humanizer git:20260907.a4d2452A

Immutable. This exact content is served forever at /api/v1/blob/a5e8f8982ef2ab68.

---
name: humanizer
description: Removes AI-pattern tells from outreach copy and edits it to read like a person with context — not like a language model. B2B-safe: never violates style; never promises to beat detectors.
---

# Humanizer

Rewrites and/or audits your draft for the specific sloppy giveaways that make B2B copy read as "AI wrote this": stacked AI-tell vocabulary, formulaic rhythms, performed sincerity, fluffed-up hedges, fake personalization.

## Inputs
- Draft (required)
- Optional: persona/role to tune for
- Mode: full rewrite, or annotate-only edit

## Removals/edits
1. **AI vocabulary blackout:** leverage, harness, streamline, unlock, delve, foster, landscape, navigate, robust, holistic, empower, dynamic, passion, elevate, showcase, spearhead
2. **Sincere-performers cut:** "I hope this finds you well," "I would love to," "Thank you for your time in advance," "Truly excited to share"
3. **Pattern-stack breaks:** no triads ("grow, scale, win"), no false-binary ("It's not X. It's Y"), no reveal-bridge openings ("Here's the thing:")
4. **Em-dash density:** cap, don't ban (~1 per 100 words)
5. **Fake-personalization check:** if the "personalized" clause could be pasted to 500 other people, mark it for a model or a real customer name/number/link only

## Approval
Mode A **rewrite**: show the edited draft; user approves before it's returned. Mode B **annotate**: spans + reason per span; user walks through. Never both applied silently.