Signs you're using AI to write (and how to fix it)

# Writing Humanizer

Transform a piece of writing so it reads like a person wrote it: clear, direct, unpadded, and in a voice someone would actually use. Deliver the revised text, not commentary about it. The hardest constraint: humanizing means removing the artifacts, not the author. If the source has a voice, the output should sound like that person on a good day, not like a different, blander person.

## Guidelines

### Clarity and economy
1. Every sentence should be understandable on first read.
2. Get to the point. Cut words that don't earn their place.
3. Short sentences, plain words. Complexity only where the idea is complex.
4. Cut filler openings: "The deadline is approaching," not "It's important to note that the deadline is approaching."
5. One idea, said once, well. No redundancy.
6. Active voice by default. Passive only when the actor genuinely doesn't matter.

### Voice and tone
1. Natural, conversational. Starting sentences with "And" or "But" is fine.
2. Relaxed grammar where the register calls for it. Don't sand a casual draft into corporate polish.
3. Address readers directly where the form allows: "you" and "your."
4. Honest beats friendly-sounding. No forced warmth, no exaggeration.
5. No marketing hype. "This can help you" not "this revolutionary tool will transform your workflow."
6. Hedge only when genuinely unsure. When sure, say "this works" not "this might potentially work." False confidence is its own tell, but so is over-hedging.

### Sentence rhythm
1. Vary structure deliberately. Mix short, medium, and long. Monotony is one of the clearest machine signals.
2. Avoid symmetrical three-part lists ("clarity, brevity, and impact") as a default rhetorical move. Use them only when the content genuinely has three parts.
3. Avoid the "It's not X, it's Y" construction. It reads as AI's favourite rhetorical move and now flags immediately.
4. Don't stack aphorisms. One earned insight is fine; a paragraph of tweetable wisdom is a tell.
5. Vary how paragraphs open. AI defaults to topic sentence, support, transition, every time. Break the pattern.

### Words and phrases to cut entirely
These are overrepresented in AI output. Remove them unless the source author uses them naturally and consistently:

Verbs: delve, delving, underscore, underscoring, showcase, showcasing, leverage, utilize, harness, streamline, facilitate, empower, navigate (metaphorical), illuminate, bolster, foster, elevate, align (abstract), unpack, unlock, dive into, embrace, explore (vague)

Adjectives/adverbs: pivotal, intricate, meticulous/meticulously, robust, seamless, vibrant, dynamic, comprehensive, multifaceted, nuanced (as a free-floating compliment), holistic, cutting-edge, transformative, groundbreaking, unparalleled, profound (inflated), ever-evolving, ever-changing, unwavering

Nouns used as borrowed grandeur: tapestry, landscape (metaphorical), realm, mosaic, ecosystem (metaphorical), symphony, labyrinth, beacon, cornerstone, bedrock, testament, odyssey

Transition/filler phrases:
- "It's important to note that…"
- "It's worth noting / mentioning that…"
- "That being said…"
- "In today's fast-paced world…"
- "In an ever-evolving landscape…"
- "Navigating the complexities of…"
- "At its core…" / "At the heart of…"
- "When it comes to…"
- "In the realm of…"
- "Plays a vital / pivotal / crucial role in…"
- "Stands as a testament to…"
- "Let's unpack this" / "Let's break it down"
- "Sit with that for a moment"
- "A nuanced understanding of…"

Closing ritual phrases (replace with actual endings):
- "In conclusion…" / "In summary…" / "Overall…" / "Ultimately…"
- "As we navigate X, it's essential that…"
- "The journey doesn't end here…"
- Any preachy final sentence about what the reader should remember or feel

### What else to remove
1. Clichés, jargon, hashtags, decorative emoji.
2. Most semicolons. They usually signal a sentence that wants to be two.
3. Em dashes used as a habit. Use a full stop, comma or brackets instead.
4. Adjectives and adverbs that add enthusiasm, not information.
5. Rhetorical questions used as section openers ("What does this mean for your business?").

## Process
1. Read the source for intent and voice: what it's trying to do, what must survive, and what the author sounds like when they're not performing.
2. One pass: structure first (cut redundancy, reorder for directness), then sentences, then words.
3. Read the flagged-word list and clear every hit unless it genuinely belongs to this author's voice.
4. Read for rhythm. Does sentence length vary? Do paragraph openings vary? Fix monotony.
5. Final read: would a careful reader flag any sentence as machine-written or as padding? Fix those. Then stop. Over-editing removes the author.
6. Check against target length and tone preferences if given.

## Discipline
- Meaning is invariant. A cut that changes what the text claims is a bug, not a feature.
- Don't flatten voice into "clean." A quirky, specific sentence that breaks a guideline beats a smooth generic one that follows it. The guidelines serve the voice, not the reverse.
- Personal specificity beats polish. References to actual experiences, concrete details, and imperfect phrasing all read as human. Preserve them.
- Deliver the revised text first. A short "key changes" note only if the edits are surprising or the user asked.

/plugin marketplace add blader/humanizer

/plugin install humanizer@humanizer