| name | humanize |
| description | Strip AI writing patterns from content stored in Payload CMS. Use when asked to "humanize", "de-AI", "make this sound human", "clean up the writing", or review content for AI patterns. Runs a two-pass rewrite on content fetched via MCP. |
| argument-hint | 'service <name>' or 'guide <slug>' |
Humanize Skill
Remove AI writing patterns from content in Payload CMS. Two-pass process: rewrite, then audit.
AI Pattern Detection List
These patterns signal AI-generated text. Find and fix all of them.
Structural patterns
- Rule of three: "X, Y, and Z" lists appearing more than once per 200 words
- Parallel sentence openings: 3+ consecutive sentences starting the same way
- Symmetric paragraphs: Every paragraph roughly the same length
- Formulaic transitions: "Moreover," "Furthermore," "Additionally," at paragraph starts
- Setup-punchline structure: "While X may seem Y, the reality is Z"
Word-level patterns
- Inflated vocabulary: "utilize" (use), "facilitate" (help), "leverage" (use), "implement" (do/set up), "comprehensive" (specific list instead), "robust" (describe what it actually does)
- Hedging clusters: "It's important to note that," "It's worth mentioning," "It should be noted"
- Vague intensifiers: "very," "really," "incredibly," "extremely," "significantly"
- AI-favorite words: "delve," "landscape," "tapestry," "realm," "paradigm," "synergy," "ecosystem" (unless technically accurate), "holistic," "seamless," "cutting-edge," "game-changer"
- Promotional adjectives: "revolutionary," "groundbreaking," "innovative," "state-of-the-art," "best-in-class"
Tone patterns
- Excessive enthusiasm: Multiple exclamation marks, "exciting," "amazing," "incredible"
- False certainty: "undoubtedly," "without question," "clearly" (when the thing isn't clear)
- Anthropomorphization: "the tool understands," "the platform knows," "the service cares"
- Reader flattery: "as a savvy user," "for discerning professionals"
- Empty empathy: "we understand how frustrating," "we know how important"
Punctuation patterns
- Em dash overuse: More than 1 per 500 words (our rule: zero em dashes)
- Semicolons: Remove all. Split into two sentences.
- Ellipses for drama: "And then... everything changed."
Process
Step 1: Fetch the content
For services: mcp__payload__findServices with name search, get description, content, features, issues.
For guides: mcp__payload__findGuides with slug search, get intro, beforeYouStart, steps, troubleshooting, outro.
Step 2: First pass — Rewrite
Go through every text field. For each one:
- Find every instance of the 18 patterns above
- Rewrite to sound natural. Shorter sentences. Specific details instead of adjectives.
- Vary sentence length: mix 6-word and 20-word sentences
- Use contractions naturally ("it's", "don't", "won't")
- Replace any "not just X, but also Y" with two plain statements
Step 3: Second pass — Audit
Read the rewritten version and ask: "What still makes this obviously AI-generated?"
Fix those remaining issues. Common second-pass catches:
- Too-perfect paragraph structure (add one short 1-sentence paragraph)
- No contractions (add 2-3 natural ones)
- Every sentence follows subject-verb-object (vary with an occasional "The hard part:" or "In practice:")
- No personality (add one slightly opinionated observation with "Worth knowing:" or "The catch:")
Step 4: Save back to Payload
Use mcp__payload__updateServices or mcp__payload__updateGuides to save the rewritten content. Only update the text fields that changed.
Step 5: Report
Tell the user:
- How many patterns were found and fixed (by category)
- Which fields were updated
- Any remaining concerns
Verification
After humanizing, the content should pass these checks:
