Humanizer
An open-source tool that rewrites AI-generated text to read naturally, based on 35 known AI-writing patterns, without touching code, data, or citations.
🔗 Visit Humanizer
Description
AI writing tends to have tells — certain phrasing patterns, an overly balanced "on one hand, on the other hand" structure, specific words used too often — that make text feel obviously machine-written even when the content itself is fine. Humanizer is a tool that specifically targets those tells, rewriting AI-generated text so it reads the way a person would actually write, without changing the facts, code, or data inside it.
Humanizer is an open-source "agent skill" — designed to run inside Claude and similar AI coding/writing tools — that detects and removes 35 specific patterns associated with AI-generated writing, based on Wikipedia's own guidelines for identifying AI text. It works in two passes: a draft, then a self-critique before producing the final rewrite, and it can match a target voice if given writing samples. Critically, it's built to preserve code blocks, data, frontmatter, and citations untouched — it only touches prose. It's installable via the Skills CLI or the Claude Desktop plugin marketplace, is MIT-licensed, and has substantial GitHub traction (37.5k+ stars), reflecting how common the "this obviously reads as AI-written" problem has become.
💬 Our review
The short version: Humanizer is a focused, open-source tool for one specific job — making AI-generated prose read like it was actually written by a person, without touching the facts, code, or citations inside it.
The two-pass approach (draft, then self-critique before finalizing) and the ability to match a given writing voice are more deliberate than most "humanizer" tools, which tend to just swap synonyms. Preserving code/data/citations untouched is also the right call for anything technical.
Where to be careful: like any AI-detection-adjacent tool, it's fighting a moving target — AI detection heuristics change, and no rewriting tool guarantees text won't be flagged by a specific detector. It's also primarily an English-language tool. If you're using AI to draft technical or long-form content and the "obviously AI" tone is a real problem for your audience, it's a well-built, free option. If you need guaranteed evasion of a specific plagiarism/AI detector, no tool can promise that reliably.
💰 Pricing
📊 Global score
🤖 AI-enriched data
Gratuit et open source (licence MIT), installable via Skills CLI ou marketplace de plugins Claude Desktop.
Pros
35 patterns d'écriture IA détectés et corrigés (base Wikipedia)
Réécriture en deux passes avec auto-critique
Préserve code, données et citations intacts
Correspondance de voix à partir d'échantillons fournis
Cons
Principalement conçu pour l'anglais
Aucune garantie contre un détecteur IA spécifique
Efficacité dépendante du style d'origine