Last week, someone copied my post. Then I realized there’s no easy fix.
A follower on Xiaohongshu messaged me: “My post from last week was screenshotted and republished by someone in my industry. They copied the comment section word for word and refused to admit it. What can I do?” After reading the recent discussions in the technical community, my conclusion was blunt: there really isn’t a good answer right now.
What an “AI text watermark” is—and why it can’t stop copying
The idea is to secretly embed a “fingerprint” in an AI-generated passage through its word choices and sentence rhythm. Later, if someone suspects a passage was written by AI, a tool could detect that fingerprint.
Tech blogger Sean Goedecke recently explained the problem directly: text watermarks are inherently rewriteable. He used an engineer named Sam as an example. Sam wrote a small script that replaced periods with exclamation marks and changed “very” to “extremely.” After one pass, the watermark detector labeled the text as human-written.
The reason is that the watermark lives inside statistical patterns. Change a few words and those patterns fall apart. Image watermarks can sometimes be removed at the pixel level; with text, changing the wording is enough to erase it. OpenAI, Anthropic, and other major labs have acknowledged that this approach isn’t reliable and have largely abandoned hard-coded watermarking at the text layer.
What I can do today—it barely costs anything
Cost: ¥0. Time: 5 minutes. Technical barrier: none, as long as I’m willing to accept reality and stop expecting a tool to handle this automatically. First step: beneath my most recent post, add a personal follow-up comment or screenshot as evidence that the post started with me.
I used to imagine that if I wrote copy with AI, at least there would be some “anti-copy code” proving I published it first. I got that part wrong. The technical community had already disproven the idea, so there’s no point in wasting more time on it.
Advice by stage
If I were just starting out and had only published a few pieces: I’d skip this entirely. Getting content out and building a publishing rhythm matter ten times more than worrying about copying. If someone copies me, that’s a sign I got something right.
If I already had one or two customers or regular readers: I’d include something only I could get my hands on in every post—a screenshot of a customer’s exact words, internal data, or a joke from my city. AI can imitate the surface, while platform rules can’t erase the source.
If I were already scaling across multiple platforms or had a team: I’d do two things: 1) create an original-publication timestamp by immediately sharing the post’s link on WeChat Moments or Weibo, where the platform records the time; 2) build a local asset library for ideas and raw data, so I could produce the original documents if a dispute ever reached court.
Watermarks can’t save us. Showing up on camera, adding a clear byline, and preserving the original source material is what works. Not everyone needs to test this tool, and it’s fine to leave it alone for now—but knowing it doesn’t work can save a lot of detours.