Content Humanizer Claude Skill
The Content Humanizer Claude skill finds AI patterns in a draft and rewrites it to sound like a real person, for content marketers and editors. It is free and open source under the MIT license, from Alireza Rezvani's claude-skills repo. It scores the text with a bundled Python script before and after the rewrite.
/plugin marketplace add alirezarezvani/claude-skills/plugin install marketing-skills@claude-code-skillsnpx skills add alirezarezvani/claude-skills --skill content-humanizerWhat the Content Humanizer skill does
Removing a few banned words does not fix a flat draft. This skill works in three modes: detect, humanize and voice injection. You can run all three together or stop after the audit if you want to see the problems first.
Detection starts with a mechanical pass using a scorer script that returns a human-ness score from 0 to 100. The skill reads low scores as a sign the draft needs a rewrite rather than a patch. It then names each pattern, such as filler vocabulary, hedging, uniform rhythm and vague claims, and explains the fix.
Humanizing varies sentence rhythm, swaps vague claims for specifics and breaks repeated paragraph structure. Voice injection then applies your brand voice, using a context file or one example of writing you like.
- Runs a Python scorer on a draft and returns a 0 to 100 human-ness score
- Flags AI patterns by severity, including filler words, hedging chains and uniform paragraphs
- Replaces filler vocabulary with plainer words instead of only deleting it
- Varies sentence length and paragraph shape to break a monotone rhythm
- Pushes vague claims toward named sources and numbers, or asks you to supply them
- Applies brand voice from a context file or one writing example
- Warns when a draft has too many tells for a patch and needs a full rewrite
- Rescores after editing, so you can check that the score moved
How it works
Score the draft
Claude runs the bundled scorer script on your draft and reports a 0 to 100 score with a breakdown by signal.
Name the patterns
Each AI pattern is listed with severity, why it reads as machine-written and a specific fix, instead of a vague note.
Fix rhythm and specifics
Sentences, paragraphs and vague claims are rewritten. You may be asked for real data where the draft has none.
Inject your voice
Using your context file or a sample, Claude applies the sentence length, tone and stance that belong to your brand.
Rescore and compare
The scorer runs again, and you see before and after paragraphs with the reasons behind each change.
Example prompts
Once the Content Humanizer skill is installed, ask Claude in plain language. These prompts trigger it:
- Score this blog draft for AI patterns and tell me if it needs a rewrite
- Humanize this article and keep our direct, slightly irreverent voice
- Audit this landing page copy for AI tells, but do not edit it yet
- Rewrite this draft and show me before and after for the three biggest changes
- Here is a post I love. Match its voice when you rewrite my draft
What you get
- A 0 to 100 human-ness score before and after
- An annotated list of AI patterns with severity and fixes
- A rewritten draft with varied rhythm and more specific claims
- A before and after comparison of key paragraphs
Who it is for
- Content marketers editing AI-assisted drafts before publishing
- Agencies keeping client voice consistent across writers and tools
- Editors who want a measurable check, not only a gut feel
How to install the Content Humanizer skill
What you need: Claude Code or another agent with a terminal and Python 3, to run the scorer script. Without a terminal, the editing method still works but the scoring step does not.
- Claude Code plugin
/plugin marketplace add alirezarezvani/claude-skills - Claude Code plugin
/plugin install marketing-skills@claude-code-skills - Skills CLI
npx skills add alirezarezvani/claude-skills --skill content-humanizer - Manual
git clone https://github.com/alirezarezvani/claude-skills.git && cp -r claude-skills/marketing-skill/skills/content-humanizer ~/.claude/skills/
After installing, restart Claude Code or start a new session. Claude loads the Content Humanizer skill automatically when your request matches it, or you can ask for it by name.
Tips for better results
- Give it one piece of writing you admire. It extracts voice patterns from a sample better than from adjectives.
- Keep your own opinions and real examples in the draft. It cannot invent specifics for you.
- Run the audit alone first if the client wants to see issues before any edits.
Limitations
- The scorer is a heuristic. A good score does not prove the writing is good or that detectors will pass it.
- The tell list is a snapshot, and newer models can show different patterns that the skill says should be refreshed.
Content Humanizer Claude skill FAQ
Yes. It is open source under the MIT license, from Alireza Rezvani's claude-skills repo. The scorer is a bundled script with no paid service involved.
Run /plugin marketplace add alirezarezvani/claude-skills, then /plugin install marketing-skills@claude-code-skills. Or use the Skills CLI line shown here, or copy the skill folder into ~/.claude/skills/.
The editing method can run in Claude.ai if you zip the skill and upload it under Settings, Capabilities, Skills on paid plans. The scorer script needs a terminal, so full scoring needs Claude Code or an agent with Python.
Give Claude your draft and a sample of your own voice. This skill scores AI patterns, removes filler, varies rhythm, adds specifics and rescores. It works best when your real opinions and examples are in the draft.
No. It aims to improve writing quality and voice, and its score is its own heuristic. It makes no claim about passing third-party AI detectors.