Regarding AI, Part Three
by Evan Braun
The Stylistic Tells
In our last post, we went through a sequence of mechanical writing devices that, when overused, tend to point towards AI use. In today’s post, we’ll turn our attention to some of the big-picture stylistic approaches.
One of the key takeaways I hope you’re getting by now is that it’s not about the overuse of particular words or punctuation. I can’t tell you how many times I’ve seen someone online suggest that a particular word is a favourite of ChatGPT. And of course everyone will likely remember the phase in 2025 when authors furiously scrubbed their writing of em-dashes, under the fear that people might come to the wrong conclusion.
As someone who has a love of em-dashes going back thirty years, let me reassure you that no form of punctuation is unwelcome, so long as it’s used the right way. And yes, that includes em-dashes. No need to avoid them.
1. Strangely generic voice.
Perhaps the most notable quality of AI-assisted books is the competent prose, free of obvious typographical errors. There aren’t many spelling mistakes—indeed, there may be zero, especially if the author copies the text more or less verbatim.
From an editorial standpoint, you might think this would be welcome! Far from it. While the absence of major grammar and spelling mistakes is obviously a good thing, the clean text that remains is interchangeable with little to no sense of the author’s personality or specific cadence. If you were to remove the author’s name, the text could have been written by anyone.
I would happily take some grammar and spelling mistakes if it meant the work was distinctive enough to stand out.
2. Saying much while saying little.
AI books also tend to have unusually low information density. What I mean by this is that you may have long, verbose paragraphs, but they don’t have very much to say. AI often uses twenty words to convey what could be more succinctly expressed in ten.
Instead of advancing an idea over the course of a paragraph, or multiple paragraphs, AI will fill space by merely restating the same idea in different words. As a result, the text often feels longer and more inefficient than it should be.
3. Repetition disguised as variation.
This is a related point, although I think it’s worth drawing out specifically. In order to expand a piece of writing, AI has a strange tendency to make the same point repeatedly, albeit phrased in alternate ways, without any self-awareness. Readers can thus feel the illusion of making progress without actually assimilating new ideas.
4. Predictable structure.
Every paragraph, section, and chapter may follow the same essential rhythm, as though it’s been composed based on the same underlying outline. The effect is that the text can feel rote and uninspired.
5. Constant hedging.
Human writers tend to commit to their arguments wholeheartedly and have to push themselves to hedge, often on the advice of editors and proofreaders. AI writing has the opposite tendency, almost subconsciously qualifying any and every stance, avoiding definitive claims with unusual reluctance.
The core effect is that the reluctance to commit to specifics can mean that the manuscript gets filled with vagaries and abstractions, with few concrete details and examples to anchor the central thesis. AI-assisted books will default to broad points and aphorisms that could apply to almost everyone.
Bizarrely, the writing can come across as both confident and uncertain of itself at the same time.
6. Overconfidence.
Indeed, the opposite problem—overconfidence—is also a symptom of heavy AI reliance. AI writing frequently comes across as extremely confident even when the information being shared is blatantly incorrect. Anyone who’s spent a lot of time dealing with AI chatbots will likely recognize this problem right away—and when challenged, AI chatbots often make the perplexing decision to double down rather than admit their own algorithmic error.
7. Artificial emotionalism.
AI writing is smart enough to know how emotions are supposed to sound, so the emotion does bleed through as intended. But the feeling often comes across as somewhat hollow, because there can be little genuine insight behind expressions of empathy.
8. Low authorial risk.
Speaking broadly, AI-assisted text will pose few surprising opinions. Its metaphors and similes will likewise tend to be well-worn and conventional, bordering on cliché. The thing about human writers is that they have the innate ability to surprise, for better or for worse.
Stepping back from this to try to get a sense of the big picture, my editorial spidey-sense is going off a lot these days. Just like AI operates on pattern recognition, the human mind does the same, albeit not as perfectly and reliably. It’s fair to say that I have read thousands of unpublished (or at least pre-published) manuscripts during my time as an editor. And in the course of a single year, I’m asked to evaluate or provide sample edits for even more.
Maybe it’s like recognizing a familiar accent when you hear it, but this distinctive AI voice pops right off the page more often than not. I typically don’t have to dive deep to find it.
Now, I’ll be the first to admit that it’s not a good idea to rely entirely on any one person’s editorial intuition, which is why we use a variety of tools and consult other editors to get second opinions. Editors aren’t supernatural AI detectors, even if they are pretty good at that. Rather, it’s all about the preponderance of evidence.
But perhaps “evidence” isn’t the right word. Ultimately this isn’t like a court case where we’re the prosecutors looking to find out whether a crime has been committed and lay charges. Far from it! Using AI to help build your manuscript isn’t a crime, of course, and Word Alive Press isn’t here to punish you. It’s a very useful tool, especially for those who may not feel they have the expertise or patience to write an entire manuscript on their own (although I’m here to tell you that you absolutely can, since I’ve been coaching inexperienced authors to do exactly that for more than twenty years).
As I mentioned in the first blog post in this series, though, it’s incumbent on us to know when AI has been used, because it could alter how a book is published and distributed. Knowing that AI has been used can also help us editorially correct for some of the ensuing problems and make it easier to ensure your book shines as brightly as possible.
That’s what we’re here for, after all.
* * *
Did you enjoy this post? Discover more helpful advice from Evan Braun here or read the entire series on AI as posts become available.
About this Contributor:
Evan Braun is a full-time author and editor. He has authored three novels, the first of which, The Book of Creation, was shortlisted in two categories at the 2012 Word Awards. He has released two sequels, The City of Darkness (2013) and The Law of Radiance (2015), completing the series. Braun is an experienced professional editor, and has worked with Word Alive Press authors since 2006. He is also a regular contributor at The Fictorians, a popular writing blog.