[NarratorTD]

Author here. I’ve published several articles on my Substack recently, but I hit a wall when readers called me out for sounding "AI-generated". Even though the ideas were mine, I was letting Claude handle the prose, and it kept defaulting to that unmistakable "AI accent"; especially the obsession with em-dashes and those perfectly symmetrical paragraph structures.

I tried a different approach to fix this: I had Claude interview me for 30 minutes to reverse-engineer a "style guide" based on my actual speech patterns. I expected a professional set of rules; instead, I got a catalog of my own neurodivergent ticks and a weird tendency to argue with myself mid-sentence.

Here is the raw Style Guide (Gist) if you want to see the specific constraints I used to "humanize" the output: https://gist.github.com/Narrator/2bd64351e3dc79a08118cb67ca7...

[ofalkaed]

Style is pretty much a set of neurodivergent ticks. having a cohesive style requires understanding them so you don't argue with yourself mid sentence, unless you have a good reason to do so which works towards your style. The lack of being neurodivergent is the lack of identifiable style and essentially AI.

[NarratorTD]

You just summarized my entire post better than I did.

That distinction between "unintentional mess" and "intentional style" is exactly what I'm trying to figure out. I think I've been treating my ticks as bugs for so long that I forgot they were actually the only unique data points I had.

[ofalkaed]

Admittedly, I skimmed over much of your post, it is a bit round about and takes the scenic route (I also have a tendency to take the scenic route). If you want to use AI in your writing but maintain your style, perhaps it would be best to use AI to help structure your post instead of structuring your post for AI to write it, this will probably save you just as much time but keep things in your own voice.

Structure is where the vast majority have issues, not style, their neurodivergent ticks are going to come out regardless of how much they fight them unless they invest a good deal of time studying style and their style.

[NarratorTD]

Fair point on the skimming. I definitely let the scenic route take over in this one; Ironically, getting lost in the weeds is how I found the voice!

I love the idea of using it for structure only. My fear was that if AI touches the structure, it inevitably sanitizes the voice too. But there's probably context I can provide to avoid that.

[ofalkaed]

Give the AI the topic you want to write about and a few of the supporting ideas you want to touch on, ask it to elaborate on them, bounce back and fourth with it and question all of its suggestions, directly ask it about them. Take what you like. Repeat a few times until you are happy with the result. ChatGPT seems better at this sort of thing than Claude in my experience, it is better at picking up on your style of developing an idea and working within it as long as you remember to question it when it moves away from it and remind it of your way of thinking about things.

10 or 15 minutes should allow you to have a basic outline you can ask it to flesh out into a more complete outline and you can do things like tell it to emphasize a point and build the outline around that point or even tell it to reach that point in a roundabout fashion which only emphasis it at the end or anything else which suits your way of developing an idea. After that you just need to write it and if you do things right that will be easy.

AI is a great tool for this stuff and the big models are smart enough to learn your style and intents, becoming more mentor than shortcut, but like a mentor you need to put ins some time with them so they can figure you out and you can figure them out.