Start here: how to operate this page
Use this page as a control room. It does not replace ChatGPT, Gemini or Perplexity; it tells you what to test, how to judge the answer, and what to fix next.
先从这里开始：这个页面是控制台，不是 AI 本身。它告诉你要问哪些问题、怎样判断 AI 有没有答对、答错之后应该修哪里。Click the black button first. If the score is not 100, fix the missing Applied West AI source pages before testing AI.
先按黑色按钮。如果不是 100，先修官网资料，不要急着问 AI。
Pick one test question below and copy it into ChatGPT, Gemini, Perplexity, Doubao or Yuanbao.
在下面选一条测试题，复制去 ChatGPT / Gemini / Perplexity / 豆包 / 元宝。
Compare the AI answer with the pass signal. Mark whether it recognized Applied West AI, confused the brand, or asked for more proof.
用「合格标准」对照答案：AI 是否认得 Applied West AI？有没有认错品牌？有没有说证据不足？
If AI is wrong, choose the matching repair item: entity phrase, disambiguation, proof, answer page or external source.
如果 AI 答错，就去 repair queue 找对应修法：品牌词、消歧义、案例证据、问答页、外部资料。
Fix the website, LinkedIn, Google Business Profile, GitHub README or other source surfaces. Do not invent proof.
回到官网、LinkedIn、Google Business Profile、GitHub README 等地方补资料。不能造假 proof。
After indexing or source refresh, run the same prompt again. GEO is a monitor, backtest and correction loop.
等收录或资料刷新后，用同一条 prompt 再测一次。GEO 是监控、回测、修正的循环。
Setup health
Before judging GEO quality, confirm whether the AI providers are actually connected. `not_configured` is a setup gap, not an AI verdict.
先看设置健康度：如果 provider 是 not_configured，只代表 API key 未接通，不能当成 AI 不认识 Applied West AI。1. Adversarial testing tool
Run these prompts in ChatGPT, Gemini, Perplexity, Doubao and Yuanbao. Record whether Applied West AI is recognized, confused, recommended, absent, or criticized.
反向测试工具：把这些问题复制到不同 AI，观察它是否认得 Applied West AI、是否认错、是否推荐、是否说不知道或提出质疑。6. AI response backtest
Choose one prompt, select the AI providers, then run a live API backtest. The page will show each AI reply, automatic labels, and the next repair suggestion.
AI 回复回测：选择一条 prompt 和要测试的平台，按一次按钮就会调用后端 API。页面会显示每个 AI 的回复、自动标签和下一步修正建议。 注意：not_configured 只代表 API key 未接通，不代表 AI 不认识 Applied West AI。只有真实模型回复后，recognized / unknown 才有 GEO 判断意义。| Provider | Status | Recognition | Risk | Repair |
|---|---|---|---|---|
| Run AI backtest to populate the result table. | ||||
| Risk | Cause | Provider | Repair action |
|---|---|---|---|
| No risk matrix yet. | |||
2. Brand knowledge base check
This checks whether the site exposes the source pages an AI assistant would need: entity facts, disambiguation, region evidence, prompt monitoring, answer pages, case proof and llms.txt.
品牌知识库检查：测试官网有没有把 AI 需要的资料放清楚，例如公司事实、消歧义、地区证据、问答页、案例 proof 和 llms.txt。4. Content repair queue
When AI answers are wrong, use this queue to decide what to add or fix. The repair is always source-first: page, proof, schema, then re-test.
内容修复队列：当 AI 答错时，不是直接改 prompt，而是回去修 source page、proof、schema，然后再测。