Chapter Humanizer — Word Add-in

A Microsoft Word task-pane add-in that rewrites AI-generated paragraphs to sound natural and human. It sits inside Word as a sidebar, lets you rewrite one paragraph at a time, scores the text before and after, and shows you exactly what patterns were removed — all without leaving your document.

Uses the Claude API (Anthropic) as the rewrite engine.

AI-generated text has fingerprints. Certain words appear far too often — leverage, pivotal, seamless, robust, furthermore. Sentences get stitched together with em dashes. Importance gets inflated with phrases like "represents a paradigm shift" or "serves as a testament to." Lists of three appear constantly. Simple verbs get replaced with roundabout constructions ("serves as" instead of "is").

Individually these patterns are minor. Together they make text feel synthetic — readable, but somehow hollow. Editors, publishers, and increasingly AI detectors notice them.

Chapter Humanizer removes them systematically, one paragraph at a time, while keeping your meaning, length, and technical accuracy intact.

Two rewrite modes:

• Humanize Paragraph at Cursor — place your cursor anywhere in a paragraph and click. No selection needed.
• Humanize Selection — highlight any span of text and click to rewrite just that portion.

Before/after scoring: Every rewrite shows a live humanizer score (7–94) for both the original and the rewrite. The score is calculated client-side by checking for banned vocabulary, em dashes, and known AI phrase patterns. A score in the green zone (71+) means the text reads human. Red (below 40) means it's heavily patterned.

Flag explanations: Below the score, you see exactly why the original scored low — which banned words were found, whether em dashes were present, which specific AI phrases were detected. After the rewrite, you see a green checklist of everything that was removed.

Accept or dismiss: Every rewrite is a suggestion. You review it in the panel, then click Accept to replace the text in your document or Dismiss to leave it unchanged.

Sample text for testing: A built-in "Insert sample AI text" button drops a heavily AI-patterned paragraph into your document so you can immediately see the scoring and rewrite flow in action.

• Node.js v18 or later
• Microsoft Word (desktop, not web) — macOS or Windows
• An Anthropic API key (sk-ant-...)

git clone <your-repo-url>
cd word-humanizer
npm install

Word requires HTTPS even for local add-ins. Run this once:

npm run install-certs

On macOS you may be prompted for your system password to trust the certificate. On Windows, accept the UAC prompt.

npm start

You should see:

✓ Chapter Humanizer running at https://localhost:3000

1. Open Word.
2. Go to Insert → Add-ins → My Add-ins → Upload My Add-in.
3. Browse to manifest.xml in this folder and click Upload.

1. Open Word.
2. Go to Insert → Add-ins → My Add-ins → Upload My Add-in.
3. Browse to manifest.xml and click Upload.

The Chapter Humanizer panel will appear in the right sidebar.

1. Enter your Anthropic API key in the panel and click Save.
2. Place your cursor in any paragraph and click Humanize Paragraph at Cursor, or select text and click Humanize Selection.
3. Review the rewrite in the panel, then click Accept to replace the text or Dismiss to skip.

Your API key is stored only in the browser's localStorage for the Word add-in — it is never logged or persisted on disk.

word-humanizer/
├── manifest.xml        # Office Add-in manifest (points to localhost:3000)
├── package.json
├── server.js           # Express HTTPS server + /api/humanize proxy endpoint
└── src/
    ├── taskpane.html   # Add-in UI
    ├── taskpane.js     # Office.js logic
    └── taskpane.css    # Styles

npm run stop
# or just Ctrl+C in the terminal running npm start

• Multi-LLM backend — swap Claude for OpenAI, Gemini, or a local Ollama model via a dropdown in the panel. Useful for air-gapped environments or cost control.
• Ollama support — run fully offline with models like Llama 3, Mistral, or Phi-3. No API key required.
• Chrome extension — bring the same rewrite flow to Google Docs, Notion, and any contenteditable field in the browser.
• Tested environments — currently verified on macOS + Word 16.x. Need community testing on Windows 10/11 + Word, and Word Online.
• Batch mode — rewrite an entire document section-by-section in one click, with a progress bar and bulk accept/dismiss.
• Custom word list — let users add their own banned words and phrases on top of the defaults.
• Score history — track average before/after scores across a session so you can see overall document improvement.

The humanization rules, pattern taxonomy, and core philosophy in this project are built on the work of Siqi Chen (@blader) and his original humanizer Claude Code skill (9.8k ★, MIT licensed).

"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." — Siqi Chen

This project wraps those ideas into a Microsoft Word add-in with a live before/after scoring UI.