Removes the traces of AI generation from text. Built for editing or reviewing copy so it sounds more natural and more human. Based on Wikipedia's wide-ranging guide to the “signs of AI writing.” It detects and fixes these patterns: inflated symbolism, promotional language, superficial analysis tacked on with -ing endings, vague attribution, overuse of em dashes, the rule of three, AI vocabulary, negative parallelism, and too many connective phrases.
For writers in Chinese: bring AI-sounding text back to a human voice
Paste in a stretch of AI-generated Chinese and get back a natural rewrite, with a point-by-point changelog and a writing-quality score out of 50.
What this tool for stripping out AI writing traces can do for you
humanizer-zh is a Chinese-language text editing tool built specifically to rewrite AI-generated Chinese into a version that reads more like a person wrote it. It suits writers, editors, and anyone who needs to publish Chinese content.
Paragraphs produced by AI tend to share a fixed set of tics: inflating small things into grand significance, piling up promotional adjectives, attributing opinions to “experts say,” leaning on em dashes and three-part lists, sentences that are tidy but have no human voice in them. Picking all of that out one word at a time takes a great deal of work.
The tool works against a checklist compiled from Wikipedia's “signs of AI writing,” scanning sentence by sentence for 24 common patterns across four categories — content, language and grammar, style, and communication markers — and then rewriting the problematic passages while keeping the original's core information and tone. It gives you usable rewritten text outright, followed by a changelog and scores in five areas: directness, rhythm, trust, authenticity, and concision.
It is aimed mainly at Chinese writing and has been tuned to Chinese conventions, so patterns that do not apply in Chinese, such as title case, are skipped. If you are hoping it will help you slip past AI detectors, that is not what it does.
How it differs from the usual AI rewriting
A general-purpose chat tool can rewrite anything and discuss writing itself along the way; this tool does one thing — it clears AI traces sentence by sentence against a single standard, and tells you what it cleared.
It scans a fixed 24-item checklist rather than improvising each time. Wherever inflated significance, vague attribution, or negative parallelism appears, it is called out and rewritten, and you can check it against the changelog instead of going on feel.
Most “de-AI” approaches stop at making the text clean; this one also puts effort into voice — long and short sentences interleaved, complicated feelings acknowledged where they belong, the first person allowed — so the rewrite reads as if a person is speaking. It reworks the expression, without touching the information or tone of the original.
Pick it if you have a piece of Chinese to send out and are unsure where it reads like a machine wrote it; if what you need is drafting from scratch or a deeper discussion of the content itself, a general chat tool is the better fit.
Common questions about removing AI writing traces
What does this skill do?
It rewrites AI-generated Chinese into a version that reads more like a person wrote it. The emphasis is not on swapping synonyms but on finding the traces of machine writing sentence by sentence and rewriting them. Afterward you get text you can use as is, plus a changelog and a quality score.
What kinds of text can it work on?
Marketing copy, academic abstracts, blog posts, and body text inside documents all work. As long as it is a piece of Chinese text, it follows the same process: identify AI patterns, rewrite the problematic passages, keep the core information, and hold the original tone.
Can I paste text straight in, or do I have to upload a file?
Both work. You can paste the text in directly, or have it work on content in a file, such as body text saved in article.md. The more complete the context you give it, the better the rewrite will match the original purpose and tone.
Will it alter the facts or figures in my article?
It will not change them of its own accord. It preserves the core meaning of the original and rewrites only the way things are expressed, and it will not add new facts or numbers on your behalf. If the original attributes a claim to a vague source such as “experts say,” it will point out the problem, but you still have to supply the actual citation yourself.
How can I tell whether the rewrite turned out well?
It scores the rewritten text across five dimensions — directness, rhythm, trust, authenticity, and concision — for a total of 50. A score of 45 to 50 means AI traces are largely gone; below 35 and another pass is recommended.
Will it tell me which parts it changed?
Yes. Besides the rewritten text, it provides a brief changelog that lists item by item what was cut, such as inflated symbolism like “proof that…,” a redundant “furthermore,” a three-part parallel construction, or an em-dash analysis sentence at the end.
How is it different from simply asking ChatGPT to rewrite something?
ChatGPT applies a different standard on every rewrite; this tool checks sentence by sentence against a fixed set of 24 AI writing patterns and gives you a changelog and a score. For the same passage it applies one consistent set of rules, which makes it easier to verify and reproduce.
Is it meant to fool AI detectors?
No. Its goal is to raise the quality of the writing, not to evade detection. By its own account, the best way to de-AI a piece of text is to put a real person behind it — opinions, rhythm, specific details — rather than disguising it as human-written.
Does it mainly handle Chinese or English?
Mostly Chinese. Its rules are drawn from English-language AI writing guides and adapted to Chinese writing habits, so some patterns that do not apply in Chinese are skipped. Purely English text is not its main target.