GEO Audit AI 搜索可见性 EN

买家问 AI 要推荐时,答案里有你吗?

绝大多数站长看不到这件事:AI 引擎从不告诉你它引用了哪些来源,访客到来时也不会带上说明来路的 referrer。这个工具在你的网站上放一台测量仪,把引擎真正说过的话报告给你。

让问给每个引擎的买家问题更贴合你的业务。

选填:品牌名,以及需要核对的事实

仅当品牌名与域名不一致时才需要填写。

每行一条,格式为 what AI might say => what is true。我们只核对你自己提供的事实,绝不替你猜。

每天三次免费扫描。每次扫描都会真实调用 AI 接口,是有成本的——限额只有这一个原因。

示例读数 acme-crm.example.com
ChatGPT Named you 2nd of four, described as “built for small teams”. Your pricing page cited. Mentioned
Perplexity Named a competitor four times, never you. Cited a two-year-old review you cannot edit. Absent
DeepSeek Answered from training memory without searching the web, so it measures what it already knows about you. Not compared
AI Overviews Requires Google Search infrastructure this deployment does not have. Not measured
Illustrative output, not a customer result. Every score we report is generated from a real scan of a real site.

扫描到底检查什么

所有检查分五组:爬虫可达性、实体清晰度、答案优先结构、实时提问、技术卫生。免费扫描和完整审计跑的是同一套检查,区别在实时提问:免费扫描问 3 个买家问题,完整审计问 10 个,并额外给出声量份额、来源分类和 AI 引荐对账。

AI 到底能不能读到你的页面

按 RFC 9309 用 15 个爬虫逐个测试你的 robots.txt,并检查页面在不执行 JavaScript 时是否可读。挡住一个回答型爬虫,就等于在那个引擎里彻底消失,而绝大多数站长从未察觉。

引擎实际说了什么

把真实的买家问题分别问给每个已配置的引擎,每次都问两遍,然后分析答案里有没有你、排在第几、那句话的语气,以及它还提到了谁。

按顺序告诉你改什么

修复清单按「每小时投入的收益」排序,每一条都附带可以直接粘贴的文件内容——robots.txt、llms.txt、JSON-LD、FAQ 区块、对比页。

审计遵守的三条规则

这三条不是关于准确性的宣传语,而是代码里的硬约束,每一条都有测试守着,行为退化就会失败。

  1. 它不会编造一个分数

    If your site blocks our crawler or does not resolve, the result is not scored with a reason — not a low score that implies we looked and found problems. We would rather show you a gap in our own measurement than a number we cannot defend.

  2. 它不会因为你做了选择而扣分

    Blocking GPTBot is a legitimate opt-out that has no effect on whether ChatGPT cites you, so it never costs points. Only crawlers that feed answers are scored. A tool that penalises this is telling you to make a decision against your own interest.

  3. 它把「知识」和「可见性」分开看

    An engine that searches the live web and misses you has a content problem you can fix. An engine answering from training memory and missing you has a knowledge problem that on-page edits will not solve. The report labels which is which instead of blending them.

从你自己的域名开始

免费扫描大约三十秒,不需要你做任何准备——不用注册,不用装代码,也不用给分析工具的权限。

Run the free scan