Analysis and Filtering of Fiction using AI
A more ruthless editor of your own work and playing with yarn.
AI’s role in fiction writing has come a long way. Back in 2022–2023, I wrote a few pieces on using it for copyediting. But that ground’s mostly covered/outdated now—for good reason. Tools like Grammarly have caught up. The real shift? Advances in large language models (LLMs) with stronger reasoning capabilities have unlocked far more valuable insights.
Here’s a quick example from a sci-fi novella I’m drafting.
The challenge: I’m juggling multiple manuscripts in a shared story universe. They rely on a consistent backstory and world logic, but the facts often diverge—sometimes on purpose—to reflect different viewpoints and narrative threads.
When I start a new manuscript, I pull together fragments—old prose, scattered notes, half-scenes. I don’t trust outlines. I need to feel the story through the prose itself, even if it’s just placeholder text I’ll toss later. I don’t trust notes—too much of what I tune into lives in the cadence and fills the spaces between bullet points.
So here I am, staring at a 25,000-word sketch—not even a real first draft yet. Once I've done that, I’ll read it through and make real plot decisions. But for now, the goal is simple: shape it into a consistent, dramatic narrative.
Take a recent example: I needed to track a minor subplot involving the genetic manipulation of pinyon jays (a long story). I wanted to know where references to it appeared, whether it held together logically, and what inconsistencies there were.
This kind of messy, detail-oriented analysis—scanning, pattern-matching, logic-checking—is where today’s reasoning LLMs add value. I used ChatGPT o4-mini in the example here.
Here’s the simple prompt I used to kick things off:
Read the following story very carefully. Catalog references to stealing the jays patterns. Do you note inconsistencies? It should work through viruses. Viruses are the pattern thieving mechanism.
It doesn’t entirely make sense yet. But I was working with the LLM in a session where it had already seen several iterations of the storyline, so it understood the context. It rephrased my request as:
(A)ll the key moments in the story where the blue jays’ own behavioral “patterns” or “neural patterns” are described as being stolen, harvested, or otherwise exfiltrated—along with notes on where the text seems to slip or contradict its own rules about how that theft actually happens:
Yeah, that is what I meant.
The full report is in the appendix—and it’s solid. I’ve run these analyses for a while, and while they’re rarely perfect, most issues come down to interpretation or gray areas. Still, for a messy 24,000-word manuscript, it outperforms any human I could realistically cajole or afford to hire.
More importantly, these reports function as checklists. I go through them, decide what (if anything) needs to be revised, tune the prompt, and rerun the analysis as often as I like. That repeatability makes the occasional imperfection less of an issue.
The prompt I gave earlier was crude—so simple I slang-ed it to the LLM. With more time spent on prompt design—clearer goals, examples, context—I could generate far more sophisticated reports.
The takeaway? The pinyon jay subplot might not make the final cut. But this kind of iterative analysis helps me sift through competing ideas and plotlines—and ultimately decide what stays.
That’s the fundamental shift—I’ve moved beyond grammar checks to using AI as a pattern-recognition engine for my blind spots. A messy 25,000-word draft that might take weeks to untangle a single thread manually becomes a weekend project. It’s not just about writing faster; it’s about thinking, testing, and failing faster. AI doesn’t improve your writing—it makes you a more ruthless editor of your work. And that, in the end, might be what makes you better at it.
Appendix — The Full Report
Catalog of Jay-Pattern–Stealing References
Viral Cross-Contamination Creates Memory-Theft in Jays
“…memory-stealing viruses designed for human subjects have somehow bonded with the cognitive enhancement protocols intended for the combat swine. Local corvid populations show unprecedented neural activity.”
(Dr. Chen’s notes on Sector 7 breach)
Viruses Copy Neural Patterns into Protein-RNA Frameworks
“As the viruses made memories inaccessible to conscious recall, they copied the neural patterns into protein-RNA frameworks—perfect biological storage that could be transferred to the biotech ecosystem.”
Jays “Infected” with Human Tactical Fragments
“They injected stolen memories into their minds. Human memories. Tactical fragments, emotional residues, identity shards.”
Jays Become “Operatives” via Viral Theft
“Minds that had been harvested, compressed, and stored inside the same viral code now running wild through the jays’ neural networks.”
Pennie’s Cog-Sci Module “Steals” Jay Patterns
“When Pennie stole their neural patterns through Runnah’s Cog-Sci module, she was robbing minds that had been accidentally weaponized…”
Jays Recognize Theft via Virus-Like “Calling to Like”
“The memory-theft viruses in their systems recognized the theft—like calling to like.”
Stolen Jay Tactics Weaponized in Penelope’s Drone Swarms
“You learned from the best teachers. Your little birds gave us such elegant lessons in revenge.” (Penelope bragging as her drones execute jay-style harassment formations.)
Noted Inconsistencies
Viruses vs. Cog-Sci Module as the Thief
Virus Mechanism: Most of the text makes it clear that the memory-stealing viruses are what harvest jay (and human) neural patterns into protein-RNA “storage.”
Manual Hack: Later, Pennie “steals” jay patterns herself via Runnah’s Cog-Sci module—implying a direct, digital data-theft rather than a viral copy.
Conflict: If viruses have already exfiltrated jays’ pattern data, why (or how) is Pennie’s Cog-Sci hack necessary or even possible? The story doesn’t reconcile whether the jays’ scheme is purely viral or also digital.
Jays’ Memory of “Pennie’s Theft”
The narrative twice attributes jays’ rage to what Pennie did (via Cog-Sci), yet earlier the jays are described as victims of viral cross-contamination. Are they upset at MarsCorp’s original viral theft, or specifically at Pennie’s secondary theft?
Scope of Viral Copying vs. Behavioral Amplification
Copying vs. Corruption: Sometimes the viruses are said merely to corrupt retrieval pathways (so memories are lost, not copied)—yet elsewhere they “copy” entire neural patterns out. That dual role (erasure and copying) is fine, but the mechanics shift between “slow corruption” and “instant archive.”
Instincts vs. Injected Tactics: At one point the text insists the jays’ tactics were “already there” as instinct, just amplified by viruses—but then it says Pennie’s module injected new stolen jay tactics into drone swarms. Are those tactics viral amplifications of innate jay skills, or fresh human-sourced modules Pennie extracted?
Agency of the Jays vs. Viral Automation
Automated Harvesting: Viruses act autonomously across species to steal memories.
Jay Consciousness: Yet the story often anthropomorphizes jays as recognizing, protesting, even organizing revenge at a meta-level (“like calling to like”). If viruses are mindless code, how do they trigger the jays’ own awareness of being stolen from, unless there’s some other “conscious” vector?
Bottom Line: the core premise—viruses that harvest and transplant neural patterns—is consistent, but the story slips between viral copy and digital hack as the means of stealing jay patterns, and between innate instinct amplified versus foreign tactics injected. If you want the jay-theft thread to be unified, you’ll need to choose one mechanism (viruses or Cog-Sci download) and stick to it, and clarify whether the jays are reacting to MarsCorp’s initial viral theft, Pennie’s secondary digital theft, or both.