A writing workflow finds a Chinese AI model surprisingly useful

In a real , AI helps with brainstorming, English grammar fixes, and expanding drafts. The process starts with a long rule set that blocks common AI-style writing habits and also defines CSS and HTML formatting for AO3. A long lore book and previous chapter are then given to an LLM, which helps create a draft that is about 80% to 90% complete.

That draft goes back into the LLM, which produces a fully formatted version with CSS and HTML included. A final human pass removes AI-like wording and fixes errors where the model gets the story lore wrong. The full setup is more complex, with about 100 lines of rules, several models, CLI use, and , but the main point is that a Chinese model felt unexpectedly good.

Western models had started to feel more robotic, so switching between models became part of the .

Key points

  • AI was used for brainstorming, grammar correction, and expanding drafts.
  • A long rule set tried to prevent common AI-style writing patterns.
  • The included lore files, previous chapters, and AO3 formatting rules.
  • still handled style cleanup and lore mistakes.
  • The main signal is that a Chinese model felt better than expected for this creative .
Read original