[{"data":1,"prerenderedAt":429},["ShallowReactive",2],{"blog-intent-mismatch-fix":3,"blog-prev-intent-mismatch-fix":116,"blog-next-intent-mismatch-fix":225},{"id":4,"title":5,"author":6,"body":7,"canonical":100,"category":101,"cover":100,"date":102,"description":103,"draft":104,"extension":105,"keywords":100,"meta":106,"navigation":107,"noindex":104,"ogImage":100,"path":108,"seo":109,"seoDescription":100,"seoTitle":100,"stem":110,"tags":111,"updated":100,"__hash__":115},"blog\u002Fblog\u002Fintent-mismatch-fix.md","意图错配让一个站两个月掉 34% 流量，我们这样修的","元析智能",{"type":8,"value":9,"toc":88},"minimark",[10,14,18,21,24,27,30,33,36,39,42,45,48,51,54,57,60,63,66,70,73,76,79,82,85],[11,12,13],"p",{},"2026 年 1 月，一个客户找到我们，自然流量两个月掉了 34%。技术侧没问题，外链没少，文章也没删。查下来是 9 个页面承接了不属于它们的搜索意图。拆开之后六周，流量回到原来的 96%。",[15,16,17],"h2",{"id":17},"症状是什么样的",[11,19,20],{},"有曝光，没点击。或者点击了，十秒就跳出。搜索后台里，这些页面的曝光在涨，点击率却从 2.1% 掉到 0.8%。",[11,22,23],{},"排名也没有完全掉，很多词还在 8 到 15 名之间。这就是意图错配的典型信号：排得上，但用户不认。",[15,25,26],{"id":26},"怎么找出来",[11,28,29],{},"把查询级数据导出来，看每个页面承接了多少个不同意图的查询。一个页面同时接“是什么”“怎么选”“报错”三类词，基本就是错配。",[11,31,32],{},"我们那次找出 9 个页面，其中一个页面承接了 47 个查询，横跨三类意图。",[15,34,35],{"id":35},"为什么会变成这样",[11,37,38],{},"大多数是历史原因。页面一开始写的是学习类内容，后来运营为了省事，把对比信息和报错 FAQ 都堆进同一个页面，标题改过三次，正文越加越长。",[11,40,41],{},"页面变成了万能页，结果哪一类用户都不满意。",[15,43,44],{"id":44},"怎么修",[11,46,47],{},"能拆就拆。原来的页面留给最大的那类意图，另外两类各开一个新页面，内容重写，内链互相挂。",[11,49,50],{},"拆不了的，改页面类型。把博客文章改成文档页或对比页，标题、结构、CTA 全部重做。",[15,52,53],{"id":53},"修完的节奏",[11,55,56],{},"先改 3 个页面，观察四周，再改剩下 6 个。一次全改，出问题查不出原因。",[11,58,59],{},"那次我们第 2 周看到点击率回升，第 4 周排名稳住，第 6 周流量回到 96%。剩下的 4% 是季节波动，不是这次修改的问题。",[15,61,62],{"id":62},"怎么预防",[11,64,65],{},"词表里每个词标意图，页面立项时确认只接一种意图。这个规矩我们在 2025 年就写过，客户没执行，于是有了这次事故。",[15,67,69],{"id":68},"另一种错配页面没错标题错了","另一种错配：页面没错，标题错了",[11,71,72],{},"有些页面内容是对的，标题承诺了另一类意图。搜“怎么选”的人点进来，看到的是功能说明，一样会走。",[11,74,75],{},"这种情况改标题和首段就行，不用重写。我们那次 9 个页面里，有 3 个属于这一类，改完一周就看到点击率变化。",[15,77,78],{"id":78},"怎么防止再犯",[11,80,81],{},"新页面立项时写两行：目标词、目标意图。发布前由第二个人核对一遍，确认页面类型和意图匹配。",[11,83,84],{},"这个检查两分钟，能省掉后面几个月的返工。",[11,86,87],{},"意图错配不会立刻爆，它慢慢积累，等到某天算法调整或者竞品发力，才一次性显出来。",{"title":89,"searchDepth":90,"depth":90,"links":91},"",2,[92,93,94,95,96,97,98,99],{"id":17,"depth":90,"text":17},{"id":26,"depth":90,"text":26},{"id":35,"depth":90,"text":35},{"id":44,"depth":90,"text":44},{"id":53,"depth":90,"text":53},{"id":62,"depth":90,"text":62},{"id":68,"depth":90,"text":69},{"id":78,"depth":90,"text":78},null,"SEO","2026-03-22","2026 年 1 月一个客户自然流量两个月掉 34%，技术没动、外链没少、文章也没删。问题出在 9 个页面承接了不属于它的意图，拆开之后六周恢复。",false,"md",{},true,"\u002Fblog\u002Fintent-mismatch-fix",{"title":5,"description":103},"blog\u002Fintent-mismatch-fix",[112,113,114],"搜索意图","流量下滑","复盘","UXVNi3Dk2O6JLxPKOT6zu6N1UmfpbmOw8nkSrWtPXNo",{"id":117,"title":118,"author":6,"body":119,"canonical":100,"category":213,"cover":100,"date":214,"description":215,"draft":104,"extension":105,"keywords":100,"meta":216,"navigation":107,"noindex":104,"ogImage":100,"path":217,"seo":218,"seoDescription":100,"seoTitle":100,"stem":219,"tags":220,"updated":100,"__hash__":224},"blog\u002Fblog\u002Ffont-weight-bloat.md","一个官网用了 11 个字重，字体治理比压缩更省",{"type":8,"value":120,"toc":205},[121,124,127,130,142,145,148,151,154,157,160,163,179,182,185,188,191,198],[11,122,123],{},"一个官网用了 11 个字重，首屏要下 5 个字体文件。这不是夸张，是今年 1 月我们在一个开发者工具站上看到的。删到 3 个字重之后，首屏字体请求从 5 个降到 1 个，体积从 1.1MB 降到 310KB。",[15,125,126],{"id":126},"字体是怎么膨胀的",[11,128,129],{},"设计师给了 4 个字体家族，每个 3 到 4 个字重，前端按设计稿逐个引入，没人算过总量。页面里真正用到的只有 3 个字重，剩下的都是复制粘贴留下的。",[11,131,132,133,137,138,141],{},"查法：跑一遍页面，用 DevTools 的 Coverage 面板看字体文件使用率，再对照 CSS 里所有 ",[134,135,136],"code",{},"font-family"," 和 ",[134,139,140],{},"font-weight"," 声明。这个站有 40 多处声明，实际组合只有 9 种。",[15,143,144],{"id":144},"治理比压缩更省",[11,146,147],{},"很多团队一上来就想压缩字体，其实先做减法收益更大。这个站把 11 个字重砍到 3 个：正文 400、标题 600、强调 700。视觉上几乎没差别，因为设计稿本来只用这三档。",[11,149,150],{},"顺序是：先砍字重，再砍字族，最后才子集化。反过来做，等于给用不上的字重做子集，白费功夫。",[15,152,153],{"id":153},"可变字体值不值",[11,155,156],{},"可变字体一个文件包含多个字重，理论上更省。实测要看用几个：只用 2 到 3 个，普通字体子集更小；用 5 个以上，可变字体才有优势。",[11,158,159],{},"我们给这个站试过可变字体，单文件 340KB，比 3 个静态子集加起来还大，最后没用。这个结论有点反直觉，但数字摆在那儿。",[15,161,162],{"id":162},"四条判断清单",[164,165,166,170,173,176],"ol",{},[167,168,169],"li",{},"页面实际用到几个字重？超过 3 个就要审",[167,171,172],{},"有没有引入图标字体？能换 SVG 就换",[167,174,175],{},"中文正文用系统字体行不行？行的话最省",[167,177,178],{},"首屏字体文件总大小控制在 300KB 以内",[11,180,181],{},"按这四条走一遍，多数站的字体体积能砍掉六成。这个站改完首屏 LCP 从 2.4 秒降到 2.1 秒，幅度不算惊人，但省下的 800KB 流量对移动端用户是实打实的。",[15,183,184],{"id":184},"顺带纠正一个误区",[11,186,187],{},"字体不是\"越少越好\"到极端。有客户问能不能只用一个字重，全部靠浏览器合成粗体。合成粗体在低分屏上边缘发虚，中文尤其明显，视觉代价比省下的 60KB 大。三个字重是多数 B2B 软件站的下限。",[15,189,190],{"id":190},"上线后的复查",[11,192,193,194,197],{},"改完字体不是一劳永逸。我们给每个站加了一个每月自动跑的检查：抓首页 HTML，列出所有 ",[134,195,196],{},"@font-face"," 和实际引用，对比总数。有一次产品团队上线新页面，又带回来两个没用上的字重，被这个检查抓到了。",[11,199,200,201,204],{},"字体文件还要设长缓存，文件名带 hash，",[134,202,203],{},"Cache-Control: max-age=31536000, immutable","。少了这条，回访用户每次还要重新协商，白等一轮往返。",{"title":89,"searchDepth":90,"depth":90,"links":206},[207,208,209,210,211,212],{"id":126,"depth":90,"text":126},{"id":144,"depth":90,"text":144},{"id":153,"depth":90,"text":153},{"id":162,"depth":90,"text":162},{"id":184,"depth":90,"text":184},{"id":190,"depth":90,"text":190},"技术SEO","2026-03-21","开发者工具站首屏要下 5 个字体文件，总体积 1.1MB。砍到 3 个字重加子集化之后，请求降到 1 个、体积 310KB。附四条判断清单，以及可变字体值不值的实测。",{},"\u002Fblog\u002Ffont-weight-bloat",{"title":118,"description":215},"blog\u002Ffont-weight-bloat",[221,222,223],"字体加载","字体治理","首屏优化","VwHYGFy9BxF9eKMtQHCWwlHAZuL3PZLuZnAPQ1w_2go",{"id":226,"title":227,"author":6,"body":228,"canonical":100,"category":416,"cover":100,"date":417,"description":418,"draft":104,"extension":105,"keywords":100,"meta":419,"navigation":107,"noindex":104,"ogImage":100,"path":420,"seo":421,"seoDescription":100,"seoTitle":100,"stem":422,"tags":423,"updated":100,"__hash__":428},"blog\u002Fblog\u002Fdomestic-ai-citation-sources.md","豆包、元宝、文心更爱引平台内容：一次 300 条引用的统计",{"type":8,"value":229,"toc":409},[230,233,236,239,242,342,345,348,355,361,367,370,373,376,379,382,385,388,391,394,397,400,403,406],[11,231,232],{},"国内 AI 搜索引用官网的比例，比海外低得多。2026 年 3 月到 4 月，我们统计了豆包、元宝、文心三个引擎的 300 条引用，官网占 21%，剩余 79% 全在平台上。",[15,234,235],{"id":235},"统计怎么做的",[11,237,238],{},"三个引擎各跑 100 条查询，覆盖 B2B 软件方向的品类词、对比词、场景词。记录每条回答挂出来的来源，按域名分类，同一条回答里同一域名只算一次。",[11,240,241],{},"300 条引用，去重后 217 个来源域名。",[243,244,245,261],"table",{},[246,247,248],"thead",{},[249,250,251,255,258],"tr",{},[252,253,254],"th",{},"来源类型",[252,256,257],{},"占比",[252,259,260],{},"典型站点",[262,263,264,276,287,298,309,320,331],"tbody",{},[249,265,266,270,273],{},[267,268,269],"td",{},"官网",[267,271,272],{},"21%",[267,274,275],{},"客户自己的站",[249,277,278,281,284],{},[267,279,280],{},"问答",[267,282,283],{},"18%",[267,285,286],{},"知乎",[249,288,289,292,295],{},[267,290,291],{},"技术社区",[267,293,294],{},"14%",[267,296,297],{},"CSDN、掘金",[249,299,300,303,306],{},[267,301,302],{},"公众号转载",[267,304,305],{},"12%",[267,307,308],{},"各类转载站",[249,310,311,314,317],{},[267,312,313],{},"门户与媒体",[267,315,316],{},"11%",[267,318,319],{},"36 氪、少数派",[249,321,322,325,328],{},[267,323,324],{},"百科",[267,326,327],{},"8%",[267,329,330],{},"百度百科",[249,332,333,336,339],{},[267,334,335],{},"其他",[267,337,338],{},"16%",[267,340,341],{},"论坛、文档站、GitHub",[15,343,344],{"id":344},"为什么官网占比低",[11,346,347],{},"三个原因。",[11,349,350,354],{},[351,352,353],"strong",{},"平台有内容池。"," 国内引擎跟内容平台有合作，抓取通道和权重都不一样。同样一篇文章，发在平台上比发在自己站上更容易被读到。",[11,356,357,360],{},[351,358,359],{},"中文语料密度。"," 一个 B2B 软件官网通常几十个页面，知乎上关于这个品类的讨论可能是几千条。引擎在语料里找答案，谁多找谁。",[11,362,363,366],{},[351,364,365],{},"官网写得不像答案。"," 我们看的 100 个官网里，有 70 多个的页面结构是\"产品价值 + 功能列表 + 客户 logo\"。这种页面没有能独立回答问题的段落，引擎抓了也没法用。",[15,368,369],{"id":369},"公众号那块要注意",[11,371,372],{},"公众号占 12%，但要看清楚：引用的多是转载到可抓取页面的版本，公众号原文本身在很多引擎里抓不到。",[11,374,375],{},"这意味着两件事。一是发公众号的时候，同步发一份到官网或知乎，让内容有可抓的入口。二是看数据时别把\"公众号发了\"当成\"被收录了\"。",[11,377,378],{},"我们有个客户每月发 8 篇公众号，发了两年，国内 AI 回答里一次没出现过。查下来是内容从没在可抓页面出现过。",[15,380,381],{"id":381},"对软件公司的含义",[11,383,384],{},"官网是事实源，平台是入口。这个分工要想清楚。",[11,386,387],{},"官网负责\"准确\"：产品参数、定价、部署方式、公司信息，写清楚，让引擎有据可查。",[11,389,390],{},"平台负责\"被读到\"：知乎回答、技术社区文章、行业媒体稿，这些决定引擎会不会在回答里提你。",[11,392,393],{},"只做官网的客户，我们说句实话：国内这三个引擎里，你的可见度会一直很低，跟你内容质量没关系。",[15,395,396],{"id":396},"一个可执行的分配",[11,398,399],{},"按我们的经验，国内 GEO 的力气这么分：",[11,401,402],{},"官网占 40%，做事实和结构。平台占 50%，做分发和场景。剩下 10% 做监测。",[11,404,405],{},"不是五五开，是官网四、平台五。因为平台内容还承担了\"被别人引用\"的功能，一篇知乎回答可能同时被三个引擎读。",[11,407,408],{},"这个比例在海外反过来，官网大概占六。做国内和做出海，不能用同一套排期。",{"title":89,"searchDepth":90,"depth":90,"links":410},[411,412,413,414,415],{"id":235,"depth":90,"text":235},{"id":344,"depth":90,"text":344},{"id":369,"depth":90,"text":369},{"id":381,"depth":90,"text":381},{"id":396,"depth":90,"text":396},"行业观察","2026-03-23","统计国内三个引擎 300 条引用，官网只占 21%，知乎 18%、CSDN 14%、公众号 12%。国内做 GEO，官网是事实源，平台才是入口。",{},"\u002Fblog\u002Fdomestic-ai-citation-sources",{"title":227,"description":418},"blog\u002Fdomestic-ai-citation-sources",[424,425,426,427],"国内AI搜索","豆包","元宝","平台内容","4xW_Ff2rK4r-EDskmVvW_gI45YGugT-ojM9hnwmABQ0",1794088107413]