natural-writinglisted

# Natural writing ## What this is for The goal is prose that reads as if a specific, thinking person wrote it — because that is also what evades AI-detectors, and it is the only version that stays evaded. Detectors are unreliable and change monthly, so tuning text to beat a particular scorer is a losing game. But the qualities that lower a detector's score — genuine voice, concrete detail, varied rhythm, an actual opinion — are the same qualities that make writing worth reading. **Optimize for the writing, not the score.** Everything below serves that. If a user explicitly wants to game a named detector, tell them plainly: the durable move is writing that genuinely reads as human, which this skill produces, and no rewrite can *guarantee* a given detector's verdict. Then do the real work. ## The two failure modes Machine-sounding prose fails in two directions at once. Fix both. 1. **It does things humans don't** — the tells: inflated diction, throat-clearing openers, rule-of-three everything, "it's not X, it's Y," uniform paragraphs, relentless positivity, hedging, vague "studies show." The full catalogue with fixes is in `references/ai-tells.md` — read it before a serious de-slop or humanize job. 2. **It fails to do things humans do** — the craft: varied rhythm, concrete specifics, a real stance, plain words, uneven structure. This is the harder half, and the more important one. You can strip every "delve" from a piece and it will still sound like a machine if it has no