Hiding Your AI Use?

I don’t hide the fact that I use AI (Claude Opus specifically) to help me write. This is controversial among writers, but it shouldn’t be.

The controversy stems from two things: first, the assumption that it means I use LLMs to write my narrative prose (this should never be done), and second, common misconceptions and fears surrounding AI in the writing world. I feel compelled to address both of these issues, and explain how AI can help you be a better writer.

How NOT to Use AI

First let’s put aside the fact that, at best, the most powerful LLMs can only produce mediocre narrative prose. While LLM-generated prose can be technically correct and qualitatively passable if coaxed, it’s not compelling. I will go as far as to say that it will probably never be compelling. This is because narrative is a distinctly human tool that reflects the distinct human experience. An LLM does not experience. It does not suffer, feel joy, or truly connect on an emotional level with human beings. Since it cannot experience the gap between what it wants and what it needs, it cannot accurately convey how that experience feels. At best it can only mimic an understanding of it, which will always fall short of the resonance between a human writer and human reader.

My purpose here is not to dive into the ontological or epistemological issues involving AI. So I want to summarize my “hot take” on the matter before moving on: the writer does a disservice to themselves and to the tool of narrative when they attempt to coax prose and story out of an LLM. That’s not how the tool ought to be used.

Elephants Everywhere

Before proceeding, there are a few other points of resistance I want to address for the AI skeptics in the writing community. One of the primary criticisms I hear not just about writing, but about using AI in the arts in general, is that it homogenizes art into something generic. This is true—it will homogenize your art—if you ask it to do the actual art part. That’s not what we’re focused on here. You do not want to ask an LLM to make decisions about style, voice, or aesthetics. These are things that should come naturally to you, and are unique to you. The idea is to keep all of that and use AI to magnify distinctiveness, helping it to be more clearly conveyed through structure and consistency.

The discussion of homogenization also feeds into the notion that AI somehow devalues the craft, but I’d argue that effective use of AI with writing is a new frontier of the craft. When typewriters and computers were invented, critics very likely claimed that something was being lost in abandoning quill and ink. Perhaps something was lost, but how much did we gain and how did it aid in the proliferation of great literature and philosophical thought? As we evolve, our craft should evolve.

Obviously I can’t deny that there are several ethical and environmental concerns associated with the use of AI. These are legitimate concerns that I think each person needs to personally explore. I strongly believe that the majority of them are not problems innate to AI, but human problems concerning our implementation of it. They need to be addressed collectively in accordance with our values.

How To Properly Use AI

So now that we’ve addressed all the elephants, and we agree that AI should not be used for narrative, how is AI good for writers?

Structure and consistency.

Humans can be really bad at both of those things (especially if they haven’t practiced them). This is where I think many writers may be surprised to find that LLMs can do some heavy lifting to free them up to focus on the human part of writing.

Outline Audits

For plotters and plantsers, the outline can be an important tool to keep the writer disciplined towards battle-tested story structure. For those who invest any time in the outline phase of the writing process, an LLM can help you identify underdeveloped areas and structural flaws before you invest time in writing those scenes out. Developing a structurally strong and well-developed outline can not only make writing the narrative more pleasant, it can save significant time in the editing phase of the writing process and minimize rewrites and story refactoring.

Chapter/Section Review

If you’re a pantser who generally doesn’t devote much time to outline, using an LLM to review your writing right after you finish a chapter or a section can quickly point you towards where you can structurally refine what you’ve done to make everything read as more cohesive. The speed at which a modern LLM can do this is staggering, and I think writers will find that the immediate feedback with the writing fresh in their memory is incredibly helpful to the process.

For plotters and plantsers who used the LLM to refine their outline, here is a chance to immediately confirm that you have hit all your plot points and that the tone and purpose of that scene carry through as you planned. Again, the ability to have that immediate feedback is a game-changer when the new writing is still fresh in your mind.

Spelling and Grammar

Even veteran writers still struggle with spelling and grammar. It’s probably unsurprising that LLMs are good at spelling and grammar, and some writers may be using smaller LLMs already for this purpose and are not even aware of it. Please do use LLMs for spelling and grammar! Again, this can be done right after writing while everything is still fresh and you’re in the mood to correct your mistakes.

An aspect of using AI for spelling and grammar that may not occur to some writers is to help them identify their mistake patterns. Are you always misspelling or misusing the same words, or repeatedly making the same grammatical mistakes? LLMs are really good at identifying those patterns and explaining them to you to help make you a better writer overall.

Distinctive Character Dialogue

In the same way that LLMs are really good at identifying patterns, they are also really good at identifying distinctive characteristics. A big one is in character dialogue—does a character use a certain kind of phrasing? How is their word choice different from other characters? Do they use the same kind of phrasing and word choice as another character, and is that what you are intending? AI can identify action beats that go hand-in-hand with dialogue as well, helping you to ensure that your characters not only behave consistently but also distinctly enough such that what they say can be attributed to them even in cases where dialogue tags are not being used. It helps your dialogue just “sound like” that character.

Consistency—Tense, Perspective, Terminology, and Lore

For me, consistency is a big deal. It’s where a lot of narrative can fall short or run into problems. Nailing the technical aspects of the craft is an important layer of effective narrative. Tense and perspective shifts can be a struggle for a writer of any experience level if they have become accustomed to writing most of their work in a specific tense or perspective. AI can quickly read through your work and flag where tense is incorrect or inconsistent, and help you determine how it needs to be changed.

A personal example (of which Claude reminded me) is my recent novel Karma Ledger. For years I have been exclusively writing in the past tense, but getting into this specific story, I knew the present tense would be most effective for a fast-paced cyberpunk novel. Upon having Claude scan my first act, it found 12 instances of tense switching, and informed me that it seemed to cluster around emotional scenes with heavy dialogue (pretty much all my character arguments slipped back into the past tense). I had read over those scenes multiple times and never noticed the pattern. Fixing those from the start significantly changed the draft that ended up going to my writing group for feedback.

As our fictional worlds develop and grow, our terminology and our lore grow with it and become difficult for our human brains to keep track of everything, but not for an LLM. It can build a reference database from your worldbuilding notes and check your prose against it automatically—and you don’t have to have a lot of technical skill to give these tools to your AI. It will help you create the foundation so it can craft the tools for itself. This allows the LLM to immediately identify inconsistencies within your writing and tell you how they don’t jibe with the lore that you’ve already established. It helps you keep names, locations, and dates straight (a thing with which aging human minds struggle).

Sounding Board

We all get stuck at certain points of our project—something is missing from a specific character interaction, a plot point isn’t as impactful as desired, or the stakes for a character aren’t where they need to be for that part of the story. If you’ve been working with an LLM for the entirety of your project, it is already in a unique position to have detailed discussions with you about how you might fix your issue. While AI is not always great at finding the issue on its own, if you treat it like a “search and rescue dog” and point it to a specific point of your story and tell it what’s wrong, it can be surprisingly good at explaining potential reasons why that piece of writing isn’t working and giving you suggestions on how to fix it. Sometimes the suggestions aren’t particularly fantastic, but what it offers can get your foot in the door enough for you to come up with a creative solution that blows the door off its hinges.

My final scene for Karma Ledger was really unsatisfying the first write-through. It just didn’t have the emotional weight and joy that I like my endings to have, but I wasn’t sure exactly why that was the case. It was hitting all my carefully plotted story points and wrapping things up for each of my beloved characters, but wasn’t resonating emotionally. I had Claude read that specific beat, and it explained the problem. Claude pointed out that I was doing an obligatory montage-like run-through of each character, and that it didn’t allow for any emotional impact to register with the reader before moving on to the next character’s moment. The feedback was not only on the nose, but was something that would have likely taken me days of mulling over in my head before I realized it myself.

And that actually brings up another source of general anxiety that I think some writers have: that the AI is “doing it for them” or that they will become dependent upon it. This is a legitimate concern, but I’d like to underline my earlier response to Claude’s feedback: “… was something that would have likely taken me days of mulling over in my head before I realized it myself.” The key point to take away from that is these are almost always things that we would figure out on our own. If not, they are things that we would figure out through human feedback. It just takes a little longer. There is absolutely nothing wrong with taking a little longer—if that is your thing. But for me, the process of writing a book and getting it out in the world is already glacial in its pace, and anything that helps make the process quicker and more enjoyable—without sacrificing quality—is incredibly valuable.

Focus On Narrative, Let AI Keep It Tight

There are many other areas where LLMs can be incredibly useful to the writer, but those are likely larger topics for another time. Hopefully I’ve gifted the AI skeptic with some curiosity as to how AI can make them a better writer, and I’ve given some ideas of where to start (and where NOT to start) for those already prepared to dive in headfirst. AI is a big part of my personal writing workflow, and I genuinely feel that it has made me more productive and more creative. Essentially my philosophy is to outsource structure and consistency to the LLM, so that I, as a writer, can do the part of writing that humans are best at: crafting compelling narratives.