top of page
Search

When I Got Tired of AI Writing Like AI

Writer: D. Patrick Dillon
D. Patrick Dillon
Nov 5
7 min read

Behind the Manuscript | The Falcon Hollow Editorial Guide


When I first started using artificial intelligence to help write my Falcon Hollow novels, I knew there would be a learning curve. What I didn't anticipate was how much time I would spend teaching AI to stop writing like AI.

I've mentioned before that I don't have any formal training in creative writing. I didn't attend writing workshops, study literature in college, or spend years learning the mechanics of novel writing. I had stories I wanted to tell, characters I wanted to develop, and a fictional town called Falcon Hollow that was gradually becoming a much larger part of my life than I'd originally imagined.

AI became part of that process. It helped me organize ideas, develop scenes, work through dialogue, and turn the stories in my head into a manuscript. But as we began working together, I started noticing certain writing habits that didn't fit the kind of novels I wanted to produce.

One of the most frustrating was the way AI structured its prose. Instead of the flowing narrative I expected from a novel, it had a tendency to break everything into short, choppy paragraphs.

A character would say something.

Another character would respond.

There might be a sentence describing someone's reaction.

Then another line of dialogue.

And another paragraph break.

Before long, what was supposed to be a chapter in a novel began looking more like a television script with a little narrative sprinkled between the conversations.

That wasn't what I wanted. I was writing mystery novels, specifically police procedurals, and I wanted readers to become immersed in the investigations. I wanted them to follow the detectives as they worked through evidence, interviewed witnesses, and tried to understand what had happened. Conversations needed to feel natural, with narrative woven into the scene rather than everything being reduced to a series of short exchanges.

So I'd tell ChatGPT to rewrite the scene. I'd explain that I wanted traditional novel prose, fuller paragraphs, and a more natural flow between dialogue and description. Usually, it would make the corrections, and we'd continue working.

The trouble was that those corrections didn't always last. We might get through another scene or two before the writing began slipping back into the same fragmented style. I'd point it out again, we'd revise the scene, and then we'd move forward until the problem resurfaced. It became a repetitive process, and I was spending far more time correcting the presentation of the story than I wanted to.

 

When the Writing Started Changing

 

The paragraph structure wasn't the only problem. As the manuscripts grew, I began noticing that the overall narrative voice wasn't always consistent from one chapter to another.

We might establish a writing style in the opening chapters that I really liked. The pacing felt right, the dialogue flowed naturally, and the story sounded like the kind of novel I wanted to write. But somewhere along the way, that style would begin to change. By the time we reached Chapter Ten, I might find myself reading prose that sounded noticeably different from what we'd established in Chapter One.

Sometimes the change happened gradually over several chapters. Sentences became shorter, descriptions were handled differently, or the pacing began to feel rushed. Other times, the shift was apparent within a single scene. I wasn't asking for a different writing style, and there was no particular reason for the change. It was simply something that kept happening as we worked.

Of course, different scenes require different approaches. A tense interview shouldn't necessarily read like a quiet conversation between friends, and an investigative discovery might call for a different pace than an ordinary day at the police department. But those variations should occur within a consistent narrative style. I wanted readers to recognize the same storytelling voice throughout the novel, regardless of which chapter they happened to be reading.

Instead, I was finding myself returning to earlier chapters to compare the writing, then revising later scenes to bring them back into line. The story itself might be progressing, but the manuscript wasn't always developing with the consistency I expected.

 

The Problem With Starting Over

 

Another frustration came from the way I was working with ChatGPT at the time. This was before I discovered Projects, and most of our work took place in individual conversations. As those conversations grew longer, the writing instructions we'd established earlier didn't always seem to carry the same weight. I might have to remind AI halfway through a chat that we'd already agreed on a particular narrative style or paragraph structure.

Starting a new conversation was even more frustrating because I often had to explain everything again. I'd describe how I wanted the prose structured, how dialogue should be incorporated into scenes, what kind of pacing I preferred, and which AI writing habits I wanted to avoid. We'd establish those expectations, begin working on the manuscript, and eventually encounter the same problems we'd addressed in the previous conversation.

It sometimes felt as though I was starting over with a writing assistant who didn't know what we'd accomplished the day before. I understood that ChatGPT wasn't a human collaborator with a complete memory of our previous work, but that didn't make the repeated explanations any less frustrating.

I needed a way to preserve those writing preferences so I could bring them into a new conversation without having to reconstruct everything from scratch. I also wanted something I could refer AI back to when the writing began drifting during an existing conversation.

That was the thinking behind what eventually became the Falcon Hollow Editorial Guide.

 

Putting the Writing Standards on Paper

 

On August 7, 2026, I established the Falcon Hollow Editorial Guide as a formal reference for the writing style I wanted to maintain throughout the series. The idea was to document the expectations we'd been discussing repeatedly, giving both AI and me a consistent reference whenever we worked on a manuscript.

As we developed the guide, I considered the kind of prose I enjoyed reading and the characteristics I wanted to preserve in my own writing. At the same time, I identified the AI writing habits that had been causing so much frustration. We addressed paragraph structure, dialogue, narrative flow, pacing, and the importance of maintaining a consistent voice from one chapter to the next.

The guide also began defining the overall editorial identity of the Falcon Hollow mysteries. I wasn't trying to write action thrillers filled with extraordinary heroes, constant confrontations, and dramatic revelations. My stories revolve around believable criminal investigations, ordinary people facing difficult circumstances, and the consequences that crime has on a small community. The detectives aren't superheroes. They're professionals doing their jobs, following evidence, interviewing witnesses, and trying to discover the truth.

I wanted the writing to reflect that approach. The investigations needed room to develop naturally, and conversations shouldn't be shortened simply to move the plot forward. Suspense should come from the investigation itself rather than from manufactured dramatic moments.

One of the principles established in the guide describes its purpose particularly well: the Editorial Guide preserves how Falcon Hollow stories are written. That's an important distinction because I wasn't trying to create a set of instructions that would teach AI how to write a novel. I wanted a reference that would remind AI how I wanted my novels written.

Having those expectations in a document meant I could upload the guide when starting a new conversation or direct ChatGPT back to it when the prose began wandering away from the established style. At least, that was the intention.

 

A Reference, Not a Cure

 

I'd like to say that creating the Editorial Guide solved the problem, but that wouldn't be accurate. It didn't, and even today I wouldn't describe it as a particularly good solution in every situation. It's a solution, and sometimes a useful one, but it doesn't guarantee that AI will follow the standards we've established.

I still encounter the same writing habits that prompted me to create the guide in the first place. Short, fragmented paragraphs occasionally return, narrative styles change unexpectedly, and scenes sometimes need to be rewritten because the prose doesn't match the rest of the manuscript. There are times when I have to stop working on a chapter and tell ChatGPT to review the Editorial Guide before generating another response.

Before I discovered ChatGPT Projects, managing the reference documents could be especially cumbersome. I'd sometimes have to upload them again when beginning a new conversation or remind AI to consult material I'd already provided. Projects eventually made organizing and accessing those references easier, although I still have to make sure they're actually being used. That's another experience worth discussing in a future article.

What the Editorial Guide did give me was a more consistent way to communicate my expectations. Instead of trying to remember every instruction we'd discussed in previous conversations, I had a document that established the writing standards for the series. When something didn't sound right, I could refer back to those standards and explain what needed correcting.

That didn't remove my responsibility for reviewing the work. I still had to read the chapters, recognize when the writing was drifting, and decide whether a scene needed additional attention. In some ways, creating the guide made me more conscious of the editorial decisions I was making, particularly as I continued learning about fiction writing and developing my own approach to storytelling.

Looking back, the Editorial Guide was one of my first attempts to establish a workable relationship between the stories I wanted to tell and the technology I was using to help tell them. Other reference documents would eventually follow, addressing different challenges as Falcon Hollow continued to develop. But the Editorial Guide began with a straightforward frustration: I wanted the prose in my novels to remain consistent, and I was tired of correcting the same AI writing habits over and over again.

The guide hasn't eliminated those problems, and I doubt I'll ever reach a point where I can simply accept every AI-generated scene without reviewing it. What it has done is give me an established set of standards to work from, even when I have to remind ChatGPT to follow them. I wanted to write novels that sounded like the stories I had in mind, not something generated from a standard AI template.

I still want that.

And every so often, I still have to remind my writing assistant.

 
 
 

Comments


bottom of page