AI Writing's Strengths and Limits
AI is astonishingly strong at material generation and organizing, but humans must hold the story's theme and emotional core. Just asking AI to "write an interesting novel" tends to be mediocre. The trick is to fix story structure and characters first.
Story-Structure Frameworks
Three-Act Structure (Hollywood Style)
- Act 1 (25%): protagonist, world, problem presentation
- Act 2 (50%): obstacles and trials, midpoint reversal
- Act 3 (25%): climax and resolution
Kishotenketsu (Japanese Style)
- Ki: introduce protagonist and world
- Sho: deepen the problem
- Ten: unexpected turn
- Ketsu: landing and resonance
Hero's Journey (Mythic)
A 15-stage journey structure (departure → trials → return). Frequent in films/RPGs.
How to Feed AI
- Present the chosen structure to AI as a format
- Bullet "what should happen" and "character emotional change" per act
- Have AI draft each act in 500-1500 chars
- Humans read and edit the plot, request rewrites
Character Design
Make a character sheet first and always feed it into AI's system prompt:
- Appearance, age, occupation
- Speech style and verbal tics: e.g., "Kansai-dialect base, overusing '~da yo ne'"
- Motivation, fear, desire
- Past wound: becomes the story's depth
- Arc of change: how they change through the story
Fixing the Writing Style
Left alone, an LLM becomes average "safe prose." Feeding a 500-1000-char sample and instructing "write in this style" can produce personality—veteran-author-like, light-novel-like, hardboiled-like.
For Screenplays
- Scene number, INT/EXT, time of day, action lines, dialogue format
- Instruction "Show, don't tell" (description-first)
- Awareness of subtext (gap between words and true feelings)
- Manage length with 1 page ≈ 1 minute
Failure Patterns
- Dumping on AI to "write an interesting story" → mediocre
- Character speech varies per scene → fix with the character sheet
- Repeating the same phrasing → instruct for diversity
- Weak climax → design "how to make the highlight" at the structure stage
Editing and Revision
- Critique from a third-person view (ask Claude to "find logical contradictions")
- Check character-speech consistency
- Align theme and ending
- Final finishing with human sensibility
Commercial Deployment
- Novel: self-publishing (Kindle Direct), note, literary-award submission
- Screenplay: indie film, VTuber scripts, drama proposals
- "AI use" disclosure: mandatory on some platforms
Summary
AI writing improves dramatically when you give AI a story-structure framework + character sheet + style sample. The strongest combination is a division where AI handles material supply and organizing, while humans hold theme, emotion, and final revision.
