[{"data":1,"prerenderedAt":361},["ShallowReactive",2],{"blog-customer-ticket-query-mining":3,"blog-prev-customer-ticket-query-mining":187,"blog-next-customer-ticket-query-mining":299},{"id":4,"title":5,"author":6,"body":7,"canonical":170,"category":171,"cover":170,"date":172,"description":173,"draft":174,"extension":175,"keywords":170,"meta":176,"navigation":177,"noindex":174,"ogImage":170,"path":178,"seo":179,"seoDescription":170,"seoTitle":170,"stem":180,"tags":181,"updated":170,"__hash__":186},"blog\u002Fblog\u002Fcustomer-ticket-query-mining.md","从客服工单里挖 AI 查询：1400 条工单找出 96 个问法","元析智能",{"type":8,"value":9,"toc":160},"minimark",[10,14,18,21,24,27,30,37,43,49,52,56,123,126,130,133,136,139,142,145,148,151,154,157],[11,12,13],"p",{},"最好的 AI 查询库不在关键词工具里，在客服工单里。2026 年 2 月我们翻了三个客户共 1400 条工单，挖出 96 个用户原话问法，其中 41 个在 AI 搜索里已经有现成回答，但回答里没有他们。",[15,16,17],"h2",{"id":17},"为什么工单比关键词工具准",[11,19,20],{},"关键词工具给的是\"别人搜过什么\"，带的是搜索框的痕迹——短、缺主语、被工具归并过。",[11,22,23],{},"工单给的是\"别人卡在哪\"。用户写工单时会把背景、环境、已经试过什么全写出来。这种句子拿去问 AI，得到的回答质量最高，因为信息完整。",[11,25,26],{},"我们对比过：同一个主题，关键词工具给出的月搜索量 2400，工单里对应的问题 37 条。37 条里的问法，没有一条能靠 SEO 关键词工具拿到。",[15,28,29],{"id":29},"三条筛选条件",[11,31,32,36],{},[33,34,35],"strong",{},"一、含疑问词。"," 怎么、能不能、为什么、是不是、有没有。陈述句多是需求和抱怨，问句才是查询。",[11,38,39,42],{},[33,40,41],{},"二、解决时长超过一天。"," 一天内解决的，说明文档已经答了，或者问题太个案。超过一天的，多半是文档没写清。",[11,44,45,48],{},[33,46,47],{},"三、重复出现。"," 同一个问题被 3 个以上客户问过，就是公共问题，值得写成内容。只出现一次的，写出来也没人搜。",[11,50,51],{},"1400 条过完，剩 96 条。",[15,53,55],{"id":54},"_96-个的分布","96 个的分布",[57,58,59,75],"table",{},[60,61,62],"thead",{},[63,64,65,69,72],"tr",{},[66,67,68],"th",{},"类型",[66,70,71],{},"数量",[66,73,74],{},"典型问法",[76,77,78,90,101,112],"tbody",{},[63,79,80,84,87],{},[81,82,83],"td",{},"配置类",[81,85,86],{},"38",[81,88,89],{},"怎么把同步任务设成增量",[63,91,92,95,98],{},[81,93,94],{},"报错类",[81,96,97],{},"27",[81,99,100],{},"报错 1045 是什么原因",[63,102,103,106,109],{},[81,104,105],{},"对比类",[81,107,108],{},"18",[81,110,111],{},"你们和 XX 有什么区别",[63,113,114,117,120],{},[81,115,116],{},"计费类",[81,118,119],{},"13",[81,121,122],{},"只跑一次任务怎么算钱",[11,124,125],{},"配置和报错占了三分之二，这跟工单的性质有关：能自己解决的不会来问。也说明这块最缺内容。",[15,127,129],{"id":128},"_41-个缺口长什么样","41 个缺口长什么样",[11,131,132],{},"我们把 96 个问法逐条拿去问 4 个引擎，看它答什么、引谁。",[11,134,135],{},"41 个已经有了像样的回答，但回答里的来源全是第三方：技术博客、论坛、竞品的文档。这 41 个最好办，因为问题已经被验证有答案，你只要写一个更好的，就有机会被替换进去。",[11,137,138],{},"剩下 55 个没有像样回答，AI 只能给一堆模糊的通用步骤。这类是首发机会，但风险是不确定有没有人搜。",[11,140,141],{},"我们的建议是先做前 41 个，做完再判断后 55 个。",[15,143,144],{"id":144},"一个副产物",[11,146,147],{},"翻工单还发现一件事：客服的问题分类错误率不低。1400 条里，有 200 多条被标错了类别，\"配置\"标签下混着计费问题。原因是客服赶时间，随手选。",[11,149,150],{},"这个错误率会直接影响客服的响应速度，也会让你以为某类问题少，其实是没被归类。顺手改标签规则比写内容收益更快。",[15,152,153],{"id":153},"谁该做这件事",[11,155,156],{},"内容团队自己翻工单，效率很低，因为看不懂技术细节。正确的做法是内容的人跟客服开一次两小时的会，让客服挑出 30 条\"最烦的问题\"，内容的人当场判断能不能写。",[11,158,159],{},"两小时能定下一个季度的选题。这比开选题会靠猜准得多。",{"title":161,"searchDepth":162,"depth":162,"links":163},"",2,[164,165,166,167,168,169],{"id":17,"depth":162,"text":17},{"id":29,"depth":162,"text":29},{"id":54,"depth":162,"text":55},{"id":128,"depth":162,"text":129},{"id":144,"depth":162,"text":144},{"id":153,"depth":162,"text":153},null,"GEO","2026-02-10","最好的查询库不在关键词工具里，在客服工单里。翻 1400 条工单挖出 96 个真实问法，41 个在 AI 搜索里已有回答但没我们的份。筛选条件只有三条。",false,"md",{},true,"\u002Fblog\u002Fcustomer-ticket-query-mining",{"title":5,"description":173},"blog\u002Fcustomer-ticket-query-mining",[182,183,184,185],"工单","选题","AI查询","内容运营","_uT9HLZ-bKCMcOTVBUDRBGDtRW3Sqpd7ZPFtvdC54ls",{"id":188,"title":189,"author":6,"body":190,"canonical":170,"category":287,"cover":170,"date":288,"description":289,"draft":174,"extension":175,"keywords":170,"meta":290,"navigation":177,"noindex":174,"ogImage":170,"path":291,"seo":292,"seoDescription":170,"seoTitle":170,"stem":293,"tags":294,"updated":170,"__hash__":298},"blog\u002Fblog\u002Fserp-intent-reverse-check.md","判断搜索意图的五分钟方法：搜一遍看五个位置",{"type":8,"value":191,"toc":276},[192,195,198,201,204,207,210,213,216,219,222,225,228,231,234,237,240,243,246,249,252,255,258,261,264,267,270,273],[11,193,194],{},"判断一个词是什么意图，最省时间的办法不是推理，是搜一遍看前十。我固定看五个位置，五分钟就能定页面类型。看错这一步，后面写什么都白费。",[15,196,197],{"id":197},"五个位置各看什么",[11,199,200],{},"前三条是不是同一种页面。有没有视频、图片、本地包这些富媒体。有没有电商或价格信息。有没有“相关问题”板块。前十里有几个是文档站。",[11,202,203],{},"这五个信号合起来，意图基本不会错。",[15,205,206],{"id":206},"三个词的实看结果",[11,208,209],{},"“API 网关”，前三条是百科和概念长文，没有价格，文档站占两个，相关问题都是“是什么”“有什么作用”。学习意图。",[11,211,212],{},"“API 网关 对比”，前三篇全是对比表和选型清单，有“相关问题”问“哪个好”。对比意图。",[11,214,215],{},"“API 网关 502 错误”，前三条全是文档和论坛排错贴，一条营销页都没有。操作意图。",[11,217,218],{},"同样是含“API 网关”的词，页面类型完全不同。",[15,220,221],{"id":221},"标题的动词也是信号",[11,223,224],{},"前三名的标题用什么动词，几乎可以反推出意图：怎么、如何、指南，偏操作；对比、哪个好、选择，偏决策；是什么、介绍、原理，偏学习。",[15,226,227],{"id":227},"富媒体出现要小心",[11,229,230],{},"一个词的前十如果出现视频或图片，说明用户更想看演示，你写一篇三千字文章也抢不到。",[11,232,233],{},"我们的做法是绕开，或者做成带截图和图解的页面。2025 年 10 月我们放弃过一个词，前十里有 4 个视频，硬写文章不划算。",[15,235,236],{"id":236},"看的时候要清环境",[11,238,239],{},"无痕窗口，别登录账号，能换设备就换。个性化会让结果偏离真实。",[11,241,242],{},"另外地域要看：同一个词在不同城市的本地结果可能不同，做本地业务时要按目标城市搜。",[15,244,245],{"id":245},"一个五分钟的清单",[11,247,248],{},"搜词，看前三条类型；看有没有富媒体；看有没有价格；看相关问题；看文档占比。五个看完，写下页面类型和理由。",[15,250,251],{"id":251},"搜索结果会骗人的一种情况",[11,253,254],{},"如果这个词刚出现、内容供给很少，前十可能是几个不相关的结果。这时候别照抄页面类型，看“相关问题”和搜索建议，那里更接近真实需求。",[11,256,257],{},"2025 年 8 月我们遇过一个新词，前十全是不相关的论坛帖，但相关问题问得非常集中，说明需求真实存在，只是没人认真写。这种词是机会。",[15,259,260],{"id":260},"一个反例",[11,262,263],{},"2025 年 6 月我们写过一个词，前十里有官网、有文档、有对比页，混杂。我们判断成对比意图，做成对比页，上线后排在第 14，一直不动。",[11,265,266],{},"后来复盘发现，这个词的学习意图更强，用户点进来是想先搞懂概念。页面没错，顺序错了。这类混杂词的稳妥做法是：一篇学习页加一篇对比页，先发学习页。",[15,268,269],{"id":269},"记录为什么重要",[11,271,272],{},"每个选题都写一句意图判断的理由和设备、地域、日期。半年后回看，你会发现自己错在哪类词上，这比记住一百个方法有用。",[11,274,275],{},"这个动作我们要求每个新选题都做一遍，记录在选题表里。理由写不出来的题，不进排期。",{"title":161,"searchDepth":162,"depth":162,"links":277},[278,279,280,281,282,283,284,285,286],{"id":197,"depth":162,"text":197},{"id":206,"depth":162,"text":206},{"id":221,"depth":162,"text":221},{"id":227,"depth":162,"text":227},{"id":236,"depth":162,"text":236},{"id":245,"depth":162,"text":245},{"id":251,"depth":162,"text":251},{"id":260,"depth":162,"text":260},{"id":269,"depth":162,"text":269},"SEO","2026-02-09","意图不用猜，搜索结果前十的页面类型就是答案。我们固定看五个位置：前三条、富媒体、价格、相关问题、文档站，五个位置看完，页面该长什么样就定了。",{},"\u002Fblog\u002Fserp-intent-reverse-check",{"title":189,"description":289},"blog\u002Fserp-intent-reverse-check",[295,296,297],"搜索意图","SERP分析","页面匹配","zhhoHVYlAaOdN6AkyMSVPwpz80490n_YhNpWVu_PnjE",{"id":300,"title":301,"author":6,"body":302,"canonical":170,"category":185,"cover":170,"date":350,"description":351,"draft":174,"extension":175,"keywords":170,"meta":352,"navigation":177,"noindex":174,"ogImage":170,"path":353,"seo":354,"seoDescription":170,"seoTitle":170,"stem":355,"tags":356,"updated":170,"__hash__":360},"blog\u002Fblog\u002Ftechnical-author-credentials.md","技术作者的履历怎么写：证书、项目、可查痕迹",{"type":8,"value":303,"toc":345},[304,307,311,314,317,321,324,327,330,333,336,339,342],[11,305,306],{},"技术作者的履历不用写长，但每一句都得能被查。我们审过 30 多个软件站的作者页，能真正查证的履历不到一半，大多数写的是\"多年互联网经验\"\"熟悉分布式架构\"这类谁也验证不了的话。",[15,308,310],{"id":309},"证书列三个以内带编号","证书：列三个以内，带编号",[11,312,313],{},"证书对技术读者有信号价值，前提是具体。写法是机构加年份，比如\"2023 年通过 CKA 认证\"\"2022 年 AWS 解决方案架构师\"。有编号的可以放编号，没有编号就别硬凑。",[11,315,316],{},"列三个以内。一堆证书堆上去，反而像培训机构的广告。如果作者没有证书，不如空着，用项目经历补，别硬编。",[15,318,320],{"id":319},"项目经历写到你改了什么这一层","项目经历：写到\"你改了什么\"这一层",[11,322,323],{},"这是最容易写虚的地方。\"参与过大型电商系统的架构设计\"没有信息量，读者不知道你做了什么。改成\"把订单系统的 P99 从 800 毫秒降到 220 毫秒，主要动作是把两处同步调用改成异步\"，信息量立刻不一样。",[11,325,326],{},"要有数字，也要有动作。数字说明结果，动作说明你真的在现场。写不出动作的项目，多半不是你亲手做的，这种就别写进履历。",[11,328,329],{},"2025 年 8 月我们改过一个客户的技术作者页，把原来那三段\"熟悉\u002F精通\"式的描述全删了，换成三个具体项目，每个项目写清楚问题、动作、结果。改完之后，这个作者的文章在站内的停留时间从 50 秒涨到 1 分 40 秒。数据不是立刻涨的，是两个月后慢慢起来的。",[15,331,332],{"id":332},"可查痕迹放哪几个平台",[11,334,335],{},"优先级从高到低：GitHub、技术社区主页（知乎、掘金、V2EX）、公开演讲或会议页面、开源项目贡献记录、专利和论文。",[11,337,338],{},"GitHub 对开发者工具类公司最重要，因为代码和 issue 记录做不了假。技术社区主页次之，看的是持续输出。演讲和会议页面适合有行业积累的作者，能直接证明\"这个人在圈子里\"。",[11,340,341],{},"这些链接要双向：作者页指向账号，账号资料里也写清楚在哪家公司、负责什么。单向的链接说服力弱一半。",[11,343,344],{},"履历这件事，被验证一次就建立了信任，被拆穿一次就全废了。宁短勿虚。",{"title":161,"searchDepth":162,"depth":162,"links":346},[347,348,349],{"id":309,"depth":162,"text":310},{"id":319,"depth":162,"text":320},{"id":332,"depth":162,"text":332},"2026-02-11","技术博客的作者页最怕写成简历。这篇讲三样东西怎么写才可验证：证书列几个、项目经历写到什么颗粒度、公开痕迹放哪几个平台，附一个改坏了的反面例子。",{},"\u002Fblog\u002Ftechnical-author-credentials",{"title":301,"description":351},"blog\u002Ftechnical-author-credentials",[357,358,359],"作者信号","技术写作","E-E-A-T","7oQuNYoIEElrRCfxLs86SRl1mf5aXNJJiP0LRgAeWCs",1794088108487]