LinkedIn Humanizer (linkedin-humanizer)

The linkedin-humanizer skill inspects raw or AI-generated LinkedIn drafts, identifies dead giveaways of machine generation, and refactors the copy into authentic, practitioner-grade human voice. It combines rigorous lexical filtering, sentence-structure variance analysis, and a multi-detector simulation framework.

CRITICAL DISCLAIMER: This skill optimizes for genuine editorial authority, human rhythm, and reader trust. It does NOT promise or guarantee 100% bypass of automated AI detection software (e.g. GPTZero, Originality, CopyLeaks). Third-party AI detectors are probabilistic, frequently update their heuristics, and generate false positives even on human text. The primary objective is to stop alienating human readers with obvious robotic patterns and generic corporate speak.

When to use this skill

Ideal Use Cases:

• Polishing drafts generated by ChatGPT, Claude, or other LLMs that sound stiff, overly formal, or robotic.
• Scrubbing corporate jargon, performed sincerity, and cringe broetry from team drafts.
• Elevating average writing into senior operator voice with concrete metrics, varied cadence, and natural flow.
• Pre-flight quality checking before loading posts into a scheduling pipeline.
• Auditing employee advocacy submissions to ensure authenticity.

Disambiguation ('Not for X, use Y'):

• Not for creating a new post from scratch -> use linkedin-post-writer.
• Not for rewriting a YouTube video or article -> use linkedin-repurposer.
• Not for reverse-engineering a viral hook template -> use linkedin-hook-extractor.
• Not for drafting comments on others' posts -> use linkedin-comment-drafter.
• Not for conducting founder interview sessions -> use linkedin-interviewer.

The AI Tells Catalog: Patterns to Eradicate

Every audit systematically scans for and removes the following 7 deadly machine-writing patterns:

1. Banned Machine Vocabulary & Replacement Table

Large language models compulsively revert to specific academic and corporate jargon. The humanizer enforces a strict zero-tolerance replacement table:

2. Staccato Fragment Stacks (Broetry Spam)

• The Tell: 10 to 15 consecutive single-sentence or 3-word paragraphs intended to manipulate mobile scrolling.
• The Problem: Disdained by senior executives and penalised by modern LinkedIn readability ranking.
• The Fix: Group related thoughts into natural 2-to-3 sentence paragraphs with coherent syntactic flow.

3. Em Dash (—) Overuse

• The Tell: Inserting em dashes in every second sentence to join thoughts (e.g., "Speed is critical—but so is quality—and teams that forget this fail").
• The Problem: LLMs use em dashes at 10x the frequency of human writers.
• The Fix: Hard cap of maximum one em dash per entire post. Zero preferred. Use parentheses, commas, or period stops.

4. Unearned Stacked Triads

• The Tell: Compulsive listing of three parallel adjectives or noun phrases in almost every sentence:
  • "It was fast, reliable, and scalable."
  • "We built it with passion, precision, and purpose."
• The Fix: Cut down to one or two strong words, or replace with a concrete metric.

5. Performed Sincerity Openers

• The Tell: False humility designed to simulate vulnerability:
  • "I am humbled and thrilled to announce..."
  • "Reflecting deeply on my journey today..."
  • "I never thought I would share this publicly, but..."
• The Fix: Delete the opener entirely. Start directly inside the action or the primary metric.

6. The Formulaic "Bookend" Conclusion

• The Tell: Closing every post with theatrical pseudo-profundity:
  • "Let that sink in."
  • "Read that again."
  • "Agree? Drop your thoughts below."
• The Fix: End with a genuine, low-friction diagnostic question or a clean takeaway sentence.

7. Monotonous Sentence Rhythm (Low Burstiness)

• The Tell: Every sentence having almost identical syllable count and grammatical structure (Subject-Verb-Object).
• The Fix: Intermix 4-word punchy statements with 20-word compound sentences with dependent clauses.

The Multi-Detector Simulation Framework

When analyzing a text, linkedin-humanizer simulates three primary statistical dimensions used by detection engines:

1. Perplexity Index (Predictability):
   • Evaluates whether words follow common statistical LLM n-gram patterns.
   • Low perplexity = predictable machine text. High perplexity = idiosyncratic human word choice.
2. Burstiness Variance (Sentence Length Standard Deviation):
   • Measures variability in sentence length.
   • Low variance (SD < 4 words) = uniform machine pacing.
   • Healthy human variance (SD > 8 words) = dynamic conversational rhythm.
3. Lexical Fingerprint Density:
   • Calculates the ratio of flagged corporate/AI trigger words per 100 words.
   • Target benchmark: < 0.1% flagged words.

How It Works Step-by-Step

When executing linkedin-humanizer, follow this 5-stage transformation pipeline:

1. Diagnostic Scan:
   • Ingest draft text.
   • Scan for the 7 deadly tells and calculate current sentence length standard deviation and tell density.
2. Structural Pruning:
   • Delete performed sincerity openers and formulaic bookend endings.
   • Collapse excessive single-line breaks into natural paragraph blocks.
3. Lexical Replacement:
   • Substitute all banned AI vocabulary (delve, tapestry, leverage, etc.) with grounded synonyms.
   • Strip redundant em dashes and dissolve stacked triads.
4. Burstiness Tuning:
   • Rewrite sentences to establish dynamic rhythm: blend short declarative statements with rich analytical descriptions.
   • Add concrete grounding: insert specific tools, versions, dollar amounts, or timeframe anchors.
5. Humanizer Audit Report:
   • Output side-by-side comparison (Before vs After).
   • Display the diagnostic audit log showing removed tells and rhythm improvements.

Approval Gate

The humanizer produces an audit report and requires user confirmation before saving or staging the revised copy:

[HUMANIZER AUDIT COMPLETE]
Original Word Count: 184 words | Revised Word Count: 162 words
Flagged Tells Removed: 4 (delve, leverage, 3 em dashes, 1 performed sincerity opener)
Burstiness Score (Sentence Length SD): Improved from 3.2 words -> 8.7 words

Draft Comparison:
-------------------------------------------------------------------------------
[ORIGINAL]:
"I am humbled to share that today we delve into the intricate tapestry of modern cloud architectures. Teams must leverage containerization—not merely as a trend—to achieve agility, speed, and resilience. Read that again."

[HUMANIZED]:
"Last month we audited our cloud infrastructure after our AWS bill hit $14,000. Most teams adopt Kubernetes because they assume containerization automatically brings agility. In our case, running 12 microservices for a 5-person team just added 20 hours of YAML debugging every sprint."
-------------------------------------------------------------------------------

Approval Token: c9910a72-6712-4ee8-90ab-123456789abc
Status: READY FOR USER CONFIRMATION.

Examples

Example 1: Eliminating Jargon and Stacked Triads

User Input: "In today’s hyper-connected business landscape, organizations must leverage cutting-edge AI tools to foster growth, streamline workflows, and drive innovation. We must delve deep into these paradigms to unlock true synergy."

[Humanized Output]
Most software teams buying AI developer tools are wasting their budget.

Adding 6 new copilot subscriptions doesn't fix unclear specifications or broken code reviews. If your senior engineers spend half their week in status meetings, automation only helps them write unreviewed code faster.

Before adding new tooling to your stack, fix the review bottlenecks first.

Example 2: Fixing Em Dash Overuse and Broetry

User Input: "Sales is hard.

Really hard.

But you know what makes it harder?

Bad data.

When your CRM is messy—and your reps are confused—your revenue suffers—every single quarter.

Agree?"

[Humanized Output]
The quickest way to burn through outbound SDR payroll is letting reps prospect in an unvalidated CRM.

When 30% of your email lists bounce and contact titles are two years out of date, even your best closers end up spending 15 hours a week doing manual spreadsheet cleanup instead of speaking with buyers.

Before hiring another rep, audit your pipeline data accuracy.

Example 3: Rewriting an Executive Announcement

User Input: "I am beyond thrilled and humbled to announce that our team has embarked on a pivotal journey to revolutionize customer onboarding. This milestone stands as a testament to our relentless passion, commitment, and dedication. Let that sink in."

[Humanized Output]
We just deployed our new customer onboarding flow after 8 weeks of testing.

Our previous setup forced new users through a 9-step registration wizard that took an average of 14 minutes to complete. The new flow gets users directly into their active workspace in under 90 seconds.

Early metrics from the first 200 accounts show activation rates climbing from 48% to 81%.