intermediate content Editor's Pick

Anti-AI Content Rewrite

Run your marketing copy through an AI-pattern detector that flags and rewrites every phrase that sounds generated — banned words, structural tells, fake enthusiasm, and hedging.

Works with: claude-code

AI-generated marketing copy has a smell. “Delve into the ever-evolving landscape” — you know it when you read it. Your audience does too.

This skill scans your content files for the patterns that make text feel generated: overused words (delve, leverage, seamless), structural tells (triple adjective lists, identical paragraph lengths), fake drama (“not just X, it’s also Y”), and hedging that avoids saying anything real.

Then it rewrites everything in a direct, human voice.

What Makes This a Claude Code Skill

It uses git status to find only new or modified content files, reads them, flags every AI pattern, and rewrites the text in place. You review a before/after diff before anything changes. No manual grep for “delve” across 50 files.

Built from research across Wikipedia’s “Signs of AI Writing” guide, GPTZero’s vocabulary analysis (3.3M texts), and community-sourced pattern lists.

What It Catches

Banned vocabulary — 80+ words and phrases that appear 10x-200x more often in AI text than human writing. “Delve” usage jumped 654% in academic papers after ChatGPT launched. “Meticulously” jumped 3,900%.

Structural patterns — Identical paragraph lengths, predictable intro-body-conclusion format, excessive bullet points, bolded “Term: Definition” lists, summary sentences restating what you just said.

Tone problems — Hedging (“it could be argued”), fake enthusiasm (“exciting!”), both-sides-ism that avoids opinions, overly formal register where casual fits better, the “whether you’re a beginner or expert” catch-all.

Sentence-level tells — Triple adjective stacking (“fast, efficient, and user-friendly”), trailing “-ing” filler (“ensuring,” “highlighting”), em dash overuse, uniform sentence length (low burstiness).

Setup

Create .claude/commands/anti-ai-rewrite.md in your project:

Rewrite unpublished content to strip the "AI-generated" feel and match my writing voice.

## Process

1. Run `git status` to find new or modified content files
2. Read each file
3. Rewrite all human-facing text fields
4. Keep factual data intact (names, URLs, pricing, dates)
5. Show me a before/after diff for approval before writing

## Voice Rules

- Use "you"/"your" — talk to the reader
- Use contractions
- Mix short and long sentences
- State opinions directly — no hedging
- Lead with what matters, not a definition
- Be specific — numbers over vague claims

## Banned Words

delve, leverage, harness, foster, embark, elevate, embrace, underscore,
unlock, unleash, empower, streamline, optimize, revolutionize, transform,
utilize, reimagine, orchestrate, meticulous, comprehensive, intricate,
pivotal, crucial, robust, seamless, vibrant, dynamic, multifaceted,
nuanced, profound, innovative, groundbreaking, cutting-edge, unprecedented,
compelling, invaluable, paramount, holistic, tapestry, landscape, realm,
beacon, journey, testament, paradigm, synergy, nexus, cornerstone, catalyst

## Banned Patterns

- "In today's..." / "In the ever-evolving..."
- "It's worth noting..." / "Whether you're X or Y..."
- "Game-changer" / "stands as a testament"
- Triple adjective lists
- Summary sentences restating what was just said
- Every paragraph the same length

Customize the voice section to match your brand. The banned words and patterns work for any marketing content.

Usage

/project:anti-ai-rewrite

Claude finds your uncommitted content, flags every AI pattern, and rewrites it. You approve each change before it’s saved.

Use this skill as step 2 in a two-step process:

  1. Research and draft with the Perplexity MCP Server — its perplexity_research tool pulls real-time data, stats, and sources. You get a factually grounded first draft with citations.
  2. Polish with this skill — run /project:anti-ai-rewrite to strip the AI patterns, inject your voice, and make it sound like a person wrote it.

Research first, rewrite second. Facts stay accurate, voice becomes yours.

Tips

  • Run this after the content is factually complete — it rewrites tone, not facts
  • Pair with the Brand Voice Linter for a full pre-publish check
  • The banned word list is based on GPTZero’s frequency analysis — these words aren’t bad on their own, but they cluster in AI output
  • Add your own banned words as you spot patterns in your content