Humanizer Pro

What this skill does

• Diagnoses briefly why a text reads generic, smooth or AI-typical (opener, buzzwords, empty benefit-chains, structure, rhythm, CTA)
• Rewrites the text more concrete, natural and channel-fit — and explains the key changes
• Protects names, numbers, dates, prices and core claims via Meaning Lock, and invents no facts or experiences
• Concretizes abstract claims only from supported info — if a detail is missing, it inserts a marked placeholder to fill in
• Offers 10 modes (Natural Clean, Casual Human, Anti-Fluff, Brand Voice, Social Native, Sales Human, Expert Human, My Voice, Executive Clear, Warm Support)
• Adapts text to 12 channels (email, website, landing page, LinkedIn, YouTube, Discord, cover letter, support, ads, product copy, newsletter, article)
• Builds a voice profile from 2–5 sample texts and writes in your own or your brand voice
• Delivers, on request, several style variants, a before/after Humanizer Score, and before/after comparisons
• Calibrates both ways — good text stays largely intact, no manufactured problems
• Builds a portable Voice Library: emits the voice as a saveable VOICE PROFILE block you reuse in any session/tool (cross-session consistency)
• Revises itself: scores the draft against a rubric, runs a faithfulness pass and fixes it once before output
• Avoids homogenization — keeps real idiosyncrasy instead of a uniform "AI-good" sound; knows when NOT to humanize (legal/medical/technical/compliance)
• Warns on risky communication (pushy, spammy, misleading) and offers a credible alternative

Description

Humanizer Pro is an AI skill for natural, credible, channel-fit text editing that turns generic or AI-typical drafts into real communication. It solves one of the most common AI problems: the first draft arrives fast but reads too smooth, too generic, too salesy — not like a real person, a real brand or a real situation. Positioning matters: Humanizer Pro is **not an "undetectable AI" tool.** It won't help evade AI detectors — that's unreliable (detectors have high error rates and are biased against non-native writers), and "humanizers" that inject statistical tricks or artificial errors are now flagged on purpose. The skill makes text **genuinely better**, not sneakier: a text doesn't read human because it contains deliberate mistakes, but because it has a clear intent, real context, concrete details, natural rhythm and a fitting voice. The flow is structured: **quick diagnosis** (why does it read generic?) → **intent & Meaning Lock** (fix the intent, protect facts/names/numbers) → **pick a mode** → **rewrite** → **quality pass** → **output** with diagnosis, improved version, a change list and variants. Five engines drive it: an **AI Pattern Scanner** (spots AI openers, empty benefit-chains, buzzword stacks, false balance, over-structuring, sycophancy, pseudo-empathy, CTA clichés — consistent with AI Signal Check), **Meaning Lock** (protects facts and adds nothing — if a concrete detail is missing it inserts a marked placeholder instead of inventing), a **Voice & Tone Engine** (a voice fingerprint from 2–5 sample texts for your own or your brand voice, without deceptively imitating a real person), a **Channel Adapter** (12 channels from email through LinkedIn to support, each with its own opener, length, tone and CTA) and a **Rewrite & Variant Engine** (one strong version plus, on request, several style variants and a before/after Humanizer Score). Three things make it reliably better than "just ask the model": a **portable Voice Library** (the voice is emitted as a saveable block you can reuse in any session — cross-session consistency), a **self-revision loop** (the skill scores its own draft against a rubric, runs a faithfulness pass and revises once before output) and **anti-homogenization** that keeps real idiosyncrasy instead of making every text sound the same "AI-good". Ten modes cover every context — Natural Clean, Casual Human, Anti-Fluff, Brand Voice, Social Native, Sales Human, Expert Human, My Voice, Executive Clear and Warm Support — because "human" means clearer, more direct, warmer, more expert or shorter depending on the situation, not uniformly "casual". Two principles set it apart: it **calibrates both ways** — an already-good, concrete text gets only minimal edits instead of a forced rewrite, and your own voice isn't smoothed flat; and it has an **abuse guard** — no academic cheating, no plagiarism laundering, no deceptive impersonation, no more-convincing spam/scam, no dark patterns or unsupported superlatives. Humanizer Pro is the repair half to the diagnostic AI Signal Check: first understand why a text reads AI-typical — then honestly write it better.

Examples

**What I changed:** cut the filler opener · replaced "innovative/optimizes" with the concrete action · one sentence instead of a claim-stack. **Meaning Lock:** invented no new facts — *which* tasks exactly, you know better than I do; tell me and it gets sharper.

**Next options:** A) shorter still · B) more email-style · C) LinkedIn version · D) 3 variants

(just added one "!" for warmth). I won't manufacture problems to look busy.

**Applied to your blurb:**
```text id="humanizer-pro-rewrite"
New: a dashboard that shows which change moved the number this week. No setup, no export to a spreadsheet — open it, see what worked, do more of that.

**Email to existing users:**
```text id="humanizer-pro-email"
Subject: The dashboard now tells you what worked

Hey — new this week: the dashboard shows which change moved your key metric that week. No setup.

Open it after your next release — you'll see what actually moved the number instead of guessing. (You'll find it under [where in the app? fill in].)

If it's not tracking your main metric, reply with which one and I'll sort it.

What it does not do

• Won't help evade AI detectors, make text "undetectable" or launder plagiarism
• Never injects deliberate errors and won't blanket-"casualize" text — "human" depends on channel and audience
• Invents no facts, numbers or personal experiences and strengthens no unsupported claims
• Won't deceptively imitate a specific real person — only your own or an authorized brand voice
• Won't make spam, scam or misleading text more convincing, and builds no dark patterns / false urgency
• Won't compulsively rewrite good text or smooth your quirks into generic "good writing"
• Doesn't replace professional, legal or medical review for such content

Compatibility & tech

Tested (internal)

10 scenarios

Modes

Natural Clean (professional but not stiff) · Casual Human (more human, less corporate) · Anti-Fluff (radical against filler) · Brand Voice (brand / own voice) · Social Native (platform-fit, not AI-LinkedIn-speak) · Sales Human (selling but not fake) · Expert Human (expert but readable) · My Voice (more like you) · Executive Clear (crisp, assured, B2B/management) · Warm Support (empathetic but not fawning)

Inputs

Text (an AI draft or your own rough text, any language) · Sample texts for your voice / brand voice (optional, 2–5) · Audience, channel, desired tone, length, taboo/preferred words, CTA (optional) · Goal/effect of the text (optional — otherwise it infers the intent)

Output format

Quick diagnosis + improved version (clean copyblock) + change list + variant picker; on request before/after, a Humanizer Score table, a voice profile, a channel pack (several channel versions), and marked placeholders for missing facts

Subcategory

AI text refinement & brand voice

License

Proprietär

Security profile

What you get

• humanizer-pro-1.1.0/12 files
  • .claude-plugin/marketplace.json
  • plugins/humanizer-pro/skills/humanizer-pro/12 files

    • SKILL.md
    • manifest.json
    • references/10 files

      • humanizer-pro-channels.md
      • humanizer-pro-checklist.md
      • humanizer-pro-examples.md
      • humanizer-pro-meaning-lock.md
      • humanizer-pro-modes.md
      • humanizer-pro-output.md
      • humanizer-pro-patterns.md
      • humanizer-pro-safety.md
      • humanizer-pro-voice-library.md
      • humanizer-pro-voice.md
  • .agents/skills/humanizer-pro/→ universal — same content (Codex, Cursor, Copilot, Gemini, Windsurf, Cline)
  • LICENSE.txt

Installation

Also works as a chat prompt

Note

Not an "undetectable AI" tool: Humanizer Pro won't help evade detectors, enable academic cheating, launder plagiarism or deceptively impersonate. It improves text honestly (clearer, more concrete, more credible) and protects facts, but doesn't replace professional, legal or medical review. Strongest with audience, channel and desired tone as context; otherwise it makes sensible assumptions.

Changelog

• v1.1.001.07.2026Stronger and still free: (1) a portable Voice Library — capture a voice once, emit it as a saveable VOICE PROFILE block, paste it back in any later session/tool (cross-session brand/own voice, which plain prompting can't do); (2) a self-revision loop — the skill scores its own draft against a rubric, runs a faithfulness pass and revises once before output; (3) anti-homogenization — keeps real idiosyncrasy so texts don't all sound the same 'AI-good'; (4) 'when NOT to humanize' (legal/medical/technical/compliance stay precise); (5) a worked before→after example library + concrete score-rubric anchors. Now 10 references.
• v1.0.001.07.2026First release: a voice, clarity and text-quality engine that makes generic or AI-typical text more natural, concrete, credible and channel-fit — not a detector-evasion tool. Flow: diagnosis → intent/meaning lock → mode → rewrite → quality pass → output. 10 humanizer modes (Natural Clean, Casual Human, Anti-Fluff, Brand Voice, Social Native, Sales Human, Expert Human, My Voice, Executive Clear, Warm Support), 12 channel modes (email/website/landing/LinkedIn/YouTube/Discord/cover letter/support/ads/product/newsletter/article), an AI pattern library (12 patterns, consistent with AI Signal Check), Meaning Lock + faithfulness (no invented facts/experiences, concretization only from the source or as a marked placeholder), a voice fingerprint from sample texts, a Humanizer Score, a variant engine, output formats. 8 references, realistic examples incl. a calibration case, internal evals. Two research-backed sharpenings: (1) anti-evasion with current evidence (detector unreliability, Turnitin bypass detection 2025) instead of a bare 'no'; (2) abuse/manipulation guard (no spam/scam/dark patterns, ad-safe). Human ≠ casual ≠ deliberate errors.

Frequently asked questions