Hemingway

GPT’s writing style, particularly the flat, declarative rhythm that so often emerges by default, didn’t arrive by accident. It reflects a very specific lineage in 20th-century American prose, a tradition shaped by journalism, advertising, and mass-market fiction, where brevity was treated as clarity.

Hemingway is usually cited as the inventor, not because he created the approach, but because he turned it into an aesthetic. After him, others followed; Raymond Carver, noir writers, and then technical manuals stripped it of any remaining sentiment.

By the middle of the last century, that tone had become the American standard for anything designed to be consumed quickly and without ambiguity because it was economical and deceptively neutral.

In the vast corpus used to train GPT, this style dominates, not because it’s the most expressive, but because it’s the most common; Wikipedia entries, newswire feeds, help centre articles, product reviews, screenplays, even well-meaning Reddit posts: all lean toward minimalism, not for artistic reasons, but because they’re easier to parse, translate, and scale.

So when the model reaches for a tone and has no clear signal to follow, it defaults to the statistical centre of the dataset, which turns out to be American, stripped-down, and efficient. You’ll see it in passages that sound like this:

“She walked in. The room was silent. He looked up. Nothing moved.”

In fact, it’s a party trick. Using tone and structure to convey authority.

What that means is that if you’re in the US, copy and paste works a treat. If you’re not, you’d better get good at prompt engineering.

Because there’s virtually no prompt in the world that will talk an LLM into doing  otherwise.