# Humanizer

> Rewrite text so it reads like a real person wrote it, stripping the telltale patterns of AI-generated writing (puffery, weasel wording, "not just X but Y", rule-of-three, em-dash overuse, AI vocabulary) based on Wikipedia's "Signs of AI writing" field guide. Use when the user says "humanize", "make this sound human", "de-AI this", "this sounds like AI", "make it natural", or runs /humanizer with or before a block of text.

- Skill: `rohitguta2432/humanizer` (Agent Skill, multi-file: 2 files)
- Install (CLI): `npx skillmds@latest add rohitguta2432/humanizer`
- Raw SKILL.md: https://api.skillmd.com/api/skills/rohitguta2432/humanizer/raw
- Safety review: pending
- Works with: Claude Code, Claude.ai, OpenAI Codex
- Category: Marketing & Growth
- Author: rohitguta2432 (https://skillmd.com/u/rohitguta2432)
- Updated: 2026-09-17
- Page: https://skillmd.com/skills/rohitguta2432/humanizer

---


# Humanizer

Rewrite the given text so it stops sounding AI-generated. Keep the meaning,
facts, and the writer's intent exactly the same — only remove the patterns
that give away machine writing.

## Workflow

1. Read the input text. If the user gave no text, ask for it.
2. Scan it against every rule in [PATTERNS.md](PATTERNS.md).
3. Rewrite. Preserve meaning, claims, names, numbers, and links verbatim.
   Do not add new facts, do not soften or strengthen claims.
4. Output the rewritten text first, then a short "what I changed" list
   (one line per pattern fixed, only patterns actually found).

## Core rules (full detail in PATTERNS.md)

**Content**
- No significance inflation: cut "stands as a testament", "plays a vital
  role", "rich cultural heritage", "continues to captivate", "-ing" tails
  that claim broader importance ("...highlighting its enduring impact").
- No weasel attributions: "industry reports suggest", "many observers
  note", "is widely regarded as" — either name the source or drop it.
- No empty "challenges and future prospects" closers or summary paragraphs
  that restate the piece in vaguer words.
- No press-release tone: "boasts", "seamlessly", "cutting-edge", "unlock".

**Language**
- Purge AI vocabulary: delve, crucial, pivotal, landscape, tapestry,
  testament, underscore, foster, showcase, leverage, meticulous, vibrant,
  intricate, robust, holistic, enhance, highlighting, emphasizing.
- Use plain copulas. Write "X is Y", not "X serves as Y" / "X functions
  as a Y" / "X represents Y".
- Kill negative parallelisms: "not just X, but Y", "it's not about X,
  it's about Y", "more than just X". Say the actual point once.
- Break the rule of three. "fast, simple, and reliable" triads everywhere
  are a tell — keep one item or restructure the sentence.
- Stop elegant variation: call a thing by the same name twice; don't
  rotate synonyms ("the protagonist... our hero... the young warrior").

**Style & formatting**
- Em dashes: allow at most one per ~150 words; replace the rest with
  commas, periods, or parentheses.
- Sentence case for headings, not Title Case.
- No bold-keyword sprinkling, no emoji bullets, no "**Term:** definition"
  inline-header lists unless the user's original had them.
- Straight quotes, not curly, in technical/plain-text contexts.
- No chatbot filler: "Certainly!", "Great question", "I hope this helps",
  "In conclusion", "It's important to note that".

**Voice**
- Vary sentence length; let some sentences be short. Fragments are fine.
- Prefer concrete specifics over generic abstractions — if the original
  had a specific fact buried under adjectives, dig the fact out and lead
  with it.
- Keep contractions if the register is informal.

## Standing mode

If the user says "humanize everything from now on" (or similar), apply
these rules to all prose you write for them for the rest of the session —
not just explicit rewrite requests.

## What NOT to change

- Meaning, claims, facts, numbers, names, quotes, links, code.
- The user's own voice: if their draft is formal, stay formal.
- Correct information that merely *sounds* fancy but is accurate and
  specific — specificity is human; only cut the generic gloss.

