How to Write a Novel with AI: From One Line to a 100k-Word Serial

The first time most people try writing a novel with AI, they prompt: "Write me a 100,000-word fantasy novel." They get back an incoherent wall of text and conclude: AI can't do long-form.

The problem isn't the AI. It's the method. A long novel is structure, not word count. Generating 10,000 words in one shot gives the AI no context, no foreshadowing, no character arc — it has to collapse. The right approach breaks "write a novel" into three stages: skeleton → chapters → serial continuation — each using AI, but differently.

1. Build the skeleton before writing a single word

The skeleton is the foundation. It holds: core setting, main characters, a three-act (or multi-volume) structure, and the main events of each volume.

In OpenNovel this step is "Generate Skeleton." You give one line — e.g. "an amnesiac sword cultivator works as a tavern busboy and discovers his bonded sword was borrowed thirty years ago" — and AI returns three skeleton variants (A/B/C). Pick one, or combine them into a fourth.

Each variant includes: world, protagonist motivation, antagonist, and three to five key turning points. At this stage, don't aim for detail — aim for "does it stand up." Is the protagonist's motivation strong enough? Are the turns surprising enough? A mediocre skeleton can't be saved by any amount of prose later.

2. Generate one chapter at a time, never all at once

With the skeleton done, you start the body. The core rule: generate one chapter per call.

Two reasons:

  1. Context window: AI can't hold 100,000 words. But it doesn't need to — it only needs "what happened in the last chapter + what this chapter must advance."
  2. Controllable quality: At 2,000–4,000 words per chapter, you can read, edit, and confirm before generating the next. This is ten times more efficient than generating 10,000 words and reworking them.

The prompt structure for a chapter: prior summary + this chapter's goal + word count. The prior summary isn't the full text — just key events and unresolved foreshadowing. The goal comes from the skeleton.

3. Serial continuation: don't let AI forget earlier setups

The enemy of serial writing is inconsistency: chapter 1 says the hero is an orphan, chapter 30 reveals a brother. Humans can track this, but past 500,000 words nobody can.

The fix is to let AI see a "continuity memo" before every generation: current state of each main character, unresolved foreshadowing, key events so far. OpenNovel's chapter generation automatically feeds the skeleton and summaries of prior chapters into context. This is the biggest difference from "raw ChatGPT."

4. Should you edit AI output?

Yes — but not all of it. AI first drafts are usually "structure right, prose watery." Your job:

  • Cut: clichés, repeated adjectives, filler dialogue.
  • Rewrite: replace "AI-flavored" sentences with your own voice.
  • Keep: plot-advancing, information-dense paragraphs.

A practical ratio: keep 60–70% of the draft, the remaining 30–40% is your rewrite. Zero editing makes AI obvious to readers; full rewrites waste the point of using AI.

5. Three traps for beginners

  • Don't skip the skeleton. 90% of abandoned novels die because there was no skeleton and the writer got lost at the midpoint.
  • Don't let AI generate more than one chapter. Quality falls off a cliff.
  • Don't hide your outline from AI. AI won't "intuit" your foreshadowing — you have to feed the foreshadowing state in every generation.

Closing

With AI writing a long novel, AI is your co-writer and typist, but you're the director. You own structure and judgment; AI turns your judgment into prose. Master the "skeleton–chapter–continuation" rhythm and one person can reliably produce a serial.

Want to try the flow? Sign up to start — new users get 200 credits, enough to build a skeleton and write the first few chapters.

Keep reading

← Back to list