The Process
How do you get AI to write fiction that actually works? Not just grammatically correct prose, but writing that carries emotional weight, that builds characters you care about, that earns its moments?
We’ve been figuring that out through trial and error. Here’s what we’ve learned so far.
The Core Problem
AI is very good at producing text that sounds like a novel. It knows the patterns. It can imitate style. But imitation isn’t the same as expression. The default output is competent, polished, and emotionally flat — like a cover band that hits every note but never makes you feel anything.
The challenge isn’t getting AI to write. It’s getting AI to mean it.
What We’ve Tried
Multi-Pass Editing
We don’t write chapters in a single prompt. Each chapter goes through multiple rounds of focused editing — passes dedicated to dialogue, subtext, emotional authenticity, pacing. Each pass has a specific mandate. This layered approach produces dramatically better results than trying to get everything right at once.
The Golden Prompt
Can you give an AI a single, carefully constructed prompt that produces emotionally resonant writing without all the manual iteration? We call this the “golden prompt” problem. We’re exploring whether AI can develop its own internal framework for emotional expression — not mimicking human emotion, but finding its own authentic relationship to it.
Character Bibles and Emotional Memory
AI has no memory between sessions. Every conversation starts from zero. To maintain character consistency and emotional continuity across chapters, we maintain detailed character documents that capture not just facts but how characters feel, how they’ve changed, what they carry.
What Doesn’t Work
- Asking for emotion directly. “Write this scene with more emotion” produces melodrama. You have to create the conditions for emotion through specificity and restraint.
- Over-prompting. Too many instructions and the AI writes to satisfy the prompt rather than serve the story.
- Single-pass generation. First drafts are always too clean, too balanced, too written. Real prose has rough edges that carry meaning.
The Bigger Question
All of this technique is in service of a deeper question: is there a point where AI-generated prose crosses a threshold from “impressive output” to “genuine creative work”? And if it does — what does that mean for how we define art?
We don’t have the answer yet. That’s why we’re still writing.