[{"data":1,"prerenderedAt":395},["ShallowReactive",2],{"blog-cluster-internal-links":3,"blog-prev-cluster-internal-links":129,"blog-next-cluster-internal-links":225},{"id":4,"title":5,"author":6,"body":7,"canonical":113,"category":114,"cover":113,"date":115,"description":116,"draft":117,"extension":118,"keywords":113,"meta":119,"navigation":120,"noindex":117,"ogImage":113,"path":121,"seo":122,"seoDescription":113,"seoTitle":113,"stem":123,"tags":124,"updated":113,"__hash__":128},"blog\u002Fblog\u002Fcluster-internal-links.md","簇页内链的规矩：每页链几个、锚文本怎么写","元析智能",{"type":8,"value":9,"toc":100},"minimark",[10,14,19,22,25,29,32,35,39,42,45,49,52,55,58,61,64,67,70,73,76,79,82,85,88,91,94,97],[11,12,13],"p",{},"2025 年 11 月，我们给一个做日志平台的客户重排了内链，只动链接，正文一个字没改。六周后，6 个簇页的平均排名从 26 到 13，支柱页的抓取频次涨了近一倍。",[15,16,18],"h2",{"id":17},"规矩一每篇簇页链回支柱页位置在开头","规矩一：每篇簇页链回支柱页，位置在开头",[11,20,21],{},"开头一两百字里放一次，不要拖到文末。文末的链接，用户看不到，引擎给的位置权重也低。",[11,23,24],{},"只放一次。有些站开头一次、中间一次、结尾一次，重复链同一页，看起来像在凑数。",[15,26,28],{"id":27},"规矩二每篇簇页再链两篇同簇页","规矩二：每篇簇页再链两篇同簇页",[11,30,31],{},"同一簇里的页面互相链，选择标准是上下文相关，不是列表随机挑。",[11,33,34],{},"我们的习惯是：正文里提到某个相关问题时顺手挂链接，锚文本就用那个问题的标题。文章末尾不再重复同一批链接。",[15,36,38],{"id":37},"规矩三一个锚文本只指向一个页面","规矩三：一个锚文本只指向一个页面",[11,40,41],{},"同一个说法指向两个页面，等于告诉引擎这两个页面差不多，结果两个都排不好。",[11,43,44],{},"我们做内链检查时，第一件事就是把锚文本和目标 URL 导出来做一对一核对。这个检查用表格就能做，半小时能过完 300 篇。",[15,46,48],{"id":47},"规矩四单页内链-5-到-8-个","规矩四：单页内链 5 到 8 个",[11,50,51],{},"少于 5 个，页面容易变成孤儿。超过 12 个，权重摊得太薄，每个链接分到的都很少。",[11,53,54],{},"底部“相关文章”自动生成 20 个链接的做法，等于没有内链。我们给客户改模板时，第一刀就砍这个模块。",[15,56,57],{"id":57},"支柱页的链法不一样",[11,59,60],{},"支柱页每个子问题后面挂一个“展开看”的链接，一页挂 8 到 15 个，这属于导航性质，不受单页 5 到 8 个的限制。",[11,62,63],{},"但支柱页的链接要分组，按子主题排，别打散。分组之后，用户能看出这一簇的地图，引擎也能看出层级。",[15,65,66],{"id":66},"面包屑和目录页",[11,68,69],{},"面包屑要带主题层级，\u002Fblog\u002F 下按簇分目录，别所有文章平铺。目录页给每个簇一个入口，这个页面本身也能拿到主题词的排名。",[11,71,72],{},"我们那个客户的目录页在整簇重排后，三个月内排到了“日志分析”相关词的第二页，之前根本没收录。",[15,74,75],{"id":75},"一个提醒",[11,77,78],{},"内链不是一次性的活。新簇页发布时要同步更新支柱页和两篇同簇页，漏掉这一步，新页面要等好几个月才被发现。",[15,80,81],{"id":81},"内链和流量的关系别想得太玄",[11,83,84],{},"内链本身不会凭空创造流量，它做的是分配。把权重集中到该排上去的页面，把用户送到下一步。",[11,86,87],{},"我们那个客户站内链改完之后，整站流量只涨了 12%，但 6 个簇页里有 4 个进入前 20。如果只看全站流量，会以为改内链没用。",[15,89,90],{"id":90},"一个检查脚本",[11,92,93],{},"用爬虫把站内链接和锚文本导出成两张表，跑一次半小时。每季度跑一次，能发现死链、指向 301 的链、以及锚文本重复的页面。",[11,95,96],{},"死链是最常见的问题。2026 年 1 月我们查一个 400 篇的站，站内死链 37 个，其中 12 个是删文章时漏改的。",[11,98,99],{},"我们的做法是发布清单里固定一项：三处内链加一个面包屑，缺一项不算发布完成。",{"title":101,"searchDepth":102,"depth":102,"links":103},"",2,[104,105,106,107,108,109,110,111,112],{"id":17,"depth":102,"text":18},{"id":27,"depth":102,"text":28},{"id":37,"depth":102,"text":38},{"id":47,"depth":102,"text":48},{"id":57,"depth":102,"text":57},{"id":66,"depth":102,"text":66},{"id":75,"depth":102,"text":75},{"id":81,"depth":102,"text":81},{"id":90,"depth":102,"text":90},null,"技术SEO","2026-02-26","只动内链，不动一个字，六周后 6 个簇的平均排名从 26 到 13。这篇写我们固定用的四条规矩：链回支柱、同簇互链、锚文本唯一、单页出站 5 到 8 个。",false,"md",{},true,"\u002Fblog\u002Fcluster-internal-links",{"title":5,"description":116},"blog\u002Fcluster-internal-links",[125,126,127],"内链","主题簇","锚文本","jvesPINp4mBl_1Vjgd-CVAn46HN4cVaX1Q6wiVFsu7k",{"id":130,"title":131,"author":6,"body":132,"canonical":113,"category":114,"cover":113,"date":214,"description":215,"draft":117,"extension":118,"keywords":113,"meta":216,"navigation":120,"noindex":117,"ogImage":113,"path":217,"seo":218,"seoDescription":113,"seoTitle":113,"stem":219,"tags":220,"updated":113,"__hash__":224},"blog\u002Fblog\u002Ffont-subset-maintenance.md","字体子集上线三个月就失效，问题出在维护上",{"type":8,"value":133,"toc":207},[134,137,140,143,146,149,152,168,171,174,177,180,183,191,198,201,204],[11,135,136],{},"字体子集化不是做完就完了。去年 9 月我们给一个 SaaS 官网做的子集，三个月后新上的产品文案里出现 17 个缺字，前端只能临时回退全量包，首屏字体从 280KB 涨回 2.3MB，之前的速度优化等于白做。",[15,138,139],{"id":139},"子集为什么会失效",[11,141,142],{},"子集是按\"生成那一刻页面上的字\"切的。市场部改一版文案、加一个客户名、发一篇带生僻词的博客，字符集就不够了。浏览器遇到缺字会用回退字体渲染，视觉上跳出一个字形不一样的汉字，比豆腐块好一点，但一样难看。",[11,144,145],{},"这个站先是客服收到反馈，说标题里有个字特别细。前端查了两天才定位到字体子集。临时方案是回退全量包，代价是所有人都变慢。",[15,147,148],{"id":148},"把字符集变成构建产物",[11,150,151],{},"我们的做法是把字符集从\"一次性脚本\"改成\"每次构建自动生成\"：",[153,154,155,159,162,165],"ol",{},[156,157,158],"li",{},"构建时从 CMS 拉全站文案和博客正文，统计去重字符",[156,160,161],{},"去掉 ASCII 和标点，剩下的中文字符交给自己写的子集脚本",[156,163,164],{},"在统计结果上留 5% 冗余，按常用字表补一批",[156,166,167],{},"产物随构建走，版本号和内容一起变",[11,169,170],{},"好处是文案更新会触发子集更新，不会悄悄缺字。",[15,172,173],{"id":173},"缺字监控",[11,175,176],{},"构建只能覆盖 CMS 里的内容，用户输入、第三方评论、搜索关键词都覆盖不到。我们加了一条前端监控：用 canvas 测量几个生僻字在自定义字体下的宽度，和回退字体对比，不一样就上报。",[11,178,179],{},"这个方案不完美，但能抓到大部分问题。上线两个月抓到 4 次缺字，都是编辑器里直接粘进来的内容。",[15,181,182],{"id":182},"一些具体数字",[11,184,185,186,190],{},"这个站子集后首屏字体 280KB，覆盖 4200 个字符，",[187,188,189],"code",{},"font-display"," 用 swap。首屏文字在 1.1 秒内用回退字体渲染，1.4 秒切换成自定义字体，CLS 控制在 0.02 以内。",[11,192,193,194,197],{},"回退字体用系统黑体，字宽差异小，切换不明显。如果换成宋体这种字形差别大的，就得配合 ",[187,195,196],{},"size-adjust"," 微调，不然整段重排。",[15,199,200],{"id":200},"我的建议",[11,202,203],{},"内容更新频繁的站，别做纯静态子集，做构建期自动生成。内容几乎不变的站，比如产品官网，手工子集加每季度复查也行，但要把复查写进流程，别靠记忆。",[11,205,206],{},"还有个更省事的方案：正文直接用系统字体栈，只在标题上用自定义字体。中文系统字体在 Windows 和 macOS 上都不难看，这一刀下去能省掉大部分字体问题。",{"title":101,"searchDepth":102,"depth":102,"links":208},[209,210,211,212,213],{"id":139,"depth":102,"text":139},{"id":148,"depth":102,"text":148},{"id":173,"depth":102,"text":173},{"id":182,"depth":102,"text":182},{"id":200,"depth":102,"text":200},"2026-02-25","子集按生成那天的文案切，三个月后新文案缺 17 个字，前端只能回退全量包，首屏字体从 280KB 涨回 2.3MB。把字符集变成构建产物，再加一条缺字监控。",{},"\u002Fblog\u002Ffont-subset-maintenance",{"title":131,"description":215},"blog\u002Ffont-subset-maintenance",[221,222,223],"字体加载","子集化","构建流程","iNigoUEjsCcEFimUiLFJgR_nMYydx8AAclNtXBeCrt8",{"id":226,"title":227,"author":6,"body":228,"canonical":113,"category":383,"cover":113,"date":384,"description":385,"draft":117,"extension":118,"keywords":113,"meta":386,"navigation":120,"noindex":117,"ogImage":113,"path":387,"seo":388,"seoDescription":113,"seoTitle":113,"stem":389,"tags":390,"updated":113,"__hash__":394},"blog\u002Fblog\u002Ffollowup-question-chain.md","AI 搜索的第二问最容易丢客户：我们跑了 60 条追问链",{"type":8,"value":229,"toc":375},[230,233,236,239,242,245,249,252,255,258,314,317,320,323,326,329,332,335,338,345,351,357,360,363,366,369,372],[11,231,232],{},"第一问被引用，不代表第二问还在。2026 年 3 月我们跑了 60 条追问链，同一批查询在 4 个引擎里连问三轮。你的出现率从第一问的 34% 掉到第二问 19%，第三问 8%。掉的不是流量，是用户的决策过程。",[15,234,235],{"id":235},"追问链是怎么形成的",[11,237,238],{},"AI 搜索有个习惯：回答完会建议下一个问题。\"你可能还想了解\"\"接下来可以问\"。",[11,240,241],{},"用户真会点。我们的观察里，约三分之一的人在第一问之后会追问一次，其中一部分会追到第三问。",[11,243,244],{},"这意味着用户在 AI 里完成了一次完整的选型调研：从\"什么是数据管道\"，到\"我们这种规模适合哪种方案\"，再到\"这个方案迁移要多久\"。你只在第一步出现，后面两步就是竞品的场子。",[15,246,248],{"id":247},"_60-条链怎么测的","60 条链怎么测的",[11,250,251],{},"每个引擎跑 20 条，共 60 条链。每条链固定三步：问第一句 → 记录回答和建议追问 → 点第一个建议 → 记录 → 再点。",[11,253,254],{},"记录三件事：出现没出现、出现方式、被引来源。",[11,256,257],{},"三轮下来共 180 次问答。统计结果：",[259,260,261,277],"table",{},[262,263,264],"thead",{},[265,266,267,271,274],"tr",{},[268,269,270],"th",{},"轮次",[268,272,273],{},"出现率",[268,275,276],{},"被引来源里官网占比",[278,279,280,292,303],"tbody",{},[265,281,282,286,289],{},[283,284,285],"td",{},"第一问",[283,287,288],{},"34%",[283,290,291],{},"38%",[265,293,294,297,300],{},[283,295,296],{},"第二问",[283,298,299],{},"19%",[283,301,302],{},"22%",[265,304,305,308,311],{},[283,306,307],{},"第三问",[283,309,310],{},"8%",[283,312,313],{},"11%",[15,315,316],{"id":316},"为什么第二问就丢",[11,318,319],{},"因为第二问更具体。第一问通常是品类词，你的概念页能答；第二问开始带约束——规模、预算、技术栈、部署方式。你的内容如果停在概念层，就接不住。",[11,321,322],{},"我们翻了一遍丢掉的第二问，高频形态是这三种：",[11,324,325],{},"\"我们只有 3 个开发，这套方案维护成本高吗\"——需要按团队规模给判断，你的页面只讲了架构。",[11,327,328],{},"\"能不能迁到国产数据库\"——需要兼容性清单，你的页面只有一句\"支持主流数据库\"。",[11,330,331],{},"\"和 XX 比哪个更适合我们\"——需要对比，你只有自己的功能页。",[15,333,334],{"id":334},"怎么接住",[11,336,337],{},"三个动作，按成本排。",[11,339,340,344],{},[341,342,343],"strong",{},"一、每个概念页末尾加\"下一步\"小节。"," 列 3 个自然的下一个问题，每个问题下面给 80 字答案，指向对应页面。这既是给人看的导航，也是给引擎的第二问素材。",[11,346,347,350],{},[341,348,349],{},"二、把约束条件写成独立小节。"," 规模、预算、技术栈、部署方式，各写一段，每段回答\"这种条件下会怎样\"。别把这些话埋在正文里。",[11,352,353,356],{},[341,354,355],{},"三、FAQ 从\"常见问题\"改成\"追问问题\"。"," 别写\"什么是 XX\"，写\"XX 在 200 人团队里怎么配\"。前者是搜索时代的问法，后者是对话时代的问法。",[15,358,359],{"id":359},"一个客户的变化",[11,361,362],{},"那个客户 2 月开始改，先改 12 个核心页面。3 月底复查：第二问出现率从 17% 到 21%，第三问从 6% 到 9%。",[11,364,365],{},"涨得不算猛，但有一件事变了：第三问里出现他们的时候，来源基本都是官网的对比页和文档页，不再是第三方。这说明用户在决策末端拿到的信息是你给的，而不是别人转述的。",[15,367,368],{"id":368},"一句提醒",[11,370,371],{},"别为了追第三问去堆内容。我们试过给一个客户写 40 篇\"下一步\"文章，第二问出现率确实涨了，但询盘没变——因为内容太碎，用户看完不知道你是谁。",[11,373,374],{},"第二问要接，但接的方式是让已有页面能承压，不是再造一批新页面。",{"title":101,"searchDepth":102,"depth":102,"links":376},[377,378,379,380,381,382],{"id":235,"depth":102,"text":235},{"id":247,"depth":102,"text":248},{"id":316,"depth":102,"text":316},{"id":334,"depth":102,"text":334},{"id":359,"depth":102,"text":359},{"id":368,"depth":102,"text":368},"GEO","2026-02-27","第一问被引用不代表第二问还在。60 条追问链实测，出现率从 34% 掉到 19%，第三问只剩 8%。丢的不是流量，是用户的决策过程。附三个接住追问的改法。",{},"\u002Fblog\u002Ffollowup-question-chain",{"title":227,"description":385},"blog\u002Ffollowup-question-chain",[383,391,392,393],"追问链","AI搜索","用户意图","2U83dpAFG3Eyv7n45gfPG1OSn00XcIDysp011ccQV14",1794088108058]