Lesson 18 — OpenClaw Humanizer Skill: Remove AI Writing Patterns in 5 Steps (2026)

Goal: Install the Humanizer Skill to detect and fix common patterns in AI-generated text, making articles read more naturally — more like a human wrote them.

8 Telltale AI Writing Patterns

Understanding these patterns is essential before installing. AI writing has a highly recognizable set of habits:

#	Pattern Type	Before	After
1	Em dash overuse	This was a critical decision — and one with far-reaching implications	This was a critical decision with far-reaching implications
2	Rule of three structure	The solution is fast, efficient, and reliable	The solution runs fast and keeps costs low
3	Inflated symbolism	A digital transformation wave is sweeping across the entire landscape of industry	More and more companies are going through digital transformation
4	Excessive transition words	Furthermore, it is worth noting that, not to mention that...	(Just make the next point directly)
5	Passive structure stacking	The approach has been proven to be viable and has been widely adopted	Many teams have tried this approach with good results
6	False balance	Of course, this method also has its limitations...	(Don't mention limitations that aren't relevant)
7	Hollow ending	In summary, the above represents our comprehensive exploration of this topic	(Delete it; just end)
8	Excessive capitalization	This is a Major Breakthrough in Artificial Intelligence	This is a new development in AI

ClawHub's 8th most popular Skill (74k downloads).

Step 1: Install the Skill

In WebChat or Telegram, send:

/install @biostartechnology/humanizer

Verify installation:

pnpm openclaw skills list
# humanizer should appear in the list

Step 2: Basic Usage — Paste an Article for Detection

Send the article directly to AI with a detection instruction:

Use the Humanizer Skill to detect AI writing patterns in the following article, list the issues, and fix them:

[paste article content]

The Skill outputs two things:

Detection Report:

AI writing patterns detected:
- Paragraph 2: Em dash overuse (3 instances)
- Paragraph 3: Rule of three structure (fast, efficient, reliable)
- Paragraph 5: Inflated symbolism ("sweeping across the entire industry landscape")
- Closing paragraph: hollow summary sentence
Severity: Medium (4 fixes recommended)

Fixed Version (outputs the complete corrected article)

Step 3: Fix Examples for Each Pattern

Fixing Em Dash Overuse

Before:

This update brings three core improvements — performance, stability, and user experience — each of which
required six months of development — ultimately culminating in this release.

After:

This update took six months and focused on improving performance, stability, and user experience —
three parallel tracks that culminated in this release.

Fixing Inflated Symbolism

Before:

Artificial intelligence is sweeping across the business world at an unprecedented pace,
reshaping the underlying logic and value chains of every industry.

After:

AI is rapidly changing the business landscape, and companies across industries are adjusting their workflows.

Fixing a Hollow Ending

Before:

In summary, the above is a complete introduction to OpenClaw configuration. We hope this has been helpful to you.

After:

(Delete it, or replace with a concrete next step: once configuration is complete, run pnpm openclaw start to verify.)

Step 4: Customize Detection Intensity

Humanizer supports three levels of correction:

Light touch (preserves original style, fixes only the most obvious issues):

Use light touch mode on this article — fix only what most hurts readability, don't make big changes:
[article content]

Standard fix (fixes all detected AI patterns, good for most cases):

Use standard mode to detect and fix AI writing patterns in this article:
[article content]

Deep rewrite (thoroughly rewrites sentence structure, preserves content, drastically changes expression):

Use deep rewrite mode — this article needs to be completely unrecognizable as AI-written:
[article content]

Step 5: Tips for Batch Processing Long Articles

For articles over 3,000 words, process in sections to avoid context window truncation causing inconsistent style across the piece:

I have a 5,000-word article to de-AI, process it in three sections:
Section 1 (parts 1–3):
[content]

After processing all three sections, ask AI for a style consistency check:

I've processed an article in sections, now check the combined version for stylistic consistency —
if any section stands out as noticeably different, flag it:
[combined article]

Batch processing multiple files (combined with Nano PDF or local file reading):

Read all .txt files in ~/Documents/articles/ directory, run each one through Humanizer
standard mode, and save the results to ~/Documents/articles/humanized/

Important Note

The Humanizer Skill is designed to make AI-assisted content read more naturally and improve the reading experience. Please use it responsibly: it's appropriate for improving the readability of published content, not for deceiving systems designed to identify AI-generated text (such as academic integrity checks).

FAQ

What are the telltale signs of AI-written text?

The 8 common AI writing patterns include: em dash overuse (connecting too many clauses with dashes), rule of three structure (always listing three parallel adjectives), inflated symbolism (using grand language for ordinary things), excessive transition words ("furthermore," "it's worth noting" appearing constantly), passive structure stacking, false balance (unnecessary "on the other hand"), hollow endings ("in summary, the above represents..."), and excessive capitalization of nouns. When these patterns appear together, AI detection tools achieve very high hit rates.

Can the Humanizer Skill make text pass GPTZero detection?

Not guaranteed. Humanizer focuses on improving the linguistic naturalness of writing and eliminating statistically recognizable AI patterns. After deep rewrite mode, most articles score significantly lower on GPTZero. However, AI detection tools are continuously updated, and no tool can promise 100% passing rates. More importantly, truly "human-sounding" writing comes from adding personal perspectives, specific experiences, and unique viewpoints — that's something you need to add beyond what Humanizer improves.

Will Humanizer change the meaning of the article?

Standard mode and light touch mode don't change meaning — they only modify expression and sentence structure. Deep rewrite mode significantly restructures sentences and may occasionally shift the emphasis of detail. After processing, carefully compare with the original to confirm core points haven't been misrepresented. If a particular section is critical and must not be changed, tell AI "don't touch paragraph 3, only process the rest."

How does it handle English text?

English processing works well, especially at identifying and fixing these common English AI patterns: overuse of transition words like "furthermore" and "it's worth noting"; overly neat parallel structures; and heavy clustering of AI-favored buzzwords like "leverage," "ecosystem," "empower," and "paradigm." Standard mode covers most English use cases.

Next Steps

Lesson 19 — Install the next Skill to keep expanding your OpenClaw capabilities