Writing Fast and Slow

How AI Adjusts to Language and Genre

About a year ago, I discovered that I could use AI to help me edit text. It felt like a small revolution. I wrote last year about how it let me finally write at the speed of myself closing the gap between my non-linear thinking and the linear prose readers expect. More recently, I noticed something that surprised me. AI does not change my writing uniformly. The effect depends on language and genre in ways I had not anticipated.

I had often struggled to get my ideas into shape, especially when those ideas crossed disciplines or did not line up neatly at first. AI suddenly helped me see them more clearly. What it did best was take my associative, web-like thinking and turn it into sequential argument in English. It smoothed inconsistencies, clarified structure, and helped me translate between cognitive modes. I also noticed something else. It subtly changed my voice. It introduced small stylistic tics that were not mine, including formulaic contrasts, tidy declarative turns, and a faint sense of rhetorical automation.

More recently, I started writing in Spanish and expected the same thing to happen. It did not.

There is one key difference. My writing in Spanish is almost entirely narrative. These are stories from my life, observations, and remembered scenes. I often dictate these pieces and use AI mainly to clean up grammar and transitions. What surprised me is how little it alters the voice. The rhythm, the tone, and even the imperfections largely remain intact.

That made me wonder whether this has to do with AI, with the kind of writing I am doing, or with both.

One possibility is that AI behaves differently when it is asked to structure arguments than when it is asked to support storytelling. Another is cognitive. English is the language in which I tend to think analytically. Spanish is the language in which I narrate. There is likely a data issue at play as well. AI has vastly more English text to draw from, which makes it easier to lean on familiar patterns and stock moves. In connectionist terms, it may be overfitting. Spanish narrative writing may simply give it less training data to work with, leaving it less room to impose formula.

Whatever the explanation, the effect is striking. AI seems to help me in different ways depending on the language and the genre. It reshapes voice far more in analytical English than in narrative Spanish.

This suggests something important about how we think about AI writing tools. They are not uniformly transformative. They intervene most where there is friction, where the gap between how we think and how we need to write is widest. Where we are already fluent and already flowing naturally from thought to prose, they add surprisingly little.

I would be curious to hear whether others who write in more than one language notice something similar. Does AI change your voice differently depending on how, and in what language, you write?