I can get a first draft remarkably quickly. Deciding whether I actually want it is another matter.
One of the pleasures of working with AI is how easily I can explore an idea that isn’t quite ready to become a story.
I can describe a situation, a possible joke or an uncomfortable choice, and the model will attempt to turn it into a scene. It may invent something I hadn’t considered. It may take a character in a direction that makes me curious. It may produce a conversation that gives me a much clearer sense of what I want to happen.
It may also get almost everything wrong.
Oddly enough, that can still be useful.
Reading an imperfect scene tells me things that thinking about the scene doesn’t always reveal. A joke I enjoyed imagining can become painfully unfunny once someone actually says it. A dramatic confrontation can feel artificial because the characters have no good reason to behave that way. Sometimes I discover that I am more interested in what happens immediately afterwards than in the scene I originally asked for.
So I change it.
And then I change something else.
The trouble with fixing one thing
The most frustrating revisions are often the ones that look successful at first glance.
I might tell the AI that a character cannot know a certain secret yet. The next version removes the revelation but leaves the character behaving suspiciously as though they know it anyway. Or I explain that a physical detail is wrong, only to receive a paragraph carefully reassuring the reader that the problem is no longer present.
The model has paid attention to my correction. Unfortunately, it has also made that correction part of the storytelling.
That is where I have learned to be careful about what I ask for.
Sometimes I need a sentence changed. Sometimes I need the AI to forget an entire version of the scene and approach the situation again.
Those are very different requests, and they can produce very different results.
Letting characters discover their own world
A particularly delicate part of the process is deciding what the characters are allowed to understand.
As the person developing the story, I may know what a strange object means, why a room exists or what a seemingly harmless event will lead to. The characters do not.
AI can be so eager to make a scene understandable that it explains more than the characters have actually learned.
I have repeatedly had to pull that knowledge back out of the prose.
The reader should sometimes notice something without knowing what it means. A character should be allowed to misunderstand a clue. An unfamiliar place should have time to feel unfamiliar.
If every mystery is translated into a neat explanation as soon as it appears, there is very little left to discover.
And discovery is one of the reasons I enjoy writing this universe.
Reading it as a book
I use AI to help write, reconsider and critique material, but there comes a point where the draft needs to be read without caring how it was produced.
Does the conversation sound natural? Is the humor still funny? Do I recognize the people speaking? Am I curious about what happens next?
Some of the more elaborate revisions have taught me that a passage can become more accurate and less enjoyable at the same time.
I can keep adding rules to prevent mistakes, but that doesn’t necessarily make the novel better. Occasionally, it makes the narration sound as though the characters are trying to justify their behavior to an invisible examiner.
That is usually a sign that I need to stop treating the scene like a problem to be solved.
I need to treat it like a story again.
The unexpected reward
AI has changed how freely I experiment. I can explore a direction without committing myself to weeks of writing, and I can discover possibilities I might not have reached on my own.
The cost is that I also have more possibilities to reject.
It is tempting to keep something because it is clever, entertaining or beautifully written. But the question I keep returning to is much simpler.
Does it make this particular story better?
Sometimes the answer is yes. Sometimes I need another attempt.
And sometimes the best thing I can do for a chapter is admit that I liked the idea more than I liked reading it.