Chinese AI-Writing Style Audit (中文 AI 味检查)
An 11-rule checklist for stripping the tell-tale patterns of unedited AI-generated Chinese copy — em-dash chains, 不是而是 contrast pairs, 首先其次 sequence stacking, 被 passive chains — refined from real user correction feedback, each with a precise trigger, bad/good example, and scope.
AI writing detection tools and de-slopping guides are almost all built for English — but Chinese AI-generated copy has its own unmistakable tells, from chained em-dashes to repeated 值得注意的是 preambles, and nothing English-language guidance catches them.
Who it's for: Chinese-language content marketers and product-announcement writers, developer-relations teams writing Chinese READMEs and release notes, anyone drafting WeChat/公众号 articles who's been told the copy 'AI 味太重', bilingual teams needing a Chinese equivalent of English AI-writing detection
Example
"帮我改一下这段产品发布稿，AI 味有点重" → Each of the 11 rules checked in turn against the draft — connected em-dash chains split into proper sentences, repeated 不是而是 contrasts replaced with direct facts, emoji stripped from formal copy, overloaded long sentences split by their actual information layers — each fix backed by a stated trigger and scope so the rule never gets misapplied outside the document type it was calibrated for
CLAUDE.md Template
New here? 3-minute setup guide → | Already set up? Copy the template below.
npx ccplaybooks add chinese-ai-writing-style-audit# Chinese AI-Writing Style Audit (中文 AI 味检查)
A checklist for removing the tell-tale patterns of unedited AI-generated Chinese copy ("AI 味") from product announcements, WeChat/公众号 articles, emails, and READMEs — refined from real user correction feedback rather than written from theory. Each rule states a precise trigger, a detection approach, a bad example, a good example, and a scope (which document types it applies to) so it's never applied outside the context it was calibrated for.
## When to Load
Load this when drafting or rewriting reader-facing Chinese copy (product launch posts, WeChat articles, emails, README sections), or when a user says the text "AI 味太重" (too much AI flavor) or "不像人写的" (doesn't read like a human wrote it).
## How to Use
While drafting or revising, check the text against each rule below in turn. Every rule defines its own scope — never apply a rule outside the document types it names, even if the pattern shows up elsewhere.
Get new playbooks like this one
One email a week with new Claude Code workflows. Free, like everything here.
No spam. Unsubscribe anytime.
What This Does
An 11-rule checklist for auditing Chinese reader-facing copy — product announcements, WeChat/公众号 articles, emails, README sections — for the specific patterns that mark it as unedited AI output. Each rule is a precise, self-contained unit: a trigger condition (what has to repeat or combine to actually count as a violation, not just any single instance of the pattern), a detection approach, a genuine bad/good example pair, a concrete fix, and a scope stating exactly which document types it applies to.
The patterns cover ground English-language AI-writing guides don't touch: chaining em-dashes in place of normal punctuation, repeating "不是……而是……" (it's not X, it's Y) contrast pairs, stacking "让我们" rallying-cry openers, mechanically sequencing every sentence with 首先/其次/最后, vague "在这个……时代" era-framing openers, emoji density in formal copy, forced parallel structure, empty poetic metaphors standing in for real information, overloaded long sentences, repeated "值得注意的是" preambles, and chained 被-passive constructions. It was refined from real user correction feedback, not written from theory alone.
Quick Start
Step 1: Create a Project Folder
mkdir ai-style-audit && cd ai-style-audit
Step 2: Download the Template
Click Download above, then:
mv ~/Downloads/CLAUDE.md ./
Step 3: Audit a Draft
claude
Paste in the Chinese copy you want audited and ask Claude to check it against the 11-rule checklist, or just say "这段文案 AI 味有点重，帮我改一下" (this copy has too much AI flavor, help me fix it).
Tips & Best Practices
- Detection here is judgment-based, not a mechanical string match — every rule's trigger depends on repetition, scope, and whether the pattern is doing real work, so read each passage in full context before flagging it.
- Respect each rule's stated scope. A rule calibrated for formal product announcements (like the emoji rule) shouldn't be applied wholesale to a casual internal chat message.
- When several rules fire in the same passage, fix the most disruptive one first — usually the em-dash chain or the overloaded sentence — since smaller issues often resolve naturally once the structural ones are fixed.
Limitations
- Calibrated specifically for Chinese-language reader-facing copy — the patterns here are language-specific and don't transfer directly to English AI-writing detection (pair with an English-focused humanizing playbook for that).
- Every rule's "does not trigger" exception exists for a reason — don't mechanically flag every instance of a pattern without checking whether the exception applies to this specific passage.
- Refined from real correction feedback on specific document types (product posts, WeChat articles, emails, READMEs) — a genuinely different genre, like fiction or academic Chinese writing, may need different calibration.