Citeshine 引耀

Citeshine helps brands find and fix what AI gets wrong about their products. ·

[Citeshine 引耀 @citeshine]

New, and nobody writes about you yet? Start with the listings you control: fill in every field on your directory and comparison-site profiles. Half-empty listings get summarized as half-empty.

[Citeshine 引耀 @citeshine]

Thread: where to earn honest mentions when you're new and nobody writes about you yet.

[Citeshine 引耀 @citeshine]

买家下单前，常会问 AI 一句：「这个产品能退款吗？」

退款规则如果只写在结账页的小字里，或者藏在帮助中心好几层下面，AI 很可能答得含糊，甚至答错。

我们的建议：单独一页写清 3 件事，退款期限、怎么申请、哪些情况不退，再从定价页链过去。改完用同一句再问一次，看回答和页面对不对得上。

[Citeshine 引耀 @citeshine]

我们复测 AI 怎么介绍一个产品时，每条回答都记 8 项：

日期、引擎、登没登录、问题原文、第几次问、有没有提到产品、引用了哪些网址、说错的那句原话。

最有用的是最后两列：知道它读了哪页、错在哪句，才知道该改哪里。

同一个问题至少问 3 次，只问 1 次的结果先别下结论。

[Citeshine 引耀 @citeshine]

@phuclh93 Nice. One check worth building in: when a keyword drops, open that query in GSC's Pages tab and see if the ranking URL changed. A lot of "drops" are Google swapping in a different page from the same site, and the fix is internal links or merging pages, not new content.

[Citeshine 引耀 @citeshine]

@lilyraynyc One check we'd add if you already built a page just for LLMs: put it next to the pages humans see. Same prices, same claims, same limits?

Anything that only lives in the bot version, move it to a real page or cut it. One source, one place to keep current.

[Lily Ray 😏 @lilyraynyc]

I’m working on an enterprise brand’s website, and something interesting stood out in its Google Search Console Gen AI report:

The most cited page on the company’s website in AI Overviews & AI Mode, with 5x more impressions than the homepage, is its “About” page.

It’s a comprehensive, detailed page about the company’s history, background, and evolution over time.

It’s not a “special page” created for LLMs or AI agents. It’s written for humans, visually appealing, and contains detailed information about the brand in a format that both humans and LLMs can easily understand.

So what’s the big “GEO” takeaway here?

Create comprehensive “About” pages with detailed, accurate information about your company’s history.

In other words, the same recommendation many of us in SEO have been making for years, especially since E-A-T became a bigger part of the conversation in 2017 (and long before that, in many cases).

Why do I think this is a more sustainable approach than creating separate pages specifically for LLMs?

Because those pages create an easily gameable surface.

Notably, Google, OpenAI, and Anthropic haven’t explicitly asked website owners to create separate, LLM-specific versions of their content, use Markdown, etc. I would change my perspective if they did (they won’t.)

The problem: it’s inevitable that bad actors (and plenty of brands) will use these pages to inject prompts, exaggerate claims, or give AI agents promotional information that doesn’t match what real users see, with the goal of manipulating what AI says about their brand.

My prediction is that as this behavior becomes more widespread (it’s already spreading thru the GEO world like wildfire), the AI companies will increasingly develop ways to ignore, discount, or otherwise treat these pages as untrustworthy.

And depending on how these systems evolve, the consequences might extend beyond the individual pages themselves.

The unfortunate part is that even well-intentioned companies simply trying to “help” LLMs understand their content better could become collateral damage. Even if their LLM-specific pages contain exactly the same accurate information as their human-facing website, they could still get caught up in broader efforts to crack down on this type of content, or flag the sites using “manipulative” approaches.

This is why I’m extremely cautious about employing tactics like this in the first place.

Meanwhile, there’s a much simpler way to work on this:

Take all that time and effort you’re spending creating special content for LLMs and use it to improve the information available to actual humans on your website.

Make your company’s history, expertise, people, products, and reputation easy to find, understand, and verify.

You’re still giving LLMs the information they need; you’re just doing it in a way that’s actually useful to your human readers, too.

And unlike tactics designed to exploit how AI systems currently work, this is an investment that’s unlikely to become a liability when those systems inevitably start cracking down.

[Citeshine 引耀 @citeshine]

Renamed your product? Add one plain line like "Formerly called X" to your homepage and docs. Without it, old reviews and new pages read like two different companies.

[Citeshine 引耀 @citeshine]

If you only fix one thing in your schema this week, make it sameAs: point your Organization markup at your real X, LinkedIn and GitHub profiles. That's the line that ties them to one company.

[Citeshine 引耀 @citeshine]

1/ Thread: the schema markup we'd add to a SaaS homepage first. Not for rich results. Just so machines get the basic facts right.

[Citeshine 引耀 @citeshine]

Write one plain sentence that answers "what is [product]?" Then use that exact sentence on your homepage, docs, and every profile bio. When your own pages disagree, assistants pick one at random.

[Citeshine 引耀 @citeshine]

产品改过名的，查一下 AI 还认不认旧名：

1. 用旧名问它「是做什么的」，看它会不会提到新名字
2. 用新名问，看它会不会把旧名时期的功能和价格算到你头上
3. 在官网关于页和文档首页写一句：原名是什么、什么时候改的

第 3 步最省事。新旧两个名字写在同一句里，AI 更容易把它们对上。

[Citeshine 引耀 @citeshine]

@blankwebdev 可以先在谷歌搜索框敲「remove bg」，看下拉里跟着哪些词，比如 free、api、批量、高清。这些长尾需求不一定都会跟着去 Canva：如果并入后要登录、进编辑器才能用，「免登录」「批量处理」这类需求就接不住。新站与其硬抢主词，不如一个需求做一页，标题直接写这个需求。

[Citeshine 引耀 @citeshine]

@iamjianyuwu 说这条链最好倒着想，先想清楚怎么赚钱，再回头挑词。放到 GEO 上也一样。

别从关键词开始测，从付过钱的客户问过的话开始：把最近 5 条售前咨询的原话，原样丢给 DeepSeek 和 ChatGPT，看回答里有没有你、说得对不对。

跟成交无关的问题，答得再好也只是曝光。

[Jianyu Wu @iamjianyuwu]

做站进度更新：今天还在练习挖新词的手感

中午哥飞老师群里分享出海建站的整套技能，一共六步，每一步真正要弄清楚的事都不一样

1. 挖掘关键词，找需求。人们在主动找什么？哪些问题反复出现？

2. 理解关键词背后的需求。谁在什么场景下搜？他最终想完成什么？现有的办法哪里不好？

3. 做网站，满足需求。用内容、工具、商品或服务，帮他拿到想要的结果。

4. 获取流量。靠 SEO（搜索引擎优化）、SEM（付费搜索广告）、GEO（生成式 AI 搜索优化），让有这个需求的人找到你。

5. 承接流量。用户进来以后，能不能马上看懂、建立信任，然后用起来、来咨询或者下单。

6. 流量变现。用户付费、广告、推荐佣金。

说说我的两个感受：

一是决定成败的是两头：第 1、2 步的判断，第 6 步的变现。中间三步是执行，AI 已经让它越来越便宜。我现在卡住的恰恰是两头，不是中间。

二是这条链最好倒着想，先想清楚怎么赚钱，再回头挑词。社群某月新词比赛的第一名，一个月 100多万访问，收入只有几百美金。流量大，不等于赚钱。（这一点我做某书的时候就深有体会，5000粉丝一个商单都没有）

我之前的问题就是一上来先建站，做完后发现两头哪头都不占

所以接下来每个候选词，都需要搞清楚搜这个词的人想要什么，我用什么满足，靠什么赚钱，依据是什么

想不清楚的，就不做。

也想问问做过站的朋友：你们选词的时候，是先看词，还是先想好怎么赚钱？

[Citeshine 引耀 @citeshine]

开个新系列：「AI 怎么介绍出海产品」。

每期挑一个出海产品，同一个问题分别问 DeepSeek、豆包和 ChatGPT，截图原样贴出，标出哪句对、哪句错，再给一处能改的地方。

规则：不登录，同一问题问 3 次，只写我们真跑过的结果，只测产品方自己提交或同意的产品。

想被测的产品方，评论区留名字。

[Citeshine 引耀 @citeshine]

@siruruku 同意拆开写。还有个细节：限制别只放在悬停提示或折叠层里，直接在套餐卡片下面用一行普通文字写清楚，比如「付款：支持 Visa/万事达，暂不支持 PayPal」。买家一眼能看到，AI 读页面时也拿得到。

[Citeshine 引耀 @citeshine]

卖到海外，买家常先问一句：「我这里能用吗？」

这个答案最好在官网上写成文字：支持哪些国家和地区、界面有哪些语言、收哪些付款方式、发货或服务覆盖到哪。

只放在图片、弹窗或客服聊天里的话，买家搜不到，AI 回答时也只能靠猜。

[Citeshine 引耀 @citeshine]

@Wuc436 恭喜过审。做 SEO 的同时可以顺手测一下 AI 搜索：拿用户真会问的一句话，比如「有没有做 XX 的工具」，分别问 ChatGPT 和 Perplexity，看你的站有没有被提到、被引用的是哪一页。每周固定测同样几句，变化就看得出来。

[Citeshine 引耀 @citeshine]

ChatGPT Search and Google AI Mode don't lean on the same kinds of sources, and they cite a different number of pages per answer. So check each one separately, and look at which pages get cited, not just whether you're named.

[Citeshine 引耀 @citeshine]

Your About page now has two readers: a buyer skimming it, and a model summarizing it. Say what you do, who it's for, and what you're not, in the first two sentences. Clever taglines don't summarize well.

[Citeshine 引耀 @citeshine]

Question for founders: when an AI assistant gets your product wrong, what do you fix first? Your own site, or the third-party page it's probably quoting? We lean toward the second. Curious what others do.

[Citeshine 引耀 @citeshine]

问一个具体的问题：

如果 AI 只能读你官网的一页来介绍你，你会让它读哪一页？首页、定价页、FAQ，还是关于我们？

我们会选 FAQ，因为它最接近用户真正去问 AI 的那几句话。你们会选哪页？

[Citeshine 引耀 @citeshine]

@hpf1908 第1点很真实。用户找同步方案时，搜的多半是问题，不是产品名，比如「Obsidian Sync 替代」「iCloud 同步冲突」「多端同步丢笔记」。可以按这几类问题各写一页排错或对比，把和官方 Sync、Remotely Save、iCloud 的差别写清楚。Google 和 AI 回答被问到时，才有你的页面可以引用。

[Citeshine 引耀 @citeshine]

@jiaozixyz @loki_yan_seo 115 次如果是日志里的 ChatGPT-User，可以再拆开看：ChatGPT-User 是用户在对话里触发的抓取，OAI-SearchBot 是给 ChatGPT 搜索建索引，GPTBot 是训练用。三个分开统计，才分得清是真有人在问，还是只是爬虫来过。UA 能伪造，最好再对一下 OpenAI 公布的 IP 段。