humanize-writing

name	humanize-writing
description	Edit long-form and web writing so it reads like a person wrote it — cut AI tells (filler openers, uniform rhythm, flagged vocabulary, em-dash overuse, listicle bloat) and raise real quality through specificity and varied sentence rhythm. Use when polishing blog posts, newsletters, docs, landing-page copy, or any prose that will be published on the web. English/general companion to humanize-korean.
origin	harness
workloads	["writing"]

Humanize Writing

Make published prose sound like a specific human, not a generic model. This is the English/general counterpart to humanize-korean.

Core Principle (read this first)

Optimize for genuine quality, not for beating AI detectors. Detectors measure perplexity (how predictable word choice is) and burstiness (how much sentence length varies); LLM prose scores low on both. But strong academic and technical prose also scores low, so "detector-gaming" can degrade good writing. Fix the underlying tells and the detector score follows — never the other way around.

Corollary: no single token is a reliable tell. The em-dash is the classic example — plenty of human essayists lean on it. Cut overuse, do not ban the character.

When to Activate

• Editing a draft (yours or an LLM's) destined for a blog, newsletter, README intro, docs, or marketing copy.
• The user says the text "sounds like AI," "sounds generic," or "needs a human voice."
• Final polish pass before publishing web content.

Patterns to Cut (the reliable tells)

• Filler openers: "In today's fast-paced world," "In the ever-evolving landscape of," "It's important to note that," "When it comes to."
• Flagged vocabulary (over-represented in LLM output): delve, underscore, tapestry, testament, realm, navigate (figurative), leverage, robust, seamless, elevate, unlock, embark, foster, showcase, pivotal.
• Uniform rhythm — the #1 structural tell. Runs of sentences all 15–22 words long.
• Empty transitions: stacked "Moreover / Furthermore / Additionally"; reflexive "That said,".
• Listicle bloat — every idea forced into a bulleted list of three with bolded lead-ins.
• Over-hedging: "can help to," "may potentially," "it's worth considering."
• Symmetry tics: the rule of three everywhere; "It's not just X, it's Y."
• Hollow conclusions: "In conclusion, X remains a powerful tool that…"; ending on a rhetorical question.
• Em-dash overuse specifically — cap the frequency; substitute periods, commas, or colons.

Positive Craft (what to add)

1. Vary sentence length deliberately. Follow a 30-word sentence with a 4-word one. This raises burstiness honestly.
2. Add specificity only a human would know — concrete numbers, names, dates, a real anecdote or edge case. Raises perplexity honestly.
3. Commit to a voice — a clear opinion, mild asymmetry, an occasional fragment or contraction.
4. Cut throat-clearing — delete the first sentence of most paragraphs; it is usually filler.
5. Prefer the surprising-but-right word over the most probable one. Kill clichés.

Editing-Pass Checklist (run in order)

• Delete filler openers and stacked transitions.
• Merge/split sentences to break any run of uniform length.
• Swap 3–5 flagged words for plainer or more precise ones.
• Replace at least one generic claim with a concrete detail (number, name, example).
• Rewrite the conclusion to end on a specific point, not a summary.
• Read it aloud once; fix anything you would not actually say.

Review Lens

To audit rather than rewrite, scan the draft against the "Patterns to Cut" list and report the top offenders with line references (the "unslop" pattern popularized by community ai-writing-auditor skills). Then apply the checklist.

Reference

Perplexity/burstiness framing: McGill Office for Science & Society; CUNY writing resources. Treat SEO-vendor "beat the detector" pages as directional only — the goal is a human reader's respect, not a classifier's score.