前几天，我在一个平台发了条动态，预告自己在做的一个写作发布工具。底下有人说没啥用，不如做个抢优惠券的；也有人说别自嗨。
我没有生气，也没往心里去，但它让我想了很久。平台上的读者来自四面八方，关心的东西差别很大，同一件事在不同的人眼里，价值完全不同，这很正常。我真正在意的是另一件事：我想写的东西，是不是只能放进推荐流里，等一个不确定的结果？
这让我重新审视了一个老问题：在 AI 时代，个人博客还有什么位置，能起什么作用。
先说难听的：我知道现在开独立博客很傻。
平台有推荐，AI 有摘要，人连搜索结果都懒得点。新域名没有权重，没有外链，没有推荐位，写一篇长文发出去，阅读量个位数是常态。这不是“内容不够好”的问题，是分发结构就这样。所以问题不是“博客还有没有价值”，而是：在一个人人都不直接访问网站的时代，我为什么还要建一个自己的站？
yishuai.wang 的首页。
这看起来确实像 49 年入国军。
这个比喻，成立一半
成立的一半是：如果把独立站当流量生意做，那它就是逆风局。 靠 SEO 等自然流量、靠广告联盟变现、靠批量更新维持权重——这套玩法在 AI 把内容供给打到无限之后，基本没有胜算。AI 一秒能生成的通稿，你写十篇也没人看，搜索引擎也未必收录。这个意义上，49 年入国军不是玩笑。
不成立的一半是：49 年入国军是押注一个正在崩塌的旧秩序；我不是押注“独立博客会重新成为流量中心”。 我不认为独立站会回到 2010 年那种“写博客—被搜索引擎收录—拿流量—接广告”的黄金年代。那个年代确实结束了。
我押的是另一件事：AI 内容越泛滥，一手经历、真实项目、可验证署名，可能越稀缺。
现在下结论还太早。AI 可以帮我查资料、改代码、润色句子，但它没有做过我做的项目，没有踩过我踩的坑，没有在四个月里对着分页排版一个像素一个像素地改。它写不出“我当时为什么决定不用现成方案”这种判断，因为它没有“当时”。但这套东西到底有没有人认，我不知道，只能先做再看。
不是没人读，可能是阅读结构变了
“AI 时代少有人类阅读”这句话对一半。更准确地说：人直接访问一个陌生独立站的概率变低了，但人通过 AI 搜索、AI 摘要、GitHub、开发者社区、RSS 找到你的路径，可能变多了。
读者未必消失，入口可能在变。以前入口是搜索引擎结果页，现在入口可能是：
- 有人问 AI“中文分页编辑器怎么做”，AI 引用了你的文章；
- 有人在 GitHub 看到 QuireKit，顺藤摸瓜找到你的站；
- 有人从 V2EX、即刻、X、掘金看到一段踩坑记录，点进来看完整版；
- 有人用 RSS 订阅了你的站，每篇都读，但从不留言。
这些都不体现在“首页访问量”里。但这也只是我的观察，不一定对。还有一个坑必须说清楚：被 AI 读到 ≠ 被 AI 引用 ≠ 被 AI 导流。 放一个 llms.txt 就以为 AI 会替你传播，是自我安慰。llms.txt 不是标准，很多 AI 不读，读了也不一定引用；被训练进模型更不等于有人会来你的站。所以我没把 llms.txt 当战略，只当导览。我试图做的是：让站更像一个可被引用的一手源，并且尽量确保引用里有我的名字、链接和工具。 做不做得成，另说。
还有一些很个人的原因
yishuai.wang 就是我的名字。我很喜欢这个域名，只是一直没好好用起来。
这几年在各个平台发文，我也有个感受：有些平台越来越平台化，推荐、热榜、流量规则排在前面，技术本身反而退到了后面。V2EX 是我觉得一直保持初心的地方，也是给我帮助最多的社区。如果有一天有机会，我也想在国内做一个像它这样的社区。现在还很远，先从把自己的站写好开始。
所以我重新把站点搭了起来，也终于把这个域名用上了。给自己留一块净土：不用围着推荐规则写，想写什么就写什么。
我正在尝试的事
都是很具体、也还没验证的事。
第一，写我想写的，但让站里有一部分内容有“为什么是我”。 我会写工程记录，也会写随笔、观点、工具分享。这些内容别人也能写，我不装作每篇都有独家价值。但站里得有一部分是别人替代不了的：QuireKit 的工程记录、分页排版的踩坑、为什么不用现成方案、四个月做到 Word 常用功能 85% 兼容是什么过程。这些不是每篇都有，但它们决定了这个站“为什么是我”。其余内容自由写，写得开心就写，只是我不拿它们当护城河。
第二，让独立站做首发站。 以后文章先发在 yishuai.wang，再同步到各个平台，平台上也是完整内容，不做删减。之后有修改和补充，先更新站内版本。能设 canonical 的平台就指向站内原文，不能设的就在文末写一句“本文首发并持续更新于 yishuai.wang”。
第三，给 AI 和人准备可引用的结构。 长文里补结论卡、事实表、FAQ、版本记录；给 QuireKit 单独做工具页，写清楚一句话定义、功能、边界、许可证、对比、路线图。AI 最爱抽取的通常不是长叙事，是结论、事实、版本、对比、代码和边界。工具页可能比博客文章更容易被 AI 推荐。这只是假设，我按这个假设在做。
文章页：开头是智能总结，右侧是目录，每篇都能直接拿到 Markdown 原文。
第四，保留 RSS。 独立站没有平台推荐，RSS 是我能想到的少数真正属于我的连接：不经过算法，更新了就能收到。中文、英文和全部内容各有一个订阅源。有没有人用，还要再看。
第五，不和 AI 对抗，把 AI 当一层分发。 我不指望拦着 AI 读我的站，也拦不住。我能做的是让 AI 读到的内容是可验证的、有署名的、有链接的。AI 的输出有人看，那我希望那个输出里，有我的名字和我的工具。
这个站，是用我自己做的工具搭的
我清楚 QuireKit 是什么、能做什么，所以愿意自己先吃这份 dogfood：这个站就是用它搭的。
QuireKit 是我在做的开源文档编辑器生态。主打的是 Classic 分页文档编辑器：在浏览器里真正分页，断行、分页、光标、选区都自己算，像 Word、WPS 一样一页一页地编辑；另一条线是 MSFE，专为 Markdown 定制的编辑器框架，Markdown 原文就是文档。
yishuai.wang 从页面到写作，用的都是它：
- 页面用 QuireKit 的组件库和 InkWash 水墨主题；
- 写文章的后台和评论框都是 MD Editor（基于 MSFE），文档、实时预览、源码三种视图随时切换；
- 每篇文章在地址后面加 .md 就能拿到原文，配上中英文 RSS、canonical、llms.txt，就是上面说的“可引用的结构”；
- 文章开头的智能总结在发布后生成一次，读者阅读时不会调用模型；
- 站点也支持 Micropub，以后可以从别的写作工具直接发文到站上。
评论框就是 MD Editor，页脚是 RSS 和 llms.txt。
站点源码也开源了（MIT）：github.com/kingshuaish… 。自己的东西先自己用，哪里不顺手，最先知道的是我。
所以，是 49 年入国军吗
我认为不是。但这只是我的个人观点，不构成对任何人的建议。
国军押的是一套正在崩塌的秩序；我押的是另一件事：当 AI 把内容供给打到无限，人的经历、判断、署名和可验证的一手记录，可能会变贵。 独立站可能没人看，但它至少是我留给自己的一个源头。域名、文章和版本记录归我，不被平台算法拿捏，也不被 AI 摘要一笔带过。
还有一层原因。好的中文技术博客本来就不多，而大模型能答成什么样，很大程度上取决于它能读到什么。如果资料大多是搬运和水文，模型再强，也会被这样的资料生态限制住。一个人的站很渺小，像一颗星；但好的内容生态，本来就是很多颗星组成的。我想做其中一颗。
没人看是常态，不是判决。我先写一批只有我能写的东西，再写更多我想写的东西，然后看它值不值得活下去。
AI 可以读它、引用它、忽略它。但源头在这里。
至于这是不是 49 年入国军——我现在不知道。我只是觉得，与其等一个确定的答案，不如先把自己的那一份留下来。
本文首发并持续更新于 yishuai.wang。
本文由 AI图鉴自动抓取并翻译，版权归原作者所有；如需查看原文请点下方链接。
A few days ago, I posted a dynamic on a platform that predicted a publish tool I was working on. Some people say it's no use, make a coupon; others say don't get high.
I wasn't angry, I wasn't thinking about it, but it took me a long time. Readers on the platform come from all sides, and the concerns vary widely, and the same thing is completely different in the eyes of people, and it is normal. What I really care about is another thing: do I want to write something that can only be put in a referral stream for an uncertain outcome?
And this brings me back to the old question of what role personal blogs can play in the AI era.
I know it's stupid to start an independent blog now.
The platform has recommendations, AI has summary, and people don't even bother looking for results. The new domain name has no weight, no external chain, no recommended position, and a long text has been written and the reading number is normal. This is not a question of “not good enough” but of distribution structures. So the question is not “blogs have no value”, but why should I build my own station in a time when everyone does not have direct access to the site?
Yishuai.wang's home page.
It does look like 49 years of military service.
That's a metaphor. Half.
Half of what was established was that if the stand was to be treated as a flow business, it would be counterwind. With natural flows such as SEO, the realization of cash through advertising alliances, and the maintenance of weights through bulk updating -- this game is almost unwinnable after AI hits the limit. It's a one-second draft, and you don't read ten of them, and the search engine doesn't have to be recorded. In that sense, 49 years of military service was no joke.
Half of the failure is: 49 years ago the National Army was betting on an old order that was crumbling; I am not betting that “independent bloggers will be back in the flow centre”. I don't think independent stations are going back to the golden age of 2010 of writing blogs -- being captured by search engines -- taking traffic -- advertising. That was the end of it.
I'm betting on another thing: the more the content of AI becomes popular, the less the experience, the real project, the less the signature can be verified.
It is too early to conclude. AI can help me check the data, change the code, colour the sentence, but it didn't do what I did, it didn't step on the pit I stepped on, it didn't do a pixel to a page layout for four months. It failed to write the judgement “why did I decide not to use the available formula” because it did not “then”. But I don't know if anyone recognizes this set, but I have to do it first.
It's not that nobody reads it. It's probably a different reading structure.
The phrase “Ai age is rarely read by humans” is in half. More precisely, the probability of people visiting a strange stand-alone station is lower, but people finding your path through AI search, AI summary, GitHub, the developer community, RSS may have changed a lot.
Readers don't necessarily disappear. The entrance may change. The entrance used to be the result page of the search engine, and now it may be:
- Some ask AI, "What does the Chinese page editor do?" AI quotes your article;
- Someone in Github saw QuireKit, and Suto found your station;
- A permutation record was seen from V2EX, instantaneous, X, gold digs, and a full version was shown;
- Someone subscribes to your station, reads every piece, but never leaves a message.
This is not reflected in the “first page visits”. But it's just my observation, not necessarily. There is also a pit that must be made clear: read by AI to be quoted by AI to be guided by AI. Put a llms.txt on it thought AI would spread for you, it was self-consolation. Alls.txt is not a standard, many AI doesn't read, read and not necessarily quoted; being trained in models is not the same as coming to your station. So I didn't put allms.txt as strategy, as guide. What I'm trying to do is to make the station more a first-hand source to be quoted and try to make sure that my name, links and tools are included in the reference. Do it or not, otherwise.
There are some very personal reasons.
Yishuai.wang is my name. I love this domain name, but it's not working.
I've been on various platforms for years, and I have a feeling: Some platforms are becoming more platform-driven, with recommendations, hot sheets, flow rules ahead of them, and technology itself retreats back. V2EX is the place where I feel I've kept my head and the community that's helping me most. If I had a chance one day, I'd like to be a community like it in the country. It's still a long way to go, starting with writing your feet.
So I reset the site and finally used the domain name. Leave yourself a piece of dirt: write without the recommended rules, write anything you want.
I'm trying.
It's all very specific and untested.
First, to write what I want to write, but to have part of the station say “Why me”. I'll write the project records and I'll write the notes, the views, the tools. It's something that others can write, and I don't pretend to have exclusive value in every piece. But there's a part of the station that can't be replaced: the project records of QuireKit, the page layout pits, why not use a ready-made solution, four months of Word's common feature 85% compatibility. These are not everything, but they decide the station “why me”. The rest is written freely, happyly, but I don't use them as a moat.
Second, get independent stations to start. The post will then be posted on yishuai.wang and synchronized to the platforms, which will also be complete without deletion. This will be followed by modifications and additions, which will be updated in-station versions. The platform that can set up canonical points to the originals of the station, and it cannot be created by writing at the end of the text, "This paper was first published and kept up to date in yishuai.wang".
Third, give AI and people the structure to quote. The long text contains summary cards, fact sheets, FAQ, version records; a separate tool page is given to QuireKit with a clear sentence definition, function, boundary, licence, comparison, road map. AI prefers not long narratives, but conclusions, facts, versions, comparisons, codes and boundaries. The tool pages may be more easily recommended by AI than blog articles. It's just a hypothesis I'm doing.
Article Page: The beginning is a smart summary, the right side is a directory, and each article is directly available in the original Markdown.
Fourth, retain RSS. There's no platform for independent stations, RSS is the few connections that I can think of that really belong to me: they'll be received without algorithms. There is one subscriber for Chinese, English and all. There's no one else to see.
Fifth, don't fight AI, put AI on the floor. I don't expect to stop AI from reading my station, nor can I. All I can do is let AI read something that is verifiable, signed, linked. AI's output is seen, and I want that output to have my name and my tools.
This station was built with my own tools.
I know what QuireKit is and what he can do, so I'd like to eat this dogfood: this station is built with it.
QuireKit is the ecology of the open source document editor I'm making. Classic Page Break Document Editor: The true page break, break, cursor, constituency calculation in the browser is done on a page-by-page basis like Word, WPS; the other line is MSFE, the editor frame customised for Markdown, the original Markdown is the document.
Yishuai.wang uses it from page to writing:
- A module library with QuireKit and an InkWash theme on the page;
- Both backstage and comment boxes for articles are MD Editor (based on MSFE), documents, real-time previews, and source-code views are always switched;
- Each article adds.md to the address and is available in its original language, in English, RSS, Canonical, llms.txt, which is the “referenceable structure”;
- An intellectual summary at the beginning of the article is produced once after publish and the reader does not call the model when reading it;
- The site also supports Micropub, which can be sent directly to the station from another writing tool.
The comment box is MD Editor and the footer is RSS and llms.txt.
Site source (MIT): github.com/kingshuaish... I'm the first to know how to use my things.
So it's 49 years old?
I don't think so. But this is my personal point of view and does not constitute a recommendation to anyone.
The National Army is in an order that is falling; I am in another: when AI supplies the content to the limit, human experience, judgement, signature and verifiable first-hand record may become expensive. Independent stations may not be looked at, but they are at least a source for me. Domain names, articles, and versions are mine, not captured by platform algorithms, and not taken by AI summary.
There is one more reason. There are not many good Chinese-language blogs, and LLM's answer depends largely on what it can read. If most of the data are porter and hydrologic, the model is stronger and the ecological limitations of such information are also present. A man's standing is small, like a star; but good content ecology is made up of many stars. I want to do one of them.
Nobody looks normal, not judgment. I'll write something that I can only write, and then write more that I want to write, and see if it's worth living.
AI can read it, quote it, ignore it. But the source is here.
As for whether it was 49 years old, I don't know now. I just think that instead of waiting for a definitive answer, we should leave our share.
This post has been launched and is continuously updated in yishuai.wang.
Auto-fetched and translated by AI Atlas. All rights belong to the original author.