[{"data":1,"prerenderedAt":425},["ShallowReactive",2],{"blog-first-party-data-article":3,"blog-prev-first-party-data-article":115,"blog-next-first-party-data-article":200},{"id":4,"title":5,"author":6,"body":7,"canonical":99,"category":100,"cover":99,"date":101,"description":102,"draft":103,"extension":104,"keywords":99,"meta":105,"navigation":106,"noindex":103,"ogImage":99,"path":107,"seo":108,"seoDescription":99,"seoTitle":99,"stem":109,"tags":110,"updated":99,"__hash__":114},"blog\u002Fblog\u002Ffirst-party-data-article.md","用自家数据写文章，取数、脱敏和口径怎么处理","元析智能",{"type":8,"value":9,"toc":87},"minimark",[10,14,18,21,24,27,30,33,36,39,42,45,48,51,54,57,60,63,66,69,72,75,78,81,84],[11,12,13],"p",{},"第三方报告谁都能引用，自家后台的数据只有你有。我们每年用客户站的数据写 8 到 10 篇，这类文章的平均阅读完成率比普通文章高 30% 左右，外链也更容易拿到。",[15,16,17],"h2",{"id":17},"数据从哪来",[11,19,20],{},"搜索后台的曝光和查询、站内搜索的词、客服工单量、产品的使用日志、报价和成交流程里的耗时。这些都不需要额外采集，日常就在产生。",[11,22,23],{},"别去找那些需要埋点三个月才能拿到的指标，等到数据齐了，选题已经过时了。",[15,25,26],{"id":26},"口径必须先写清楚",[11,28,29],{},"同一个“转化率”，按访问算、按注册算、按成交算，能差三倍。文章里第一次出现必须写清口径，写在数字后面或者脚注里。",[11,31,32],{},"2025 年 10 月我们写过一篇讲内容页转化的文章，因为没写口径，被读者在评论区指出前后矛盾，最后改了两次。",[15,34,35],{"id":35},"脱敏的三种做法",[11,37,38],{},"改绝对值，只写比例。模糊时间，写“过去 12 个月”而不是具体月份。合并样本，把几个客户的数据合成一组。",[11,40,41],{},"绝对不要出现客户名、域名、可识别的业务特征。这三条踩一条，客户关系就没了。",[15,43,44],{"id":44},"样本少的时候怎么办",[11,46,47],{},"样本少于 30 个别写百分比，改成写具体案例。三个客户的经历写成三个故事，比“33% 的客户”更可信。",[11,49,50],{},"我们有个规矩：n 小于 10 不出百分比，n 小于 30 不出对比结论。",[15,52,53],{"id":53},"一个具体的例子",[11,55,56],{},"2026 年 2 月我们写过一篇讲长尾词优先级的文章，用了 4 个客户站、共 1260 个词的数据，说明月搜 50 以下的词里，有 37% 在前三页出现过我们的页面。",[11,58,59],{},"这个数字是从后台导的，别人引用不了，这就是独家。",[15,61,62],{"id":62},"为什么值得做",[11,64,65],{},"这类文章很难被抄。抄的人拿不到原始数据，只能转述结论，转述的东西没有可信度。",[15,67,68],{"id":68},"数据文的结构",[11,70,71],{},"开头给最反常识的那个结论，然后写数据怎么来的，再给两三个细分结论，最后写适用边界。",[11,73,74],{},"别把数据文写成报告。报告是给人做决策用的，数据文是给人判断用的，后者的结论要更锐利。",[15,76,77],{"id":77},"一个容易被忽略的步骤",[11,79,80],{},"数据跑完先让自己人看一遍，问三个问题：这个数字符合直觉吗，有没有可能算错，结论能不能被反过来解释。",[11,82,83],{},"2026 年 1 月我们有一篇差点发出去，数据没错，但结论被相反的样本推翻了，幸好发布前有人提了一句。",[11,85,86],{},"AI 引擎也更愿意引用带具体数字和口径的页面。我们那几篇数据文被引用的频率，比其他文章高。",{"title":88,"searchDepth":89,"depth":89,"links":90},"",2,[91,92,93,94,95,96,97,98],{"id":17,"depth":89,"text":17},{"id":26,"depth":89,"text":26},{"id":35,"depth":89,"text":35},{"id":44,"depth":89,"text":44},{"id":53,"depth":89,"text":53},{"id":62,"depth":89,"text":62},{"id":68,"depth":89,"text":68},{"id":77,"depth":89,"text":77},null,"内容运营","2026-04-12","第三方数据谁都能引，自家数据只有你有。我们每年用客户站的后台数据写 8 到 10 篇，取数、脱敏、口径三件事都做对，文章才会有信息增量。",false,"md",{},true,"\u002Fblog\u002Ffirst-party-data-article",{"title":5,"description":102},"blog\u002Ffirst-party-data-article",[111,112,113],"一手数据","内容写作","信息增益","4luaucfchBOqfWsfgtwq8GBbsuc90kpqBgFVkzY7hvo",{"id":116,"title":117,"author":6,"body":118,"canonical":99,"category":188,"cover":99,"date":189,"description":190,"draft":103,"extension":104,"keywords":99,"meta":191,"navigation":106,"noindex":103,"ogImage":99,"path":192,"seo":193,"seoDescription":99,"seoTitle":99,"stem":194,"tags":195,"updated":99,"__hash__":199},"blog\u002Fblog\u002Fgooglebot-vs-baiduspider.md","Googlebot 和百度蜘蛛的抓取习惯差在哪",{"type":8,"value":119,"toc":179},[120,123,127,130,133,137,140,144,147,151,154,157,160,163,167,170,173,176],[11,121,122],{},"两个蜘蛛不是同一个物种。Google 主动、快、肯执行 JS；百度保守、慢、更依赖你主动提交。拿优化 Google 的经验直接套百度，会误判得很惨。",[15,124,126],{"id":125},"差异一抓取频次差一个数量级","差异一：抓取频次差一个数量级",[11,128,129],{},"5 个自有客户站的日志里，Googlebot 的日请求量平均是 Baiduspider 的 6 到 15 倍。新站差距更大，Google 上线第二周就可能来抓，百度可能一个月才零星来几次。所以百度的收录慢，很多时候不是你的问题，是它的节奏。",[11,131,132],{},"我们给客户做季度复盘时，会把两个引擎的抓取量、收录量、点击量做成三条曲线，任何一条背离另外两条，就开始查原因。这个动作做上几个季度，对两个蜘蛛的脾气就有数了。",[15,134,136],{"id":135},"差异二抓取路径不一样","差异二：抓取路径不一样",[11,138,139],{},"Google 会顺着内链往深处爬，深度 5 层的页面也常被抓。百度在新站阶段更集中——首页、sitemap 里的 URL、你主动提交的 URL，深度超过 3 层、又没被提交过的页面，可能一次都不来。",[15,141,143],{"id":142},"差异三渲染能力不同","差异三：渲染能力不同",[11,145,146],{},"Google 用接近最新版 Chrome 的环境渲染，对 SPA、Web Component 的支持比较完整。百度有自己的渲染流程，复杂前端耗时更长、失败率更高，所以对百度来说，服务端渲染和静态 HTML 的价值更高。",[15,148,150],{"id":149},"差异四对新站的耐心不同","差异四：对新站的耐心不同",[11,152,153],{},"Google 对新站有观察期，但通常几周内就会开始抓核心页。百度对新站的观察更久，尤其是没有任何外链、没有备案信息、内容量又少的站，两个月不来抓都算正常。这时候与其等，不如把提交、内链、服务器响应这三件能控的事做到位。",[15,155,156],{"id":156},"一个站该怎么分别对待",[11,158,159],{},"对 Google：把内链结构做深、渲染质量做好，剩下的交给它自己爬。",[11,161,162],{},"对百度：强化主动提交。普通收录 API 是最直接的入口，sitemap 其次；服务器响应时间要压住，百度对 TTFB 比 Google 敏感；正文尽量出现在初始 HTML 里。",[15,164,166],{"id":165},"_2025-年-11-月那个反差站","2025 年 11 月那个反差站",[11,168,169],{},"客户站 Google 收录 92%，百度收录只有 31%。查日志：百度蜘蛛 30 天的请求集中在 200 个 URL 上，深度超过 3 层的页面一次没进。补了内链、用 API 提交了 1800 个 URL、把服务器响应从 1.4 秒压到 0.4 秒，三个月后百度收录到 68%。",[11,171,172],{},"同期 Google 的收录和排名几乎没变，说明这套改动对 Google 来说是锦上添花，对百度是雪中送炭。",[15,174,175],{"id":175},"立场",[11,177,178],{},"别用一个指标看两个引擎。百度收录低不代表站的技术水平差，Google 收录高也不代表百度就该跟上。分开看日志、分开定优化动作，才不会被平均数骗。",{"title":88,"searchDepth":89,"depth":89,"links":180},[181,182,183,184,185,186,187],{"id":125,"depth":89,"text":126},{"id":135,"depth":89,"text":136},{"id":142,"depth":89,"text":143},{"id":149,"depth":89,"text":150},{"id":156,"depth":89,"text":156},{"id":165,"depth":89,"text":166},{"id":175,"depth":89,"text":175},"技术SEO","2026-04-11","同一个站，Googlebot 一天来 1.2 万次，百度蜘蛛来 800 次，这不是资源分配的问题。我们从 5 个站的日志里对比出三个行为差异，以及一个站该怎么分别对待。",{},"\u002Fblog\u002Fgooglebot-vs-baiduspider",{"title":117,"description":190},"blog\u002Fgooglebot-vs-baiduspider",[196,197,198],"爬虫","日志分析","百度","-zTdZ6mNErozQIa6tz30ul4dvcmD9B_rdXTwt1HypdQ",{"id":201,"title":202,"author":6,"body":203,"canonical":99,"category":412,"cover":99,"date":413,"description":414,"draft":103,"extension":104,"keywords":99,"meta":415,"navigation":106,"noindex":103,"ogImage":99,"path":416,"seo":417,"seoDescription":99,"seoTitle":99,"stem":418,"tags":419,"updated":99,"__hash__":424},"blog\u002Fblog\u002Fai-visibility-three-levels.md","AI 可见性分三层：被读到、被提到、被推荐，三种算法不同",{"type":8,"value":204,"toc":403},[205,208,212,215,218,221,224,228,231,234,237,241,244,247,250,253,256,312,315,318,321,379,382,385,388,391,394,397,400],[11,206,207],{},"AI 可见性不是一件事，是三件事。被读到、被提到、被推荐，三层的数据来源完全不同，混在一起算会得出自相矛盾的结论。上个月我们给一个客户做完整测量：读到 72%，提到 34%，推荐 9%。",[15,209,211],{"id":210},"第一层被读到","第一层：被读到",[11,213,214],{},"指标是 AI 爬虫有没有抓你的页面。数据来源是服务器日志，看 GPTBot、OAI-SearchBot、PerplexityBot 的访问记录和路径。",[11,216,217],{},"这层是基础，但容易被跳过。没被抓到，后面两层都是零。",[11,219,220],{},"判断方法：拉一个月日志，看目标页面有没有被访问、被访问几次、被抓的是不是最新版本。",[11,222,223],{},"一个常见问题是抓的是旧版本。CDN 缓存放了 30 天，爬虫拿到的还是改版前的页面。",[15,225,227],{"id":226},"第二层被提到","第二层：被提到",[11,229,230],{},"指标是 AI 回答里有没有出现你的品牌名或产品名。数据来源是人工跑查询、抄回答原文。",[11,232,233],{},"这层可以无链接。很多文章只算\"挂了链接\"的引用，漏掉了大量\"提了名字但没给链接\"的情况。后者对品牌认知一样有用。",[11,235,236],{},"这层的测量要点是固定查询池。不固定就没法环比，见我们另一篇讲分母的文章。",[15,238,240],{"id":239},"第三层被推荐","第三层：被推荐",[11,242,243],{},"指标是你在回答里处于什么位置。三种形态：首位推荐、列举中的一项、备选位置。",[11,245,246],{},"数据来源还是回答原文，但要读结构。首位通常是一句判断句，\"如果只能选一个，我建议 X\"。",[11,248,249],{},"这层最接近生意。我们统计过，首位推荐的询盘率是备选位置的 5 倍以上。",[15,251,252],{"id":252},"三层的转化关系",[11,254,255],{},"那个客户的数：",[257,258,259,275],"table",{},[260,261,262],"thead",{},[263,264,265,269,272],"tr",{},[266,267,268],"th",{},"层级",[266,270,271],{},"数值",[266,273,274],{},"说明",[276,277,278,290,301],"tbody",{},[263,279,280,284,287],{},[281,282,283],"td",{},"读到",[281,285,286],{},"72%",[281,288,289],{},"目标页 25 个，18 个被 AI 爬虫抓过",[263,291,292,295,298],{},[281,293,294],{},"提到",[281,296,297],{},"34%",[281,299,300],{},"60 个查询里 20 条回答出现品牌名",[263,302,303,306,309],{},[281,304,305],{},"推荐",[281,307,308],{},"9%",[281,310,311],{},"其中 5 条在首位或主答案",[11,313,314],{},"读到 72% 说明抓取没问题，卡在第二层：抓到了但没用上，内容可引用性不足。这种情况改内容结构，不用管技术。",[11,316,317],{},"如果读到只有 20%，那要先修技术：内链、sitemap、robots、CDN。",[15,319,320],{"id":320},"症状对应修法",[257,322,323,336],{},[260,324,325],{},[263,326,327,330,333],{},[266,328,329],{},"症状",[266,331,332],{},"卡在哪层",[266,334,335],{},"先修什么",[276,337,338,348,358,368],{},[263,339,340,343,345],{},[281,341,342],{},"日志里没爬虫",[281,344,283],{},[281,346,347],{},"抓取通道",[263,349,350,353,355],{},[281,351,352],{},"有抓取没提到",[281,354,294],{},[281,356,357],{},"内容块和事实句",[263,359,360,363,365],{},[281,361,362],{},"提到但不靠前",[281,364,305],{},[281,366,367],{},"对比页和场景结论",[263,369,370,373,376],{},[281,371,372],{},"靠前但说错",[281,374,375],{},"提到，准确性",[281,377,378],{},"语料纠错",[11,380,381],{},"这张表贴在我们每次复盘的墙上。看到数字先定位层，再谈动作。",[15,383,384],{"id":384},"一个提醒",[11,386,387],{},"别拿第三层的数字去汇报，然后被问\"为什么流量没涨\"。推荐位带来的是认知和信任，直接点击很少。要证明价值，得看你自己的注册表单和销售记录，不能只看网站分析。",[15,389,390],{"id":390},"三层怎么排采集频率",[11,392,393],{},"日志每天自动跑，不占人力。提到层每周跑一次 60 个查询，一个人两小时。推荐层每月跑一次，因为位置变化慢，周频看不出趋势。",[11,395,396],{},"推荐层要人读原文，这是唯一不能自动化的部分。\"它把我们放在第几个\"这件事，机器判不准，因为它不理解那句话的语气。",[11,398,399],{},"我们试过用脚本抓品牌名出现的位置序号，跟人工判断对比，一致率 68%。剩下的 32% 是\"提到了但其实是贬义\"或者\"列举里排第三但实际是推荐\"。这种活儿还是得人看。",[11,401,402],{},"三层数据放一张表里，但结论要分层下。哪层掉了查哪层，别一看到总数下降就全站改一遍。",{"title":88,"searchDepth":89,"depth":89,"links":404},[405,406,407,408,409,410,411],{"id":210,"depth":89,"text":211},{"id":226,"depth":89,"text":227},{"id":239,"depth":89,"text":240},{"id":252,"depth":89,"text":252},{"id":320,"depth":89,"text":320},{"id":384,"depth":89,"text":384},{"id":390,"depth":89,"text":390},"工具评测","2026-04-13","被读到看日志，被提到看回答原文，被推荐看位置和链接，三层的数据来源完全不同。一个客户读到 72%、提到 34%、推荐只有 9%，卡在哪一层就修哪一层，别混着算。",{},"\u002Fblog\u002Fai-visibility-three-levels",{"title":202,"description":414},"blog\u002Fai-visibility-three-levels",[420,421,422,423],"AI可见性","监测方法","GEO","指标","ngcZ9a0Vk1IjGHGoyYEcBK8bTH2-ryaF8oxpuEDSZyk",1794088106825]