Using AI Without Compromising the Story

By James M. Jackson

I belong to several writing groups that have online forums for their members. Recently, a highly respected member on one of the forums went on a rant about how any use of large-language-model artificial intelligence (like Claude, ChatGPT, and others that analyze text and identify patterns—LLMs, for short) is totally unacceptable for authors. The rationale pointed to LLMs' unethical use of copyrighted material to learn, and that it’s dumb to do anything to help a product that undercuts your product. I’ve paraphrased a much longer and more erudite analysis. In this view, writers should feed the backlash and shut AI down.

Standing on its grandest pedestal, this is the equivalent of saying that eating anything produced by American farmers on United States soil is unacceptable and should be banned because all the land used to produce the food was stolen or otherwise obtained illegally or immorally from the indigenous peoples. What’s more, if we only stopped using this product, farmers would stop producing it.

Of course, we need food to eat and don’t need generative AI to live—at least not yet.

Maybe a better analogy is that writers should go back to scribbling longhand because science indicates that doing so uses a larger percentage of our enfeebled brains than keyboarding does. But I choose to keyboard instead of writing longhand because it’s much faster (and I can actually read the result two hours later, instead of scratching my head about the intended meaning of those ink marks on paper).

My firm line with AI is that I will not have anyone or anything do my creative writing for me. I do accept human and AI suggestions to improve my work, and I will absolutely allow LLMs to produce a first draft synopsis for my 95,000-word novel. And I’ll certainly allow it to generate multiple suggestions for the book’s blurb. Even in those instances where LLMs did the initial draft, the final words and styling will be my choices, even if composed mostly of words chosen by the AI. If AI can help me perform a routine task faster, I’m all in.

Here are three examples of how I recently used Claude to perform structural analysis.

In my current WIP, Niki Unbound (Niki Undercover #3), I realized Cindy (Paddy’s wife and Seamus’s daughter-in-law) went missing for part of the story. In the before-days, as part of the read-through of the entire novel, I would have set up a little worksheet to keep track of where Cindy was and what she was doing. Since I knew it was a problem, I gave Claude this prompt:

Where's Cindy: One of the areas to check is Cindy's location in each scene. Please review Niki Unbound draft 2 2026-02-21.docx and provide a chart with the following columns: 1. Scene number 2. Cindy present Y/N 3. Cindy's location [Chicago area, D.C. area, Shank Lake vicinity, Other (specify), Unknown]. Please provide results in a downloadable Excel file. Questions?

On its own, Claude color-coded the lines to mark location (making it visually very easy to see where Cindy was), provided a short note for each scene, and provided commentary on a few scenes it thought I should look at because they were somewhat ambiguous. The graphic below shows the first thirteen scenes.

With this information, I can pinpoint where to fix this issue and do that before doing my read-through.

The second area I’ve experimented with is having Claude do a first pass at analyzing scenes for their impact on the story. Normally, before I hand the second draft over to Jan, I do a read-through and capture on a spreadsheet a number of characteristics for each scene. I want to know if it’s Action (A) or Reaction (R) or a combination. I also want to know which scenes have increased Twists (T), Dangers (D), and Reversals (R)—a concept I picked up from Jane Cleland. I also record Jess Lourey’s approach: A.R.I.S.E. (Action, Relationship, Information, Suspense, Emotion). After providing the basic definitions to Claude, I asked it to do a first pass at analyzing each scene.

Here’s what it did for the first few scenes.

I’m not claiming Claude got this 100% correct, but this makes it easy to spot potential issues that I can then investigate. It took me only a few minutes to formulate the query and give Claude the manuscript, and it has already saved me hours of work.

The third task was also in preparation for the read-through. Because I write the first draft mostly by the seat of my pants, following characters where they take me, I create plot holes and cul-de-sacs that I must fix in the second draft. Normally, I discover the problems as part of my read-through. This time I queried Claude:

Please review the Niki Unbound draft document for any and all continuity issues. Be thorough. Questions before proceeding?

It produced a four-page Word document with “issues.” It grouped them into Critical – Plot Holes & Logic Gaps, Significant – Technical Errors & Story Problems, Moderate – Tone, Pacing & Character, Missing / Unverified, and (unrequested) Typos & Copy Errors. Here’s the top part of the report:

Claude was good, but not great. It identified sixteen issues, of which four were misinterpretations on its part. An example is #2 above. It correctly identified eight people but missed the ninth. When I pointed out who I thought the ninth was, it checked and agreed I was correct. Back when I was in school, a 75% on a test was a C-minus. Until I do the read-through, I won’t know if it should have flagged other issues.

But even with the mistakes, it found a dozen issues I could fix before I did my read-through. That, I believe—I haven’t done it yet—will make my read-through faster and more valuable.

I’m not trying to convince writers how they should or should not use AI. But if you are AI-curious and tolerant, these three examples may spark some ideas for you. Writers and readers, where is your firm line in using AI?

* * * * *

James M. Jackson writes justice-driven thrillers with brains and bite, including the Niki Undercover Thriller series and the Seamus McCree series. To learn more information about Jim and his books, check out his website, https://jamesmjackson.com. You can sign up for his newsletter (and get to read Low Tide at Tybee, a novella featuring Seamus, his darts-throwing mother, and six-year-old granddaughter, Megan).