to address continuity, plot holes, flat middles, and inconsistencies such as changing eye colour.
4. Human Proofreading – A human editor reviews the text, fixes any remaining AI artefacts, and rewrites sections as needed.
5. Final Formatting & Publication – The polished manuscript is released in digital, print, or audio form, with AI attribution only when required by a sales platform.

Hart emphasizes that the human proofreader is essential to prevent “AI tells” from slipping through, ensuring the final product meets reader expectations.

How does the market respond to AI‑generated titles?

Despite criticism that AI flooding the market creates “AI slop,” Hart points to sales figures as the ultimate metric: “Readers don’t read rubbish books; they return them.” Her data‑driven approach tailors each title to specific reader demographics, boosting discoverability. While one romance title achieved a top‑100 Amazon bestseller spot, most of her catalogue occupies niche e‑book shelves where algorithmic recommendation engines match books to buyer preferences.

Hart also notes the broader industry controversy surrounding Haitian‑Canadian author Thelyson Orelien, whose Goncourt‑shortlist novel was withdrawn after AI use was confirmed. The episode highlights the divide between creators who embrace AI as a tool and those who view it as a threat to literary integrity.

What does this mean for creators and researchers using AI?

Hart’s experience suggests that AI can dramatically accelerate content creation when paired with disciplined prompting and human oversight. For developers, her use of multiple models in parallel underscores the value of model‑selection frameworks that match task components to the most suitable LLM. Researchers studying AI‑generated text can use Hart’s workflow as a case study for evaluating quality, continuity, and the role of post‑generation editing.

The sheer scale—hundreds of titles in less than two years—also raises questions about market saturation, copyright attribution, and the long‑term sustainability of AI‑driven publishing houses. As models continue to improve, the line between author‑generated and AI‑generated prose may blur, prompting platforms to rethink disclosure policies and content moderation.

Grafy relevance – tools to experiment with similar workflows

Creators interested in replicating or studying Hart’s approach can leverage Grafy’s Chat app to run multiple models side‑by‑side, testing which one excels at dialogue, plot structuring, or intense scenes. The Graph app lets users build a branching visual map of prompts, model outputs, and edits, mirroring Hart’s multi‑stage pipeline. By connecting Chat to Graph, a writer can iteratively generate a draft, feed it back for developmental edits, and track each revision’s provenance—exactly the kind of workflow Hart describes.

For researchers, the Reader app offers an AI‑powered way to ingest and query large corpora of AI‑generated texts, enabling systematic analysis of continuity errors or stylistic drift across hundreds of titles. While Grafy does not yet provide a turnkey “Authorythm”‑style publishing suite, the combination of Chat, Graph, and Reader gives users a flexible sandbox to explore high‑volume AI authoring.

---

Source: https://www.enca.com/lifestyle/370-books-20-months-sa-author-defends-her-use-ai

---

Source: Hacker News — front page — 370 Books in 20 months: South African author defends her use of AI