Why I Won't Read LLM Authored Fiction: Statistical Profiles and the Median Writing Problem

Chris McCormick, an independent software developer, has a clear position: he won't read fiction authored by LLMs. In a recent blog post, he explains why — and it's not about quality or plot, but about the statistical fingerprint of the prose.

The core argument

McCormick points out that every writer has a unique statistical profile — a tendency to choose certain words over others. When you read someone's work, you absorb that profile, and it nudges your own writing toward new directions. This effect is especially strong with fiction, which tends to be written from the heart and closer to raw, natural creativity.

LLMs, however, sample from a distribution that represents a "median way of writing." As McCormick puts it: "The statistical profile of LLM writing is close to the normal. That's literally how LLMs work." For him, reading fiction should push your mind toward the unexpected, not toward the average.

He writes: "I simply don't want my mind pushed towards the statistically normal when I am reading fiction. That's the exact opposite of what I want from the experience."

Why this matters for developers and AI users

For developers building or using LLM-based tools, this is a useful lens: the "voice" of an LLM is inherently average. If you're using the same model for all your writing — whether in code comments, docs, or creative projects — your output may drift toward a homogenized style. That's fine for boilerplate, but if uniqueness matters, consciously varying your inputs (including reading human authors) can help.

McCormick even notes that he's stopped reading fiction entirely at times, but always returns, because it refreshes his own statistical profile. "Your words begin to sparkle slightly," he says.

The human requirement

His bottom line: "For any piece of fiction that I read from the 2020s and beyond, I want to know 'this text was written by a human being in their own words.'"

The post is a first-person essay rather than a technical guide, but it raises a practical question for anyone using AI writing tools: how do you preserve your own creative bias?