The site read like an LLM because it was

For a year the prose on my site read like a machine made it, for the honest reason that a machine often did. So I calibrated a voice-fit meter on my own writing from before I used models, then built a rewriter that ranks candidates against it. The tell wasn't the vocabulary. It was the rhythm: every sentence the same confident mid-length, no short jabs, no clauses that trail off the way I actually think.

The tell was never the vocabulary. It was rhythm. Every sentence on my site landed at the same confident mid-length, because a model wrote a lot of it. So I calibrated a voice-fit meter on my pre-model writing and had a rewriter score candidates against the old me.

For a year my site read like a machine wrote it, because a machine often did. I built a voice-fit meter calibrated on my writing from before I used models. Turns out the giveaway isn't fancy words. It's rhythm: every sentence the same steady length, no short jabs, no half-finished thoughts.

For a year the prose on my site read like an LLM made it, for the honest reason that an LLM often did. The fix wasn't a better prompt. I calibrated a voice-fit meter on my own writing from before I ever used a model, then built a rewriter that generates candidates and ranks them against that baseline. The surprise was what the meter flagged. Not vocabulary, not the em dashes even. Rhythm. Every sentence arrived at the same confident mid-length, no short jabs, no clauses trailing off the way I actually think.

The tell was never the vocabulary. It was rhythm: every sentence on my site the same confident mid-length, because a model wrote a lot of it. So I calibrated a voice-fit meter on my pre-model writing and made a rewriter score against the old me.

My site read like a machine wrote it because one often did. The giveaway wasn't the words though. It was rhythm: every sentence the same steady mid-length. So I calibrated a voice meter on my writing from before I used models and scored the rewrites against it.