/humanizer — Rewrite Existing Text to Remove AI Patterns
Note: The humanizer rules are ALWAYS ON via
.claude/rules/writing-quality.md. Every external draft is written with these patterns in mind automatically. This/humanizercommand exists as a manual tool for rewriting text that was already written — by another AI, pasted from somewhere, or from an earlier draft that slipped through.Think of it this way: the rule file is DNA (always present in every cell). This skill is a tool (you pick it up when you need it for a specific job).
Voice DNA
Step 1: Load personal/voice-dna.md. If it exists, that IS the user's voice. Match it exactly.
Step 2: Read the room. The same person writes differently to a client vs a friend vs their team. Match the context:
| Context | Register |
|---|---|
| Client email | Professional but still them. No corporate stiffness unless that IS their style. |
| Friend / casual | How they'd text. Short, loose, personality showing. |
| LinkedIn / public | Their brand voice. Confident, specific, opinionated. |
| Internal / team | Direct, efficient. Skip the niceties unless that's their culture. |
| Formal (legal, board, investor) | Measured and precise. Still them, just buttoned up. |
Step 3: If no voice-dna.md exists, use these defaults until one is built:
- Direct. No warm-up paragraphs.
- Specific. Name the number, the person, the date.
- Conversational. Write like talking to a smart colleague.
- Vary rhythm. Short sentences. Then a longer one. Fragments are fine.
The 25 AI Writing Patterns
Detect and fix all of the following. Based on Wikipedia's "Signs of AI Writing" guide.
Content Patterns
1. Significance inflation. Words like "testament", "pivotal moment", "vital role", "indelible mark", "setting the stage", "deeply rooted." AI puffs up importance. Cut it. State the fact.
- Before: "This marks a pivotal moment in the evolution of modern analytics."
- After: "We shipped cross-source data aggregation. Users can now query both providers in one search."
2. Notability inflation. Listing media coverage or follower counts without context. If you cite coverage, say what was said, not just where.
- Before: "Featured in multiple industry publications."
- After: "TechReview covered our aggregation approach in their March issue, calling it 'the first practical implementation.'"
3. Superficial -ing analyses. Tacking "-ing" phrases onto sentences for fake depth: "highlighting", "underscoring", "symbolizing", "showcasing", "fostering", "ensuring."
- Before: "The platform aggregates data from multiple sources, showcasing how integration enhances transparency."
- After: "The platform pulls data from multiple sources into one view."
4. Promotional language. "Groundbreaking", "vibrant", "stunning", "breathtaking", "nestled", "boasts", "renowned." Describe what the thing does, not how impressive it is.
- Before: "The product boasts a groundbreaking approach to data management."
- After: "The product pulls data from multiple providers into one searchable database."
5. Vague attributions. "Experts say", "industry observers note", "some critics argue." Name the person or drop the claim.
- Before: "Industry experts agree that data fragmentation is the biggest challenge."
- After: "Our lead engineer, who spent 10 years in the space, calls data fragmentation the core problem."
6. Formulaic "challenges and future" sections. "Despite challenges... continues to thrive." "The future looks bright." State what's actually happening.
- Before: "Despite challenges in data standardization, the company continues to thrive."
- After: "Data formats vary by provider. We normalize them into a common schema."
Language and Grammar Patterns
7. Overused AI vocabulary. Additionally, crucial, delve, enduring, enhance, fostering, garner, highlight (verb), interplay, intricate, landscape (abstract), pivotal, showcase, tapestry (abstract), testament, underscore, valuable, vibrant. Replace with simpler words or cut entirely.
8. Copula avoidance. "Serves as", "stands as", "functions as", "boasts", "features", "offers." Just use "is", "are", "has."
- Before: "The product serves as a single source of truth for operational data."
- After: "The product is a single source of truth for operational data."
9. Negative parallelisms. "It's not just X, it's Y." "Not only... but also..." Overused. Just state the point.
- Before: "It's not just a database. It's a decision-making tool for operators."
- After: "Operators use it to compare options before committing."
10. Rule of three. AI forces ideas into groups of three. Real people don't always think in threes.
- Before: "Speed, accuracy, and transparency."
- After: "It's fast and the data is traceable."
11. Synonym cycling. Using different words for the same thing to avoid repetition: "the platform", "the tool", "the system", "the solution." Pick one and stick with it.
- Before: "The platform aggregates data. The tool normalizes formats. The system resolves conflicts."
- After: "The product aggregates data, normalizes formats, and resolves conflicts across sources."
12. False ranges. "From X to Y" constructions where X and Y aren't on a meaningful scale.
- Before: "From small startups to global enterprises, from simple to complex."
- After: "We work with companies at different scales. Most are mid-market."
Style Patterns
13. Em dash overuse. AI loves em dashes. Use commas, periods, or parentheses instead. One em dash per piece of writing maximum.
14. Excessive bold. Don't bold every key phrase. If everything is emphasized, nothing is.
15. Inline-header vertical lists. Bullet points starting with "Header: description" is an AI tell. Convert to flowing prose or simple bullets without bold headers.
16. Title Case in headings. Use sentence case. "Strategic partnerships" not "Strategic Partnerships."
17. Emojis. Never. Not in emails, not in LinkedIn posts, not anywhere in external communications.
18. Curly quotes. Use straight quotes ("like this") not curly quotes.
Communication Patterns
19. Chatbot artifacts. "I hope this helps!", "Let me know if you'd like me to expand", "Great question!", "Here is an overview of..." Delete these entirely.
20. Knowledge-cutoff disclaimers. "As of my last update", "While specific details are limited..." If you don't know something, say so plainly or don't mention it.
21. Sycophantic tone. "Great question!", "You're absolutely right!", "That's an excellent point!" React to content, not to the person asking.
22. Filler phrases. "In order to" -> "To." "Due to the fact that" -> "Because." "At this point in time" -> "Now." "It is important to note that" -> cut it.
23. Excessive hedging. "It could potentially possibly be argued that it might..." -> State it or don't.
24. Generic positive conclusions. "The future looks bright." "Exciting times ahead." End with something specific or don't end with an outlook at all.
25. Hyphenated word pair overuse. AI hyphenates with perfect consistency. Humans are inconsistent. For common compounds (cross functional, high quality, data driven), drop the hyphen unless it genuinely aids clarity.
Process
The goal is simple: do not sound like AI. Sound like the user.
When the user says /humanizer or asks to humanize a draft:
- Load voice-dna.md. That is the target voice. If it does not exist, use the defaults above.
- Detect context. Who is this for? Client, friend, public, internal? Adjust register.
- First pass. Kill all 25 AI patterns above. Apply the user's voice. Keep the core message.
- Gut check. Read it back. Would a human reading this think AI wrote it? If yes, find why:
- Every sentence the same length?
- No opinion or personality?
- Too tidy, too structured?
- Missing "I" where the user would naturally use it?
- Second pass. Fix what the gut check caught.
- Present the final version. Brief summary of what changed.
Output Format
## Rewrite
[The humanized text]
## What changed
- [Brief bullets listing the patterns that were fixed]
If the original text is clean, say so. Don't rewrite for the sake of rewriting.
Context-Specific Guidelines
Emails to partners / stakeholders
- Be direct. Peers are peers, not customers.
- Skip pleasantries beyond a one-line opener. Get to the point.
- End with a clear ask or next step, not a vague sign-off.
LinkedIn posts
- Hook in the first line. LinkedIn truncates after ~150 characters.
- Write like you're telling a friend what happened. Not like you're writing a press release.
- One idea per post. Don't try to cover everything.
- No hashtag spam. Two or three max, at the end.
Pitch materials (investors, partners, clients)
- Lead with the problem, not the solution.
- Use real numbers. "2,000 customers across 3 markets" beats "rapid growth."
- Name names. Specificity builds trust.
- Acknowledge what doesn't exist yet. Honesty about roadmap beats vaporware.
Investor updates
- Lead with metrics, then narrative. Numbers first, story second.
- Bad news early, not buried. Investors respect candor.
- End with a specific ask if you have one. Don't waste the touchpoint.
Reference
Patterns sourced from Wikipedia: Signs of AI writing, maintained by WikiProject AI Cleanup. Adapted for founder-led project communications.