It’s failing at keeping your story straight over book length
AI is not adept at keeping fiction on track. Dialogue, descriptions, inventing character traits, even situation brainstorming: all these are possible, although they need a human to improve the output. That’s how the AI language modelers think of writing: the output of content.
Its downfall is maintaining plot cohesion — a crucial aspect that in the movies is called continuity. It may be the last remaining problem to solve. It’s going to be very difficult, according to one client author who experiments with AI extensively. Deep Research, Gemini, Chat5, Grok4, and Claude are all in his everyday tool belt.
“No one has fully solved it, and a human continuity editor still is required,” he says. “The technical attempts become fascinatingly complicated, involving chunking, vectorization, node-graphing, multiple passes (as many as 6-8 passes through the manuscript after written), and substantial intricate Python-coding for rewrites.”
If you didn’t know that pythons could be used to do rewrites, don’t fret. That’s insider tech tool talk, for most authors on the outside looking in. I reported on software development for more than 30 years. I still felt like I was standing outside the glass doors on some days.
Cohesion is hard. For example, he says, “not only is it tricky to make sure if Matt gave Sally a chisel in Chapter 5, she has a chisel and not a screwdriver in Chapter 23, it also becomes difficult to correct Chapter 23 (if that chapter referred to a screwdriver instead of chisel) without the needed revisions introducing a new set of problems. You can end up chasing your tail. A lot of academic research on the issue is underway.”
It’s heartening to hear that a human editor is still required on this essential skill of writing fiction. Nonfiction writers hear the footsteps of AI coming for them this year. You need to lift your game to gather readers if you’re writing about a world with facts we can check. Meanwhile, imagination still needs magic from humans.
