If your brand has been quietly testing AI models, AI avatars or AI-written notes on Xiaohongshu, this one is for you. The platform just made that choice visible to every consumer who lands on your page.
What Xiaohongshu announced on 7 August 2026
On 7 August 2026, Xiaohongshu extended its AI labeling rules from the note to the account. Any account operated as an AI virtual persona must now state this on its profile page and switch on the “AI虚拟人” (AI virtual human) tag. It is not a suggestion. It is a required field.
Accounts that skip it and get detected face a sliding scale of penalties. At the mild end, notes sink in the feed and recommendation goes to zero. At the serious end, the account is demoted and its traffic is cut in half. The announcement was reported in detail by Sohu on the day it landed.
The numbers behind the rule are worth pausing on. Xiaohongshu’s AI governance team says it found more than 24,000 high-fidelity AI virtual human accounts over the previous six months. Complete personas. Steady posting rhythms. Emotional interaction with followers. Some of them even switched tone depending on whether the comments were warm or hostile.
As of July 2026, the platform auto-detects and labels over 300,000 AI-generated notes per day. When AI involvement in a note passes 30%, a semi-transparent floating label appears in the lower right corner reading “本文含AI生成内容”, meaning this post contains AI-generated content.
There is a carrot too. Creators who declare early keep their traffic and get first access to the platform’s new AI tools.
What the announcement does not say
Read the timing first. China’s 《人工智能生成合成内容标识办法》, the Measures for Labeling AI-Generated Synthetic Content, took effect on 1 September 2025. Xiaohongshu’s move lands almost exactly one year later. That gap is the interesting part. The law existed for twelve months while platforms did very little with it in public. What we are watching now is enforcement catching up with legislation, and it is arriving at account level rather than post level.
Account level changes everything. A label on a single note is a footnote. A label on a profile is an identity. Every visitor who checks your page before deciding whether to trust a review sees it first.
The scale of clean-up already done tells you how serious this is. In July 2026 alone, Xiaohongshu banned around 420,000 accounts it classified as AI-managed and non-compliant, a figure reported in the Chinese press alongside a wider look at how AI virtual influencer accounts must now operate. Fully automated accounts get permanent bans. Partly automated ones get restrictions. That distinction matters to any brand running a hybrid workflow.
Three things stay genuinely unclear, and I would keep an eye on all of them.
- How the 30% threshold is computed. Is it word count, image pixels, a model confidence score, or a blend? Nobody outside the platform knows. A note that is 100% human-written but illustrated with one AI-upscaled product shot may or may not trip it.
- Whether AI-assisted editing of real photography counts. Background removal, colour correction and generative fill are standard in every fashion studio. If generative fill registers as AI generation, a large share of legitimate e-commerce imagery is affected.
- Whether the tag will spread to brand accounts. Today the rule targets accounts run as a virtual persona. A brand account posting AI-generated model shots is not quite that. Not quite is a thin place to stand.
Who this affects, and the window you have
Your brand is affected if any of the following is true. You run a virtual brand ambassador on Xiaohongshu. You commission AI-generated model imagery for lookbooks or product notes. Your agency or your local team uses AI to draft note copy at volume. Or you work with seeding partners who batch-produce “grassroots” reviews.
You are not affected if your notes come from real photography, real copywriters and real creators, even if someone used AI to tidy up a first draft. Which is most likely where you should want to be anyway.
Here is the part I would sit with. On a platform whose entire value is peer trust, a visible AI label on a luxury brand’s note is not a compliance checkbox. It is a conversion problem. A Chinese consumer scrolling Xiaohongshu before spending 8,000 RMB on a bag is looking for evidence that a real person held it. A grey floating tag saying this content contains AI is the exact opposite of that evidence.
Which is why I think the cost saving was always illusory for premium positioning. AI model imagery is cheaper per asset. It was never cheaper per conversion. The rule did not create that gap. It just made it visible.
A European accessories brand we advise, mid-size, had approved an AI-model shoot for autumn three weeks before this announcement. The maths looked good on paper: one third of the cost of a Shanghai studio day. They have now moved the budget back to a real shoot with two mid-tier creators. Not because of the penalty. Because their category is trust-led and the label would have sat on their best-performing note format.
Doing nothing costs you in two directions. Reach, if you are detected and not declared. Trust, if you declare and the tag sits on content meant to feel personal.
What to do, in order
Start with an inventory. Pull every asset published on your Xiaohongshu account in the last twelve months and mark each one: fully human, AI-assisted, fully AI-generated. Your team can do this in-house in about a day. It is tedious and it is the only step you cannot skip.
Then check your account status. If you operate any persona account, declare it now and enable the tag. Early declaration protects your traffic. Late detection does not.
Third, audit your seeding partners. This is where most foreign brands have exposure they do not know about. Ask each agency or KOC network in writing whether AI drafting is used at any point in note production, and get the answer on paper. If you buy volume seeding at a very low unit price, assume the answer is yes until proven otherwise.
Fourth, rewrite your content brief. Add a clause on AI use, one on the 30% threshold, and require named human authorship for review-format notes. Keep the raw files from every shoot, since being able to prove a real photograph exists is quickly becoming an asset.
Fifth, rebuild the imagery you have to replace. This is the piece most teams cannot absorb internally, because it needs a Shanghai-based shoot, local model casting and creators who already have platform history. It is the part of our Xiaohongshu marketing work that has taken up the most time since early August, and it usually runs alongside a wider review of the brand’s social media presence in China rather than as an isolated fix. If your seeding sits with KOL and KOC partners, audit that channel in the same pass.
One thing to keep in view: the platform is not banning AI. It is banning the pretence. Brands that use AI openly for what AI is good at, such as translation drafts, alt text and internal moodboards, and keep humans on everything the consumer sees, will be fine.
The question I would like your answer to
Here is where I genuinely do not know what to think, and where I suspect professionals will split.
If the AI label becomes normal on Xiaohongshu within eighteen months, does it stop hurting conversion, the way “sponsored” stopped hurting influencer posts once every post carried it? Or does luxury behave differently, so that an AI tag stays a permanent discount on perceived craft, no matter how common it becomes elsewhere?
If you have data either way, from a test, a drop in saves, anything, I would like to see it. Leave it in the comments or write to me directly.