Last week at 2 AM, I was helping my Xiaohongshu friend Xiaoya check her posts
Last week at 2 AM, I was helping my Xiaohongshu (Little Red Book) friend Xiaoya look at her notes — she clearly typed them out herself late at night, but the platform flagged them as "Suspected AI" and almost throttled her reach. She was ready to file an appeal. I said: hold on, let's run our own test first.
What this is, and who's already using it
Last month Substack added an AI detection button right in the backend (not a new concept, just made public). Zhihu, WeChat public accounts, and Xiaohongshu are quietly rolling it out too. The principle isn't complicated: AI-generated text tends to be too "neat" — uniform sentence length, few transitions, a strong formula feel — and detectors latch onto those patterns.
I made this mistake myself: I used ChatGPT to polish a client draft, and when the client ran it through a detector, it came back 92% AI. The client didn't scold me, but I turned red for half a day — my own original draft had been sanded into "machine flavor" by AI.
This is what people in the detection circles call the "cat-and-mouse game": detectors catch "over-polishing," and most grammar-polishing tools (including ChatGPT's polish mode) happen to manufacture exactly that signature.
Now my friends doing personal IP all use them — Ajie, who runs paid communities, and Xiaomin, who runs a WeChat public account matrix — both run every post through GPTZero or ZeroGPT before publishing, and keep the "AI rate" under 30%.
What it costs to replicate today
Money: Free. GPTZero (gptzero.me) and ZeroGPT (zerogpt.com) both offer daily free quotas — single articles under 1000 words fit fine.
Time: 3-5 minutes. Paste, click, wait for the result.
Technical barrier: Zero. As easy as using Baidu Translate.
First step: Open gptzero.me → click "Try Free" → paste your article in → check the percentage.
Quick tip: if the AI rate comes back high, reverse-engineer it — drop in a personal mini-rant, delete an overly smooth summary, throw in a few casual words. Detectors don't recognize "did you use AI," they recognize "is there a human in there."
Advice for friends at different stages
Just starting out: Don't panic right now. Focus on writing your real feelings — detectors won't catch you. Don't burn time studying tools yet.
Got 1-2 stable clients or readers: I'd suggest self-testing one "high-risk" piece per month (deep long-form, paid posts). One false flag is cheaper than losing a client.
Scaling up (matrix accounts / team): Write "pre-publish self-test" into your SOP (Standard Operating Procedure). One team I know already requires all final drafts to come in under 30% AI rate, or they get sent back for rewrite. It eats time short-term, saves accounts long-term.
One last thing: this tool isn't for everyone. If you just write a journal on a small account and log your life — detectors have nothing to do with you. But if you make a living from content — it's worth 3 minutes of your time.