How to write AI novels

Working with AI — structure, characters, text

What to hand over, what to keep, and how to phrase it.

Why AI cannot hold a story structureAsk a model for a story with a real turning point and you get the shape without the drop. That is not a quality problem: next-token prediction has no way to sacrifice the present scene for a future one. Where it breaks, and the two lines a human has to supply.Why every AI character sounds the sameGive a model three characters and they arrive with different names and identical interiors. The model does not remember a persona; it rebuilds one from your text every single turn — and that text is under one percent of the prompt. What to hand it instead.Why AI writes a summary when you asked for a hookAsk for a synopsis and you get events in order — accurate, complete, and completely uninviting. Two causes: the word means summary to a model, and withholding information is structurally unavailable to next-token prediction.Why you still stall when AI writes for youSupply is solved; text is infinite now. People still stop. What stops them has moved from production to judgement — and judgement is the part a model is worst at. Why candidates that are all correct are impossible to choose between.

By genre and setting

Where AI is strong, where it fails, and what only you can supply.

BL and AI — what the genre built, and what cannot be handed overWhat BL accumulated was not prose quality but resolution of relationship: power, distance, and what goes unsaid. That is the exact region a model is weakest in. Why it fails there, and what changes when you can build your own reading.

Hand it one line only you could write