Xiaohongshu prohibited-word checker
Check Xiaohongshu note titles, bodies, comment scripts and product copy against a built-in list of common prohibited and sensitive wording: Advertising Law absolute claims, medical-aesthetics and efficacy claims, off-platform funnels, counterfeit and dropshipping phrases — with positions, context and suggested rewrites, plus width/traditional normalisation, whitelist, custom words and masking.
Runs in your browserEvery computation happens in your browser — your data never leaves this device.
The list is a curated set of common prohibited/sensitive wording, not Xiaohongshu's official full list; platform rules change often, so results are for self-checking only and are not legal advice.
Uncheck a category to skip it — the list is organised by category.
Detection report
Matched locally against the built-in list with leftmost-longest overlap handling; for pre-publish self-checks only.
What this tool does
- Pre-publish self-check: paste a note title, body, comment script and product description and get every hit at once — absolute claims, medical-aesthetics and efficacy claims, off-platform funnel codes, counterfeit and dropshipping phrases, income promises — each with a suggested rewrite.
- Built for beauty and medical-aesthetics notes: project names such as 水光针, 热玛吉, 玻尿酸 and 溶脂针, plus special-cosmetics claims like 祛斑, 美白, 防脱 and 减肥, are grouped under Medical claims.
- Filter by category: keep only the funnel phrases (微信号, vx, 暗号, 私信) or only the absolute wording by unchecking the others; the risk score and per-category counts follow the filter.
- Maintain your own whitelist and extra words: whitelisted brand or clinic names are reported separately as ignored, and internal banned words can be added line by line, then masked for a publishable version.
Example
Input
这瓶水光针打完真的无副作用，姐妹们私信我拿暗号，链接在主页。
Output
Risk level: high (4 hits, 0 ignored) By category: Medical claims 2; Platform violations 2 Hits: 1. 水光针 [Medical claims] chars 3-5 -> suggest: 补水护理 context: 这瓶水光针打完真的无副… 2. 无副作用 [Medical claims] chars 10-13 -> suggest: 温和配方 context: …光针打完真的无副作用，姐妹们私信… 3. 私信 [Platform violations] chars 18-19 -> suggest: 评论区留言 context: …作用，姐妹们私信我拿暗号，链… 4. 暗号 [Platform violations] chars 22-23 -> suggest: 删除该表述 context: …妹们私信我拿暗号，链接在主页… Masked preview: 这瓶***打完真的****，姐妹们**我拿**，链接在主页。
Positions are 1-based, inclusive character ranges; the masked preview replaces each hit with * by default. Hit titles show the slice from your original text (traditional or full-width input stays as typed) — normalisation is used for matching only.
Frequently asked questions
Is this Xiaohongshu's official word list?
No. The built-in list is a curated set of common prohibited and sensitive wording drawn from Article 9/17/25 of China's Advertising Law, the medical-aesthetics service rules, the special-cosmetics and health-food rules, and Xiaohongshu's community, brand-partnership and commerce note rules — 209 entries (58 absolute wording, 44 medical claims, 68 platform violations, 26 finance, 13 superstition). It covers common phrasing only and is not an official full list. Platform rules change often, so treat the result as a pre-publish self-check, not legal advice.
Why are project names such as 水光针 or 热玛吉 flagged?
Medical aesthetics is a medical procedure, so recommending a specific project, implying a treatment outcome or steering readers to a private clinic is high-risk content. The list therefore includes common project names (水光针, 热玛吉, 光子嫩肤, 玻尿酸, 溶脂针, 美白针) under Medical claims. This is a text-level hint only — whether you may publish and whether it needs a partnership disclosure depends on the current platform rules and your account.
Does it catch variants like vx, 威信 or 暗号?
Yes. The list includes the common homophone and abbreviation spellings (vx, v信, 威信, 薇信, 扣扣, qq号, 暗号, 私信, 小红书号). Before matching, the text is converted from full-width to half-width, from traditional to simplified and case-folded, so full-width letters or traditional characters used to dodge filters are still caught.
How does the matching work — will one word be reported many times?
It uses Aho–Corasick (a trie with failure links) so the text is scanned once regardless of dictionary size, then resolves overlaps leftmost-longest: 销量第一名 is reported once as 销量第一. Normalisation handles full-width to half-width (％ to %), traditional to simplified (333 built-in single-character pairs, 網 to 网) and case folding; all three are 1:1 substitutions, so positions map back to the original text exactly. Normalisation is used for matching only and never rewrites the display: for the input 無副作用 the hit title stays 無副作用 [Medical claims] at chars 1-4, in line with the highlight, the context and the report.
How is the risk level calculated, and is my text uploaded?
Each hit scores by category: 2 points for absolute wording and platform violations, 3 for medical claims and finance. A total of 0 is "none", 1–3 "low", 4–9 "medium" and 10 or more "high"; whitelisted hits score nothing. Everything runs locally in your browser — no network requests, no logging — and the limit is 20000 characters per run.
Keywords：小红书违禁词小红书敏感词检测违禁词检测医美违禁词笔记限流自查xiaohongshu prohibited wordsbanned words checkernote compliance