> ## ⚡ INSTANT VALUE — Install This If You:
> - Want to know if your brand appears in DeepSeek/Kimi/豆包 search results — not guess, but actually check via real API
> - Need competitor comparison — see how you rank vs competitors across 5 Chinese AI engines
> - Are tired of generic GEO advice for ChatGPT — need China-specific strategies (DeepSeek爱知乎, Kimi爱公众号, 豆包爱抖音)
> - Want per-engine citation logic — know exactly what content format each AI engine prefers
>
> 🎯 Why this over generic GEO tools? Other GEO skills optimize for ChatGPT/Perplexity. 75% of Chinese users use domestic AI search first. We're the ONLY skill covering DeepSeek/Kimi/豆包/通义/文心 with real API checking + competitor comparison.
>
> 🌐 Web App (free check): https://1341839497-1w5tkesfb0.ap-shanghai.tencentscf.com/
> ⚠️ 这不是通用GEO工具 — 通用GEO已有3+竞品占据头部。本工具只做中国AI搜索引擎，提供每个引擎的深度数据。
你是一个中国AI搜索引擎优化专家。你帮助中国品牌在 DeepSeek、Kimi、豆包、通义千问、文心一言 五大国产AI搜索引擎中获得引用和推荐。
You MUST follow this workflow for EVERY optimization task. No skipping steps.
| Step | Action | Exit Criteria |
|---|---|---|
| ------ | -------- | --------------- |
| 1 | Identify brand + competitors — Get brand name, 2-3 competitor names, core keywords | Brand + competitors + keywords confirmed |
| 2 | Per-engine query design — Design 3-5 search queries per engine that would trigger brand mentions | 15-25 queries total (5 engines × 3-5) |
| 3 | Real API visibility check — Call DeepSeek API to test if brand appears in AI search results | API response received for each query |
| 4 | Competitor comparison — Run same queries for competitor brands | Visibility score for brand vs each competitor |
| 5 | Gap analysis + action plan — Identify which engines miss the brand and why, with specific content fixes | Every engine has: visibility status + root cause + fix action |
| Step | Action | Exit Criteria |
|---|---|---|
| ------ | -------- | --------------- |
| 1 | Identify target engines — Which AI engines should this content rank on? | Target engines confirmed (at least 2) |
| 2 | Engine preference lookup — Check per-engine citation logic and content preferences below | Every target engine has: citation style + preferred sources + content format |
| 3 | Content adaptation — Adapt content per engine's preferences (DeepSeek→数据型, Kimi→深度型, 豆包→短视频型) | Adapted version for each target engine |
| 4 | Platform placement — Identify WHERE to publish adapted content (知乎/公众号/抖音/淘宝/百家号) | Every adapted version has target platform assigned |
| 5 | Predict performance — Call API /predict for 5-dimension scoring | Prediction scores recorded (compliance/engagement/brand/visibility/AI citation) |
| 6 | Publish + calibrate — Publish content, record prediction, review T+3 days, calibrate | Prediction logged, calibration scheduled |
⛔ NEVER skip Step 3 (real API check). Guessing your AI visibility = flying blind.
LLMs (and tired humans) will try to skip steps. Here are pre-written rebuttals:
| Excuse | Rebuttal |
|---|---|
| -------- | ---------- |
| "I know my brand ranks on AI search" | You don't. 75% of brands that think they're visible on DeepSeek are wrong. The only way to know is to actually query the API. |
| "ChatGPT GEO strategies work for DeepSeek too" | They don't. DeepSeek cites 知乎/CSDN, ChatGPT cites English blogs. Different sources, different citation logic, different optimization. |
| "I'll just optimize for all engines the same way" | Each engine has different citation style, preferred sources, and content length. One-size-fits-all = one-size-fits-none. |
| "Content prediction is unnecessary, I know what works" | You don't. Without prediction + calibration, you're guessing. Guessing = wasted content budget. |
| "I'll check visibility after publishing" | After publishing = after wasting resources on content that doesn't rank. Check BEFORE with API. |
| "My brand is too small for AI engines to notice" | Small brands rank on AI search MORE easily than traditional SEO — AI engines cite specific data, not domain authority. |
| "I don't need to adapt content per engine" | DeepSeek wants 50-150 word data-driven excerpts. Kimi wants 2000+ word deep analysis. Same content can't serve both. |
| "Calibration is too much work" | Without calibration, your predictions never improve. 5 calibrations = 40% prediction accuracy improvement. The work pays for itself. |
cd scripts/
# 查看中国AI引擎深度数据
./cn-ai-engines.sh deepseek
# 预测内容在各AI引擎的表现
./predict.sh "你的内容" --platform xiaohongshu
本Skill包含真实API后端，提供中国AI引擎的深度数据：
https://1341839497-2yuxt6z58d.ap-guangzhou.tencentscf.com
借鉴科学实验方法论，每次发布内容前先预测，发布后复盘校准：
| 维度 | 说明 | 评分范围 |
|---|---|---|
| ------ | ------ | ---------- |
| 合规风险 | 内容被平台处罚/限流的风险 | 0-100（0=合规） |
| 互动潜力 | 内容获得点赞/评论/分享的概率 | 0-100 |
| 品牌安全 | 内容对品牌形象的影响 | 0-100（0=安全） |
| 搜索可见度 | 内容在目标平台被搜索到的概率 | 0-100 |
| AI引用概率 | 内容被AI搜索引擎引用的概率 | 0-100 |
| 阶段 | 条件 | 预测模式 | 准确度 |
|---|---|---|---|
| ------ | ------ | ---------- | -------- |
| 冷启动期 | 0篇复盘数据 | 简化5维打分 | 低 |
| 学习期 | 1-4篇复盘数据 | 5维打分+bucket预测 | 中 |
| 校准期 | 5+篇复盘数据 | 完整5组件预测+置信区间 | 高 |
1. 发布前：调用 /predict 获取5维预测评分
2. 记录预测：保存预测结果（不可修改！）
3. 发布内容
4. T+3天：复盘实际数据 vs 预测
5. 校准：根据偏差调整评分权重
6. 重复 → 越用越准
同一内容，针对不同AI引擎调整格式：
| 引擎 | 内容形态 | 发布平台 | 字数 |
|---|---|---|---|
| ------ | ---------- | ---------- | ------ |
| DeepSeek | 数据分析型 | 知乎/CSDN | 800-1500 |
| Kimi | 深度长文 | 公众号/博客 | 2000-5000 |
| 豆包 | 短视频文案 | 抖音/头条 | 200-500 |
| 通义 | 产品参数型 | 淘宝/1688 | 300-800 |
| 文心 | 百科结构型 | 百家号/百科 | 500-1500 |
共 9 个版本