Teaching AI to write like a person, not a press release.
HUMANIZE.MD is a small, open writing standard: one markdown prompt you hand to any language model so its output reads the way people actually write. Plainspoken. Specific. Free of the tics that give machine text away.
It started as a personal correction file, the same notes pasted ahead of every draft, patched so many times it hardened into something closer to a method. The full rulebook lives on Method, and the research it leans on lives on Research.
The rulebook below is the whole method. Not an essay about writing well, not a list of tips: the actual prompt, word for word, that gets pasted ahead of a draft before it goes anywhere.
It grew out of annoyance more than theory. The same corrections kept recurring, so eventually they got written down once instead of retyped for the hundredth time. What follows is organized by the kind of mistake it corrects, not by importance, because in practice they matter about the same amount.
Copy it. Paste it ahead of a prompt. Adjust nothing unless a section clearly doesn't apply to what you're writing.
{{ beforeText }}
{{ afterText }}
{{ r.body }}
The rulebook isn't a matter of taste. It's built on a review of how synthetic and human writing actually diverge, from word-level predictability up to how a story gets told. What follows is a working summary. The full paper, with sources, is downloadable at the bottom.
{{ s.body }}
{{ s.body }}
| Metric | Human writing | AI writing (typical) |
|---|---|---|
| {{ row.metric }} | {{ row.human }} | {{ row.ai }} |
The gap between machine fluency and human writing keeps narrowing, but the seams are still visible if you know where to look. Closing them isn't about hiding that a tool was used. It's about using it well enough that the writing keeps the friction, the specificity, and the point of view that make it worth reading in the first place.
Download the paper (PDF)