Humanizer Language and Grammar Patterns: The 7 AI Cues Targeted for Natural Rewrites

Humanizer targets seven specific language and grammar patterns—including overused AI filler words, passive voice constructions, and forced rhetorical triads—to detect and rewrite mechanical prose into authentic human text.

The blader/humanizer repository defines a curated set of language and grammar patterns that identify AI-generated prose characteristics. These patterns are documented in the project's README.md (lines 68‑79) and SKILL.md (lines 110‑138), serving as the detection logic for the Claude Skill’s rewrite engine.

The 7 Targeted Language and Grammar Patterns

According to the source documentation, Humanizer applies transformations after content-level checks to eliminate seven specific linguistic cues. Each pattern targets a distinct marker of synthetic writing.

1. Overused AI Words

This pattern flags frequent AI-typical filler words such as actually, additionally, crucial, enhance, and showcase. The rewrite removes these hedges to produce direct statements.

• Example transformation: "Actually, the system enhances performance." → "The system improves performance."

2. Avoiding "is" and "are"

Humanizer detects nominalizations that replace simple copular verbs—phrases like "serves as," "features," or "boasts"—and restores concise "is/are" constructions.

• Example transformation: "Gallery 825 serves as LAAA’s exhibition space." → "Gallery 825 is LAAA’s exhibition space."

3. Not X but Y and Clipped Endings

This cue catches overused contrast constructions (not just... but...) and terse negative endings. The rewrite unravels the forced dichotomy into straightforward description.

• Example transformation: "It’s not just a song, it’s a statement." → "The song conveys a clear statement."

4. Forced Groups of Three

Artificial triads ("innovation, inspiration, and insights") signal AI-generated rhythm. Humanizer collapses these into natural pairs or singular concepts.

• Example transformation: "The event offers innovation, inspiration, and insights." → "The event offers innovation and insight."

5. Changing Names and Repeated Openings

Inconsistent entity naming (switching between "the protagonist," "the main character," and "the hero") and repetitive sentence beginnings trigger this pattern. The rewrite standardizes terminology.

• Example transformation: "The protagonist… The main character… The hero…" → "The protagonist…"

6. False "from X to Y" Ranges

This detects impractical ranges (from X to Y) that do not form true intervals, such as "From the Big Bang to dark matter." The rewrite converts these into logical lists or scopes.

• Example transformation: "From the Big Bang to dark matter." → "Covers the Big Bang, star formation, and dark matter."

7. Passive Voice and Missing Subjects

Passive constructions that hide the actor are flagged and converted to active voice with explicit subjects.

• Example transformation: "No configuration file needed." → "You do not need a configuration file."

CLI Installation and Usage

Install Humanizer via the Skills CLI to apply these language and grammar patterns to your content.

# Install globally via npx

npx skills add blader/humanizer --global

For Claude Desktop users:

/plugin marketplace add blader/humanizer
/plugin install humanizer@humanizer

Practical Transformation Examples

The following snippets demonstrate how Humanizer applies the seven patterns to real inputs.

Overused AI words cleanup:

Input:  Actually, the new algorithm enhances performance significantly.
Output: The new algorithm improves performance significantly.

Restoring "is/are" verbs:

Input:  Gallery 825 serves as LAAA's exhibition space.
Output: Gallery 825 is LAAA's exhibition space.

Resolving Not X but Y constructions:

Input:  It's not just a song, it's a statement.
Output: The song conveys a clear statement.

Breaking forced triads:

Input:  The conference featured keynote sessions, panel discussions, and networking opportunities.
Output: The conference featured keynotes, panels, and networking.

Standardizing entity names:

Input:  The protagonist ... The main character ... The hero ...
Output: The protagonist ...

Correcting false ranges:

Input:  From the Big Bang to dark matter, ...
Output: Covers the Big Bang, star formation, and dark matter.

Activating passive voice:

Input:  No configuration file needed.
Output: You do not need a configuration file.

Source Code Structure

The rewrite logic and language and grammar patterns are defined in two core files:

• SKILL.md (lines 110‑138): Contains the complete skill definition, including all 35 detection patterns with the seven language-specific cues enumerated above.
• README.md (lines 68‑79): Provides human-readable documentation summarizing the grammar patterns in tabular format for quick reference.

These files dictate how Humanizer parses input text, identifies synthetic linguistic markers, and generates concisely rewritten output without altering factual content.

Summary

• Humanizer identifies seven distinct language and grammar patterns that characterize AI-generated text.
• Overused AI words and forced groups of three target mechanical rhythm and filler vocabulary.
• Avoiding "is/are", Not X but Y, and false "from X to Y" ranges correct unnatural syntactic constructions.
• Changing names and passive voice patterns enforce consistency and clarity by standardizing references and activating subjects.
• Pattern definitions reside in SKILL.md and README.md, applied via the Claude Skills CLI.

Frequently Asked Questions

What specific words does Humanizer flag as overused AI terms?

Humanizer targets high-frequency AI filler vocabulary including actually, additionally, crucial, enhance, and showcase. These words appear in the pattern definition within SKILL.md and are stripped or replaced with direct alternatives during rewriting.

How does Humanizer handle passive voice detection?

The tool identifies sentences with missing subjects or passive constructions (e.g., "No configuration file needed") and rewrites them into active voice with explicit actors (e.g., "You do not need a configuration file"). This pattern is enumerated as the seventh grammar rule in the source documentation.

Can I customize which language and grammar patterns Humanizer applies?

The current implementation in blader/humanizer applies the full set of 35 patterns defined in SKILL.md, including the seven language-specific cues. Users invoke the skill via CLI arguments, but individual pattern toggling requires modifying the skill definition file directly before installation.

Where are the transformation examples documented?

concrete input and output examples for each of the seven patterns are documented in README.md (lines 68‑79) and demonstrated within the skill’s usage instructions in SKILL.md (lines 110‑138), providing authoritative reference material for expected rewrite behavior.

{
  "mcpServers": {
    "instagit": {
      "command": "npx",
      "args": ["-y", "instagit@latest"]
    }
  }
}