Humanizer: The Open-Source Tool That Makes AI Writing Sound Human

If you've spent any time reading AI-generated text, you've felt it. The prose is polished, grammatical, and somehow empty. It leans on 'delve,' 'testament,' and 'landscape.' It reaches for three-item lists and fake-candid openings. And no matter how accurate it is, it never sounds like a person wrote it.

Humanizer is an open-source skill that fixes exactly that. It rewrites AI-sounding text so it reads like a human wrote it, while preserving the meaning, tone, and factual claims. Because it is just Markdown, it works with any AI agent that supports skills, including Claude Code and Claude Desktop.

How Humanizer works

The project is based on 35 patterns from Wikipedia's 'Signs of AI writing' page, maintained by WikiProject AI Cleanup. In the first pass, it rewrites the text without treating the original structure as sacred. Then it checks the draft against the pattern list and the original claims. Anything that still sounds artificial gets rewritten again.

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.

It doesn't make things up. A name, number, date, quote, citation, or other factual detail must come from the source or the writer. For personal writing, Humanizer keeps the writer's style. Technical and reference prose stays neutral and plain. If you provide a writing sample, Humanizer follows that sample instead of its default style rules.

35 patterns of AI writing

The 35 patterns fall into five buckets: content, language and grammar, style, chatbot phrasing, and filler words. Here are a few highlights:

There are more patterns in the repo, but this gives you a feel for how specific and practical they are.

From AI to human: the Lisbon example

The repository includes a first-person travel story about Lisbon. The AI version is full of 'breathtaking panoramic views,' 'truly special,' and an emoji. The Humanizer version keeps the same trip but sounds like a real traveler with mixed feelings. Instead of 'this city completely stole my heart,' it says 'I spent five days in Lisbon last October and still have mixed feelings about it.' That's the core move: state the point directly, keep the facts, and cut the sales pitch.

The rewrite even makes a joke about calves and chooses a specific pastry shop in Graça instead of the famous tourist one. It's more specific, more honest, and far more readable.

Match your own voice

If you want the result to sound like you, include a writing sample. Humanizer follows the sample's rhythm, word choice, punctuation, and deliberate quirks. The skill can be invoked like this:

You can also write plain language requests like 'Please humanize this text: [your text]' or point it at a file: 'Humanize the prose in docs/launch-post.md.' It only alters prose, leaving code, data, frontmatter, and link targets alone.

Installation

Installing Humanizer is simple. With the Skills CLI:

Leave off --global for current project only. Add --agent to choose specific agents. In Claude Code 2.1.142 or newer, you can install the plugin instead:

The plugin command is /humanizer:humanizer. In Claude Desktop, download the repository as a ZIP and upload it as a skill. For a manual install, just copy SKILL.md into the agent's skill folder. The project is MIT licensed.

Why this matters for AI creators

At our AI image and video generation platform, we're all for AI-assisted production. But we've seen too many AI draft posts that feel like they were written by a committee. Humanizer is a great final layer for scripts, captions, blog posts, and video descriptions. It doesn't just swap synonyms; it checks a real list of AI tells and removes them while preserving what the text actually says. The result is writing that doesn't set off your own AI radar.

If you generate content with AI, you know the tell. Humanizer is a free, open-source way to take the edge off it. Check out the project on GitHub, read the 35 patterns, and try the Lisbon example if you want to see the difference. Your readers won't thank you for it. They just won't know why your AI copy finally sounds human.