Humanizer: Remove 24 AI Writing Patterns and Add Real Voice to Your Text

44,231 downloads and 348 stars — the Humanizer skill by @biostartechnology is the second most downloaded skill on ClawHub. That number tells you something real: a lot of AI-generated content is getting used in contexts where it can't sound like AI-generated content.

This skill doesn't just scan for buzzwords. It applies a systematic 24-pattern detection framework derived from Wikipedia's actively-maintained "Signs of AI writing" guide — and then it pushes further, requiring the agent to inject actual personality into the result.

The Problem It Solves

AI text has a fingerprint. Not just the obvious ones — "delve into," "I cannot and will not," "as an AI language model" — but subtler patterns that aggregate into something that just reads as machine-generated:

• Sentences all the same length and structure
• Neutral on everything, opinionated on nothing
• Significance inflation ("marks a pivotal moment in...")
• Copula avoidance ("serves as" instead of "is")
• Em dashes everywhere, in threes
• Vague attributions ("experts believe")

Even technically clean text fails the human test if it has no voice. The Humanizer addresses both problems.

How It Works

The skill loads a comprehensive editing framework into your AI agent. When you ask it to humanize text, it runs two passes:

1. Pattern detection — Scans for all 24 documented AI writing signals
2. Voice injection — Adds opinions, varied rhythm, first-person perspective, and specificity

The 24 patterns split across four categories:

Content Patterns (6)

• Significance inflation — "marking a pivotal moment" → specific facts
• Notability name-dropping — citing sources without context
• Superficial -ing analyses — "symbolizing... reflecting..."
• Promotional language — "nestled within the breathtaking..."
• Vague attributions — "Experts believe..."
• Formulaic challenge framing — "Despite challenges... continues to thrive"

Language Patterns (6)

• AI vocabulary words — "Additionally... testament... landscape..."
• Copula avoidance — "serves as" instead of "is"
• Negative parallelisms — "It's not just X, it's Y"
• Rule of three — forcing ideas into groups of three
• Synonym cycling — excessive substitution
• False ranges — "from X to Y" on non-meaningful scales

Style Patterns (6)

• Em dash overuse
• Boldface overuse
• Inline-header lists
• Title Case Headings
• Emoji decoration
• Curly quotation marks

Communication Patterns (6)

• Chatbot artifacts — "I hope this helps!"
• Cutoff disclaimers — "While details are limited..."
• Sycophantic tone — "Great question!"
• Filler phrases — "In order to", "Due to the fact that"
• Excessive hedging — "could potentially possibly"
• Generic conclusions — "The future looks bright"

The Soul Problem

What makes this skill interesting is what it does beyond the checklist. The skill prompt includes an explicit section called "PERSONALITY AND SOUL" — because removing AI patterns without adding voice just produces sterile text, which is a different kind of obvious.

The framework tells the agent to:

• Have opinions — "I genuinely don't know how to feel about this" beats neutral reporting
• Vary rhythm — Short punchy sentences. Then longer ones that take their time getting where they're going.
• Acknowledge complexity — Real humans have mixed feelings
• Use "I" when it fits — First person signals a real person thinking
• Be specific about feelings — Not "this is concerning" but "there's something unsettling about agents churning away at 3am"

Before and After

Before (AI-sounding):

The new software update serves as a testament to the company's commitment to innovation. Moreover, it provides a seamless, intuitive, and powerful user experience—ensuring that users can accomplish their goals efficiently.

After (humanized):

The software update adds batch processing, keyboard shortcuts, and offline mode. Early feedback from beta testers has been positive, with most reporting faster task completion.

The difference: specifics over abstractions, plain verbs over "serves as," no em dash, no "Moreover."

Usage

Install and invoke:

clawhub install humanizer

Then in your agent:

Please humanize this text: [paste your text]

Or integrate it into editing workflows:

Review this draft blog post and humanize it. Focus especially on the introduction and conclusion — those tend to be the most AI-sounding sections.

The skill uses Read, Write, Edit, Grep, and Glob tools — so it can work directly on files in your project:

Humanize the content in docs/blog-draft.md and save the result back to the same file.

Comparison

Practical Tips

Focus on the bookends. AI patterns are most concentrated in introductions (preamble, significance inflation) and conclusions (generic wrap-ups, "The future looks bright"). Humanize those first.

Preserve technical content. The skill is instructed to maintain meaning and voice. For technical sections with precise terminology, tell it explicitly: "Don't change the code blocks or technical specs, just humanize the surrounding prose."

Batch process documents. Since it has Glob and file access, you can point it at a directory:

Humanize all .md files in the content/ directory. Process them one at a time and save each result.

Run it after your first AI draft, not instead of writing. The best results come from AI-drafted content that goes through Humanizer and then a light human pass — not from expecting fully finished output.

Check for overcorrection. Aggressive humanization can introduce inconsistencies in tone. Ask the skill to preserve the intended audience (formal vs. casual) explicitly.

Considerations

• Context window cost — The 24-pattern framework is detailed. For very short texts, you're loading a large instruction set for a small task.
• Subjectivity — "Voice" is inherently subjective. The skill follows the Wikipedia guide, which reflects one community's consensus. It may not match every brand's tone guide.
• Not a detection bypass — This is for making text read naturally to humans, not for fooling AI detectors. Detector bypass and human-quality writing are different goals.
• Scope creep — On long documents, the skill might over-edit. Giving it specific sections works better than a whole 10,000-word document at once.

The Bigger Picture

The Humanizer skill reflects something uncomfortable about the current AI writing moment: we generate more text than we can individually voice-check, and the patterns are legible enough that readers notice even if they can't name exactly what's wrong.

At 44,000+ downloads, it's clearly a workflow that a lot of people have settled into — generate with AI, humanize with this skill, polish by hand. Whether that's ideal is a separate conversation. As a tool, it does exactly what it says.

View the skill on ClawHub: humanizer