Account suspended for 7 days. No explicit violation notice — just a suspension. Content was tech tutorials, no product promotion.
After reviewing flagged posts, the pattern was clear. The problem wasn't:
It was tone words:
| High-risk phrase | Why it's flagged |
|-----------------|-----------------|
| 颠覆级 (revolutionary) | Superlative marketing language |
| 必学 (must-learn) | Creates urgency, ad pattern |
| 绝对需要 (absolutely need) | Prescriptive, ad copy structure |
| 神器 (magic tool) | Hype word |
| 冲 (go for it) | Call-to-action, ad pattern |
The platform's content classifier treats these as commercial promotion signals, regardless of whether there's an actual product being sold.
Content that consistently passes moderation shares these traits:
1. Specific scenario: "I spent 3 days debugging this" not "Here's a must-know trick"
2. Reproduce the problem: Show the error message, the environment, the failed attempt
3. Use data: Timestamps, error codes, screenshots — concrete evidence
4. No prescriptive framing: Describe what you did, not what readers "must" do
| Original (risky) | Rewrite (safe) |
|-----------------|----------------|
| 这个神器你必须收藏 | 我用了三天，记录一下踩过的坑 |
| 颠覆级工具来了 | 试了一下，有几个地方和文档不一样 |
| 绝对需要学这个 | 我在这里卡了两小时，原因是 |
The rewrite shifts from "recommending to readers" to "documenting your own experience."
Based on observation:
Rapid-fire posting on the same topic triggers the commercial promotion classifier even without any high-risk words.
Xiaohongshu's moderation model isn't primarily a keyword filter — it's a pattern classifier that looks at the overall tone and posting behavior. Writing like a person documenting their experience is the most durable strategy.