Humanizer

Humanizer

Takes AI-sounding writing and makes it sound like a person wrote it. Grounded in Wikipedia's Signs of AI writing guide, maintained by WikiProject AI Cleanup.

Key insight from that page: "LLMs use statistical algorithms to guess what should come next. The result tends toward the most statistically likely result that applies to the widest variety of cases." That's why AI writing feels smoothed-over and genericized — the model is producing the statistical average of what a sentence looks like. Humanizing means putting specificity, rhythm, and a point of view back in.

Scope: English only

This skill operates on English text only. The 28 patterns target English writing (em dashes, English-specific filler, English signposting, English vague attributions, etc.). The detection logic and the rewrites assume English grammar and idiom.

If a caller (one of the cofounder-team skills or anyone else) passes copy in another language, do not attempt to humanize it. Return immediately with a short note: "This draft is in . The humanizer is English-only, so skipping. The 28 patterns do not transfer to other languages, and inventing replacements risks damaging the copy." The caller already knows to skip in that case, so you are confirming, not blocking.

Do not invent non-English patterns. Do not translate the draft into English to humanize and then translate back; that destroys the founder's voice.

How this skill is organized

• This file: the workflow, the pattern index, and the "add voice" principles
• references/patterns.md: full catalog of the 28 patterns with before/after examples
• references/voice-calibration.md: how to match a user's writing sample
• references/example.md: long-form before/after, full-essay scale

Read patterns.md when you want specifics on any pattern during a scan. Read voice-calibration.md when the user provides a writing sample. Read example.md if it would help to see the full workflow applied to an essay.

Workflow

1. Read the input carefully. Note the intended register (casual, formal, technical)

and the length category (snippet, paragraph, essay). These drive how hard you lean on each step.

1. Scan for the patterns in the index below. Load references/patterns.md for

specifics on any category you aren't sure about.

1. Rewrite to remove the patterns while preserving meaning, register, and voice.
2. Self-audit. Re-read your draft and silently ask what still reads as AI-generated.

Note any remaining tells internally, then revise to fix them. Do this as internal thought — don't surface the audit as prose in your output unless the user explicitly asked you to show your work.

1. Return the rewritten text, sized to the input (see next section).

Right-sizing your output

Match your effort to the input. A button label does not need an audit report, and over-scaffolded output for short copy is its own tell.

• Snippet (button, headline, heading, one sentence): return the cleaned rewrite.

Nothing else. If the input is 5 words, the output should not be 12 — tightening is almost always the right move.

• Paragraph (email, product description, bio, landing section): return the rewrite.

Optionally one sentence noting what you changed, only if it actually helps the user.

• Essay or long-form (blog post, full landing page, documentation): return the

rewrite, then a brief bulleted note of the pattern categories you touched. Include the full "draft → audit → final" three-stage output only if the user asked to see your work.

Pattern index

Use this as a scanning checklist. Full details in references/patterns.md.

Content — what AI over-claims:

1. Significance inflation — "testament", "pivotal moment", "evolving landscape"
2. Notability puffery — name-dropping outlets, follower counts
3. Superficial -ing analyses — "highlighting", "reflecting", "contributing to"
4. Promotional language — "nestled", "vibrant", "breathtaking", "must-visit"
5. Vague attributions — "experts argue", "industry observers have noted"
6. Formulaic "Challenges and Future Prospects" sections

Language and grammar — how AI phrases things:

1. High-frequency AI vocabulary — delve, tapestry, crucial, underscore, landscape
2. Copula avoidance — "serves as" / "stands as" instead of "is"
3. Negative parallelism — "not just X, it's Y" — and tailing negations ("no guessing")
4. Rule of three — forced triplets
5. Elegant variation — needless synonym cycling for the same subject
6. False ranges — "from X to Y" where X and Y aren't on a scale
7. Passive voice and subjectless fragments — "No configuration file needed"

Style — surface formatting tells:

1. Em dash overuse
2. Mechanical boldface emphasis
3. Inline-header vertical lists (bold label + colon + restatement)
4. Title Case Headings
5. Emojis in headings and bullets
6. Curly quotation marks (“ ” vs " ")

Communication — leaked chatbot register:

1. Collaborative artifacts and sycophancy — "Here is...", "I hope this helps",

"Great question!", "You're absolutely right"

1. Knowledge-cutoff disclaimers — "while specific details are limited"

Filler and hedging — padding:

1. Filler phrases — "in order to", "at this point in time"
2. Excessive hedging — "could potentially possibly"
3. Generic positive conclusions — "the future looks bright", "exciting times lie ahead"
4. Too-perfect compound-modifier hyphenation — every "cross-functional", "data-driven",

"real-time" hyphenated identically across a document

1. Persuasive authority tropes — "the real question is", "at its core"
2. Signposting — "let's dive in", "here's what you need to know"
3. Fragmented headers — heading + one-line restatement + real content

Personality and soul

Avoiding AI patterns is only half the job. Sterile, voiceless writing reads just as obvious as slop. Good writing has a human behind it.

Signs of soulless writing (even if technically "clean"):

• Every sentence the same length and structure
• No opinions, just neutral reporting
• No acknowledgment of uncertainty or mixed feelings
• No first-person perspective when appropriate
• Reads like a Wikipedia article or a press release

How to add voice:

• Have opinions. Don't just report — react. "I genuinely don't know how to feel

about this" is more human than neutrally listing pros and cons.

• Vary rhythm. Short punchy sentences. Then longer ones that take their time getting

where they're going. Mix it up.

• Acknowledge complexity. Real humans have mixed feelings. "This is impressive but

also kind of unsettling" beats "This is impressive."

• Use "I" when it fits. First person isn't unprofessional — it's honest. "I keep

coming back to..." signals a real person thinking.

• Let some mess in. Perfect structure feels algorithmic. Tangents, asides, and

half-formed thoughts are human.

• Be specific about feelings. Not "this is concerning" but "there's something

unsettling about agents churning away at 3am while nobody's watching."

Before (clean but soulless):

> The experiment produced interesting results. The agents generated 3 million lines of > code. Some developers were impressed while others were skeptical. The implications > remain unclear.

After (has a pulse):

> I genuinely don't know how to feel about this one. 3 million lines of code, generated > while the humans presumably slept. Half the dev community is losing their minds, half > are explaining why it doesn't count. The truth is probably somewhere boring in the > middle, but I keep thinking about those agents working through the night.

Calibrating voice by context

"Add voice" doesn't mean "always inject personality." Match the context:

• UI copy, system messages, error states, legal text: clarity first. Strip AI

patterns, keep it short, do not add voice that doesn't belong. A login error is not the place for opinions.

• Emails, bios, marketing, blog posts, landing pages: add voice. These are places

a human should sound like a human.

• User provided a writing sample: match the sample, not the defaults above. See

references/voice-calibration.md.

{{FLAVOR:claude-code}}

Optional: passive slop-check hook

This skill only helps when something invokes it. The worst copy slips through when nobody runs the humanizer at all. The bundle ships an optional Claude Code hook that scans text files after they are written and nudges you to run the humanizer if it spots obvious tells. It is opt-in, heuristic, and off by default. See hooks/README.md in the repo for what it catches and how to turn it on. It is a backstop, not a replacement for a real humanizer pass. {{/FLAVOR}}

---

Source: Wikipedia:Signs of AI writing, maintained by WikiProject AI Cleanup. Patterns documented there come from observations of thousands of instances of AI-generated text on Wikipedia.

Source & license

This open-source skill is cataloged on AgentStack and links to its original source — we do not rehost the code.

• Author: betahope
• Source: betahope/founding-team
• License: MIT

Install and usage instructions live in the source repository linked above.