[{"data":1,"prerenderedAt":444},["ShallowReactive",2],{"blog-delete-post-301-or-410":3,"blog-prev-delete-post-301-or-410":125,"blog-next-delete-post-301-or-410":217},{"id":4,"title":5,"author":6,"body":7,"canonical":109,"category":110,"cover":109,"date":111,"description":112,"draft":113,"extension":114,"keywords":109,"meta":115,"navigation":116,"noindex":113,"ogImage":109,"path":117,"seo":118,"seoDescription":109,"seoTitle":109,"stem":119,"tags":120,"updated":109,"__hash__":124},"blog\u002Fblog\u002Fdelete-post-301-or-410.md","删掉的文章该 301 还是 410，我们的判断只有三条","元析智能",{"type":8,"value":9,"toc":97},"minimark",[10,14,19,22,25,29,32,35,39,42,45,48,51,54,57,60,63,66,69,72,76,79,82,85,88,91,94],[11,12,13],"p",{},"删文章的时候，301 和 410 的选择比大多数人想的更有讲究。我们的判断只有三条：这个页面上有没有外链、有没有排名、有没有历史流量。三条都是否，才走 410，其余一律 301。",[15,16,18],"h2",{"id":17},"_410-适合什么页面","410 适合什么页面",[11,20,21],{},"从来没被收录、从来没曝光、没有任何外链、内容本身没有价值的页面。测试页、空标签页、批量生成的垃圾页，这类走 410。",[11,23,24],{},"410 的意思是“永久删除，别再来了”。它比 404 更明确，爬虫放弃得更快。",[15,26,28],{"id":27},"_301-适合什么页面","301 适合什么页面",[11,30,31],{},"有外链的，有排名的，有过流量的，被外部引用过的。哪怕现在没流量，只要有一个外链，就 301。",[11,33,34],{},"2025 年那次审计，88 篇里只有 12 篇走了 410，剩下 76 篇都 301 到最相关的一篇。判断标准就是这个。",[15,36,38],{"id":37},"_301-指向哪","301 指向哪",[11,40,41],{},"指向内容最接近的那篇，不是首页，不是分类页。",[11,43,44],{},"实在找不到相关页面的，指向同主题的支柱页。全站指向首页的做法，等于把一堆不相干的权重混在一起，搜索引擎会给这种改法打折。",[15,46,47],{"id":47},"别一次性批量删",[11,49,50],{},"我们按批次删，每批 20 篇以内，隔两周看一次抓取和收录。",[11,52,53],{},"一次性删 80 篇，日志里会出现一大批 404，抓取预算会被这些错误占掉。2024 年有个客户这么干过，收录恢复花了两个多月。",[15,55,56],{"id":56},"删之前做三件事",[11,58,59],{},"导出 URL 清单。查每个 URL 的外链数。在搜索后台提交移除请求，加速下架。",[11,61,62],{},"这三件事做完再动手，顺序反了容易漏掉有外链的页面。",[15,64,65],{"id":65},"一个我们踩过的坑",[11,67,68],{},"2025 年 8 月，我们把一个 410 用在了有 3 个外链的页面上。当时判断是“内容没用”，忽略了外链。",[11,70,71],{},"两个月后客户发现有个行业目录站还在链那个地址，权重全落空。后来补了 301，但延迟已经造成了。",[15,73,75],{"id":74},"_404-和-410-的实际差别","404 和 410 的实际差别",[11,77,78],{},"对搜索引擎来说，两者都表示页面不存在。410 更明确，一般下架更快；404 会被反复重试，有时候要几周才彻底移除。",[11,80,81],{},"但对用户来说，410 和 404 长得一样。所以这个选择的技术意义大于体验意义，别在这上面纠结太久。",[15,83,84],{"id":84},"一个可以偷懒的做法",[11,86,87],{},"拿不准的时候，用 301。301 至少有把权重转给别的页面的机会，410 什么都没有。",[11,89,90],{},"我们内部有个更简单的判断：如果这篇内容在退订邮件、销售资料、渠道目录里可能被提到过，就别用 410。链接一旦出现在别人的文档里，你就控制不了它什么时候生效。",[11,92,93],{},"我们给客户删页面时的默认选项就是 301，只有确认边缘价值为零，才改成 410。",[11,95,96],{},"应该是：先查外链，再判断内容价值，最后决定 301 还是 410。",{"title":98,"searchDepth":99,"depth":99,"links":100},"",2,[101,102,103,104,105,106,107,108],{"id":17,"depth":99,"text":18},{"id":27,"depth":99,"text":28},{"id":37,"depth":99,"text":38},{"id":47,"depth":99,"text":47},{"id":56,"depth":99,"text":56},{"id":65,"depth":99,"text":65},{"id":74,"depth":99,"text":75},{"id":84,"depth":99,"text":84},null,"技术SEO","2026-03-06","2025 年我们给客户删了 88 篇文章，其中只有 12 篇走 410，其余全部 301。判断依据就三条：有没有外链、有没有排名、有没有历史流量。判断错了就是白丢权重。",false,"md",{},true,"\u002Fblog\u002Fdelete-post-301-or-410",{"title":5,"description":112},"blog\u002Fdelete-post-301-or-410",[121,122,123],"删除页面","301重定向","内容审计","ZPRFj_HA8gjqu4Wl_HHu11YwU3l-mQ7paseRzo43nMs",{"id":126,"title":127,"author":6,"body":128,"canonical":109,"category":110,"cover":109,"date":206,"description":207,"draft":113,"extension":114,"keywords":109,"meta":208,"navigation":116,"noindex":113,"ogImage":109,"path":209,"seo":210,"seoDescription":109,"seoTitle":109,"stem":211,"tags":212,"updated":109,"__hash__":216},"blog\u002Fblog\u002Ffont-display-swap-vs-optional.md","font-display 选 swap 还是 optional，两周实测的数据",{"type":8,"value":129,"toc":198},[130,138,141,144,147,150,153,156,159,162,165,171,174,177,180,183,186,189,192,195],[11,131,132,133,137],{},"font-display 选 swap 还是 optional，这事没有标准答案，但可以测。今年 2 月我们给两个站各跑了两周，结论是内容站选 optional，营销站选 swap 加 ",[134,135,136],"code",{},"size-adjust","，两者对留存的差别看不出来。",[15,139,140],{"id":140},"三种取值的实际差别",[11,142,143],{},"swap：文字立刻用回退字体渲染，自定义字体到了再换。好处是内容不会空白，坏处是切换时有一次视觉跳动，中文站比英文站明显。",[11,145,146],{},"optional：字体在极短时间内可用才用，超时就用回退字体直到下次访问。好处是几乎不跳动，坏处是首次访问的用户可能一直看到系统字体。",[11,148,149],{},"block：短暂隐藏文字等字体到位。这个基本别用，首屏文字空白三秒比字体丑更糟。",[15,151,152],{"id":152},"两周实测数据",[11,154,155],{},"站 A 是文档站，正文为主。swap 的 LCP P75 是 1.6 秒，optional 是 1.5 秒，差别很小，因为 h1 用哪个字体渲染出来都是最大元素。但 CLS 差别大：swap 是 0.09，optional 是 0.01。",[11,157,158],{},"站 B 是营销首页，大标题承担品牌感受。swap 的 LCP 是 1.9 秒，optional 是 2.3 秒。原因是大标题在 swap 下会被重绘，但至少先出现在首屏；optional 下首次访问看不到自定义字体，下一次访问才生效。CLS 方面 swap 是 0.14，optional 是 0.02。",[15,160,161],{"id":161},"我们的选择",[11,163,164],{},"内容站选 optional。文档、博客、术语表以阅读为主，字体视觉一致性没那么重要，稳定更重要。",[11,166,167,168,170],{},"营销站选 swap，但必须配两件事：字体文件 preload，回退字体加 ",[134,169,136],{}," 校准。这两件做完，swap 的 CLS 从 0.14 降到 0.03 左右。站 B 改完两周，真实数据稳定在 0.04。",[15,172,173],{"id":173},"一个反例",[11,175,176],{},"我见过一个站把所有字体都设成 block，理由是\"避免闪烁\"。结果移动端首屏空白 2.8 秒，用户以为网站挂了。这种做法在 2026 年没有任何理由。",[11,178,179],{},"还有人给所有字体都 preload，一次 preload 四个字体文件，等于把首屏带宽分给四个资源，LCP 反而更差。preload 只给首屏真正用到的那一个。",[15,181,182],{"id":182},"怎么自己测",[11,184,185],{},"不用搞复杂，两个页面、各跑两周，看 GSC 的真实数据就够。重点看两个数：LCP 的 P75 和 CLS 的 P75，别只看实验室。",[11,187,188],{},"如果两周内数据波动超过 20%，说明样本不够，需要更长时间或者更大的流量。这种时候别急着下结论，先看有没有其他改动混进来了。",[15,190,191],{"id":191},"落地时的两个细节",[11,193,194],{},"一是字体文件要和 CSS 在同一层 CDN 上，别一个在源站一个在 CDN，跨层回源会把收益吃掉。我们有个客户字体放在对象存储，没开缓存头，每次访问都要回源，改了缓存策略后 P75 从 700 毫秒降到 90 毫秒。",[11,196,197],{},"二是用 web vitals 的 JS 埋点看真实用户数据，别只看实验室。实验室环境是本地缓存加高速网络，测不出用户手机上的真实情况。我们现在的做法是每周拉一次真实用户数据，超过阈值就回看那次发布改了什么。",{"title":98,"searchDepth":99,"depth":99,"links":199},[200,201,202,203,204,205],{"id":140,"depth":99,"text":140},{"id":152,"depth":99,"text":152},{"id":161,"depth":99,"text":161},{"id":173,"depth":99,"text":173},{"id":182,"depth":99,"text":182},{"id":191,"depth":99,"text":191},"2026-03-05","两个站各跑两周：文档站用 optional，LCP 1.5 秒、CLS 0.01；营销站用 swap，LCP 1.9 秒、CLS 0.14，加上 preload 和 size-adjust 后 CLS 降到 0.03。",{},"\u002Fblog\u002Ffont-display-swap-vs-optional",{"title":127,"description":207},"blog\u002Ffont-display-swap-vs-optional",[213,214,215],"font-display","CLS","字体加载","i91VTSe9rZjqFi8WpwD4uWU9JeBiHiOsWAGAcWVLP0g",{"id":218,"title":219,"author":6,"body":220,"canonical":109,"category":431,"cover":109,"date":432,"description":433,"draft":113,"extension":114,"keywords":109,"meta":434,"navigation":116,"noindex":113,"ogImage":109,"path":435,"seo":436,"seoDescription":109,"seoTitle":109,"stem":437,"tags":438,"updated":109,"__hash__":443},"blog\u002Fblog\u002Fcross-engine-citation-gap.md","同一篇文章在 6 个 AI 引擎里的引用差异：一张对照表",{"type":8,"value":221,"toc":424},[222,225,228,231,234,344,347,354,360,366,369,372,378,384,390,393,396,412,415,418,421],[11,223,224],{},"同一篇文章，在六个引擎里表现完全不同。2026 年 3 月我们拿一个客户的 20 篇内容跑 60 个查询，记录被引来源。ChatGPT 偏爱文档，Perplexity 偏爱对比表，AI Overviews 偏表格和定义段，豆包偏知乎，元宝偏公众号，DeepSeek 偏技术博客。",[15,226,227],{"id":227},"测试怎么做的",[11,229,230],{},"20 篇内容覆盖三类：8 篇文档页、6 篇对比页、6 篇博客。60 个查询，每类各 20 个。六个引擎各跑一遍，记录回答里出现的目标内容，以及挂出来的来源。",[11,232,233],{},"样本不算大，但差异明显到不需要统计检验。",[235,236,237,256],"table",{},[238,239,240],"thead",{},[241,242,243,247,250,253],"tr",{},[244,245,246],"th",{},"引擎",[244,248,249],{},"最常引的来源",[244,251,252],{},"官网占比",[244,254,255],{},"特点",[257,258,259,274,288,302,316,330],"tbody",{},[241,260,261,265,268,271],{},[262,263,264],"td",{},"ChatGPT",[262,266,267],{},"文档页、报错页",[262,269,270],{},"41%",[262,272,273],{},"按句挂源，读得深",[241,275,276,279,282,285],{},[262,277,278],{},"Perplexity",[262,280,281],{},"对比页、第三方评测",[262,283,284],{},"34%",[262,286,287],{},"多源交叉，来源多",[241,289,290,293,296,299],{},[262,291,292],{},"AI Overviews",[262,294,295],{},"表格、定义段",[262,297,298],{},"46%",[262,300,301],{},"只挂少数链接",[241,303,304,307,310,313],{},[262,305,306],{},"豆包",[262,308,309],{},"知乎、CSDN",[262,311,312],{},"23%",[262,314,315],{},"中文平台权重高",[241,317,318,321,324,327],{},[262,319,320],{},"元宝",[262,322,323],{},"公众号、门户",[262,325,326],{},"19%",[262,328,329],{},"平台内容池",[241,331,332,335,338,341],{},[262,333,334],{},"DeepSeek",[262,336,337],{},"技术博客、GitHub",[262,339,340],{},"27%",[262,342,343],{},"技术细节偏好强",[15,345,346],{"id":346},"三个反常识",[11,348,349,353],{},[350,351,352],"strong",{},"一、官网占比高的引擎，不一定是好消息。"," AI Overviews 官网占比 46%，但它挂的链接最少，一条回答平均 2.1 个。占比高是因为它几乎不引第三方，不是因为它特别喜欢你。",[11,355,356,359],{},[350,357,358],{},"二、Perplexity 的来源数量在下降。"," 我们 2025 年底测的时候平均 6.4 个，2026 年 3 月是 5.2 个。它开始收敛，意味着单条来源的权重变高，进去更难，进去之后更值钱。",[11,361,362,365],{},[350,363,364],{},"三、同一引擎中英文表现能差一倍。"," ChatGPT 英文查询引官网的比例 48%，中文查询 31%。中文查询它更多去找中文平台，因为你的中文内容密度不够。",[15,367,368],{"id":368},"怎么用这个差异",[11,370,371],{},"别写六版内容，那样成本撑不住。三种做法：",[11,373,374,377],{},[350,375,376],{},"底层事实句通用。"," 带数字、带条件、带时间的那句话，在所有引擎里都好使。这部分写一份，全平台复用。",[11,379,380,383],{},[350,381,382],{},"形态按引擎调。"," 同一批事实，在官网做成表格，在知乎写成长回答，在公众号写成场景故事,在 GitHub 写进 README。不是复制粘贴，是同一事实的四种容器。",[11,385,386,389],{},[350,387,388],{},"平台投放看引擎。"," 如果你的客户主要在中文引擎里做调研，就把资源压到知乎和公众号；如果客户是海外技术团队，官网文档和 GitHub 优先。",[15,391,392],{"id":392},"一个具体的分配",[11,394,395],{},"我们给一个客户排 2026 年 Q2 的内容，是这么分的：",[397,398,399,403,406,409],"ul",{},[400,401,402],"li",{},"官网：12 篇文档页改造 + 4 篇对比页",[400,404,405],{},"知乎：8 个高赞问题下写回答，每篇带一个可引用表格",[400,407,408],{},"公众号：6 篇场景文，讲客户现场",[400,410,411],{},"GitHub：README 重写 + 4 个常见报错的 Issue 模板",[11,413,414],{},"四块内容讲的是一批事实，形态完全不同。三个月后六个引擎的覆盖率从平均 12% 到 24%。",[15,416,417],{"id":417},"一句实在话",[11,419,420],{},"跨引擎差异这件事，知道就好，别过度优化。真正决定你被不被引用的是内容里有没有可搬走的句子，这件工作在哪个引擎都一样。",[11,422,423],{},"引擎的偏好会变，去年和今年就不一样。你追着它改，永远慢一步。",{"title":98,"searchDepth":99,"depth":99,"links":425},[426,427,428,429,430],{"id":227,"depth":99,"text":227},{"id":346,"depth":99,"text":346},{"id":368,"depth":99,"text":368},{"id":392,"depth":99,"text":392},{"id":417,"depth":99,"text":417},"行业观察","2026-03-07","20 篇内容、60 个查询、6 个引擎。ChatGPT 引文档，Perplexity 引对比，AI Overviews 引表格，豆包引知乎，元宝引公众号。一份内容通吃是幻觉。",{},"\u002Fblog\u002Fcross-engine-citation-gap",{"title":219,"description":433},"blog\u002Fcross-engine-citation-gap",[439,440,441,442],"跨引擎","AI引用","GEO","平台差异","vA62lvr-AdPl0Ra4CThbC775hIOAXeYVtZG4nuHo2MQ",1794088107839]