A Deep Dive into Modern SEO: Principles, Algorithms and Practice

1.1 Crawl Frontier & URL Discovery#

Search engines rely on a system called Crawl Frontier (the crawl frontier).
The crawl frontier is a list that manages URLs; crawlers fetch pages according to the URLs in the queue.
URLs are discovered mainly in four ways:

• Internal Links: the crawler discovers new pages through the <a> tags in pages it has already crawled. For example, if your Arch Linux Installation Guide contains a link to A Detailed Guide to GRUB Boot Configuration, the crawler will discover the latter page through this link
• External links Backlinks: links from other sites to yours are also an important discovery path. For example, if a rhythm game community, in an article rounding up the works of the composer xi, links to your Which songs does xi have in maimai DX?, then that is an external link For the SEO of external links, see the PageRank explanation
• Sitemap: the site owner provides a sitemap.xml that lists all the important pages of the site. Search engines can use this file to quickly learn what content exists under your site and index it
• History & Prediction: based on previous crawl data, search engines predict new URLs. For example, suppose your URL structure is /chongxi/1/; the crawler may infer that /chongxi/2/ might exist, and keep crawling

1.2 Crawling Strategies: BFS & Priority & Adaptive Scheduling#

Faced with a huge number of URLs, crawlers need to manage crawl order efficiently The following may have already been covered in the previous search engine deep dive, so I’ll keep it brief here as a review

• BFS: the most basic crawling strategy. Starting from a seed page, it prioritizes pages closer to the current one (fewer link levels)

• Priority Scheduling: modern crawlers dynamically adjust priority based on various signals, for example:

  1.PageRank: if your maimai DX Beginner's Guide is linked by many high-quality sites, the crawler may consider this page more important and crawl and update it first, but PageRank has long been a thing of the past, replaced by more advanced authority evaluation; see 3.3 for details

  2.Update frequency: for example, a page like maimai DX attendance live feed may be crawled more frequently

  3.User demand: if compiling a new OpenWrt version becomes a hot search, related tutorial pages may be crawled first

1.3 Crawl Budget#

Crawl Budget refers to the time and resource limits a search engine places on crawling a site. It consists of two parts:

1. Crawl Rate Limit: the limit on how frequently the search engine crawls your server. If your server responds quickly and doesn’t produce large numbers of 404 or 5xx errors, the crawler will crawl more frequently
2. Crawl Demand: how much value the search engine believes a page has. For example, if your page is very valuable, or its content updates often such as a frequently updated maimai DX song list, the crawler will crawl it more frequently

1.4 Layers and Logic of Crawl Control Protocols#

Site owners can control crawlers in several ways

• robots.txt file: located in the site’s root directory; tells crawlers which paths must not be accessed Configuration example:

1
User-agent: *
2
Disallow: /tmp/

• <meta name="robots">: page-level control, placed in the HTML <head> section Configuration example:

1
<meta name="robots" content="noindex">

This means the page can be crawled, but must not be shown in search results

• X-Robots-Tag: controls crawling and indexing through HTTP response headers; suitable for non-HTML files (such as PDFs and images)

2.1 Text Preprocessing Pipeline#

Crawled page content goes through a series of processing steps:

1. Content extraction: strip templated content such as ads, sidebars and footers, and focus on the core body text
2. Tokenization: split the article content into individual tokens token. For example, the Chinese query 小米路由器刷 openwrt would be split into 小米, 路由器, 刷入, openwrt
3. Normalization: includes lowercasing, lemmatization (e.g. “running” to “run”), and removal of stop words (such as “的”, “了”)

2.2 Core Data Structure I: Inverted Index#

The inverted index is one of the core data structures of search engines. It maps tokens → the list of document IDs containing that token, allowing fast lookup

NOTE

EXAMPLE

Suppose your blog has the following three articles:

• Article 1 ID: 1: title The Ultimate Guide to Optimizing Steam Proton Performance on Arch Linux
• Article 2 ID: 2: title Installing and Configuring i3wm on Arch Linux
• Article 3 ID: 3: title A Few Steam Indie Games That Run Smoothly on Linux

After text preprocessing (tokenization, normalization), the search engine builds the following inverted index table (Posting List) for some key tokens:

Token Token	List of document IDs containing the token Posting List
arch	[1, 2]
linux	[1, 2, 3]
steam	[1, 3]
guide	[1]
game	[3]
install	[2]

Now let’s see how the inverted index works when users perform different searches:

Scenario one: a simple query

When a user searches Arch, the search engine goes straight to the inverted list for arch and instantly returns [Document 1, Document 2]

Scenario two: a boolean query

When a user searches Arch Linux Steam, the search engine performs the following efficient intersection operation: 1. Take the list for arch: [1, 2] 2. Take the list for linux: [1, 2, 3] 3. Take the list for steam: [1, 3]

Next, merge these sorted lists to compute the intersection:

PRTCL // PLAINTEXT

1
- [1, 2] AND [1, 2, 3]  => [1, 2]
2

3
 *then use the result of the previous step to continue the intersection*
4

5
- [1, 2] AND [1, 3] => [1]

COPY

At this point, the search engine has within milliseconds precisely pinpointed the only relevant Article 1, without any need to scan the hundreds of millions of documents irrelevant to the query

2.3 Core Data Structure II: Vector Index & Semantic Space#

Modern search engines have introduced vector indexes to understand the semantics of a query. Through deep learning techniques, each word or document is converted into a high-dimensional vector

NOTE

EXAMPLE

Your two articles How to Set Up Home Cloud Storage with OpenWrt and How to Set Up a VPN Server with OpenWrt don’t share exactly the same keywords, but their topics are related.
Through vector indexing, the search engine can turn their semantic relationship into distance in vector space and know that they both belong to the topic OpenWrt advanced configuration. So when a user searches OpenWrt VPN, the search engine may also recommend the OpenWrt home cloud storage article

2.4 Content Deduplication and Normalization#

The internet is full of duplicate content. Search engines use deduplication algorithms such as SimHash to identify similar content and index only one authoritative version

NOTE

EXAMPLE

• Article A: How to install the OpenClash service on OpenWRT
• Article B: OpenWrt OpenClash installation and configuration guide

If the content is highly similar, the search engine will consider them duplicates. At this point the rel="canonical" tag comes into play: it can explicitly tell the search engine which one is the primary version, and the weight of the other duplicates is consolidated into the primary version

3.1 The Multi-stage Funnel Model of Ranking#

The following was covered earlier, so here is a brief summary

The ranking process can be seen as a funnel model, progressively filtering and refining results:

1. Recall Recall: through the inverted index and vector index, find several thousand of the most relevant candidate documents out of billions
2. Coarse ranking Coarse Ranking: use a computationally cheap algorithm (such as BM25) to give candidates an initial score, filtering down to a few hundred most promising results
3. Fine-grained ranking Fine-grained Ranking: use complex machine learning models that combine hundreds of signals to rank candidates more precisely
4. Re-ranking Re-ranking: on top of the fine ranking, make final adjustments based on the user’s personalization needs, recency and other factors

3.2 Core Coarse Ranking Algorithm: BM25#

BM25 is a coarse ranking algorithm commonly used by search engines. It estimates the relevance between a document and a query based on a probabilistic model, and it is an improved version of TF-IDF

NOTE

EXAMPLE

Query: OpenWrt proxy service configuration

• Document A: How to configure a proxy service on OpenWrt, 1500 words, the keywords “OpenWrt” and “proxy service” appear 10 times

• Document B: OpenWrt Basic Setup Guide, 8000 words, also mentions “proxy service” a few times, but spends more of its length on DHCP, DNS and so on

How BM25 scores:

1. Term frequency (TF): document A has a higher keyword frequency, so it scores higher.
2. Document length: BM25 penalizes overly long documents. Document B also contains the keywords, but because it’s so long and its topic isn’t focused enough, its score is dragged down, which prevents long documents from gaining an unfair advantage by containing more scattered keywords
3. Term frequency saturation: even if document A repeats “proxy service” 100 times, its score won’t grow without bound, which effectively prevents keyword stuffing

3.3 Link Authority Models: The Sudden Death of PageRank & Modern Authority Models#

PageRank is Google’s early, revolutionary algorithm. It was the first to turn the chaotic links of the internet into a huge, computable network of trust votes

A page’s authority is partly inherited from the authority of the other pages that link to it. A link is a recommendation ballot

NOTE

EXAMPLE

Suppose you wrote an article on A Study of Composer xi’s Famous BMS-era Tracks

• Case one: the article is published only on your personal blog
• Case two: the article is cited by BEMANI WIKI and the blogs of some veteran music critics

From the perspective of traditional PageRank: in case two, every link pointing to your article is a trust vote. The weight from an expert like BEMANI WIKI is far higher than that of 100 ordinary forum signature links. So the article in case two is assigned a much higher PageRank value and is considered a more authoritative, more trustworthy resource

CAUTION

You’re right, but

That public PageRank score from 0 to 10 stopped being updated as early as late 2013, and was formally sentenced to death by Google in 2016!

Why did it die?

Its publicness and simplicity spawned a huge, distorted link trading market. Black-hat SEOs stopped focusing on creating high-quality content and obsessed over buying and exchanging high-PR links, trying to manipulate this voting system

It was like an election with large-scale bribery and fake ballots, leaving the result untrustworthy. Public PageRank was alienated from a measure of authority into a vulnerability that could be directly attacked and exploited, polluting the entire search ecosystem

To preserve the fairness of search results, Google had to personally execute the public form of this once-flagship algorithm (good riddance, meow)

The modern authority model, PageRank reborn

PR is dead, but the core idea of calculating authority through link analysis didn’t disappear. Instead, in a more complex, smarter and harder-to-manipulate form, it merged into the ocean of hundreds of ranking signals used by modern search engines

We can think of the modern authority model as PageRank-ADVANCED. It no longer cares only about who voted, but more about:

1. The topical relevance of the vote Topical Relevance How the modern model thinks: the value of a link depends largely on the topic of the page it sits on

NOTE

EXAMPLE

A link from an Arch Linux Chinese Wiki page to your Arch Linux Installation Guide is far more valuable than a link from a food blog directory, even if the latter might have a higher traditional PR value

Because the former tells Google: an authority in the Linux field recommended another Linux resource
That is a highly relevant, professional endorsement

2. The surrounding context of the link Surrounding Context How the modern model thinks: the text around the link — was it carefully placed in the body by an editor, or is it buried in a pile of links in the site footer?

NOTE

EXAMPLE

High value: In the body of an article reviewing an OpenWrt router, the author writes: for detailed configuration of a bypass router, I strongly recommend this Ultra-detailed OpenWrt Bypass Router Configuration Guide on Bulianglin; he explains every pitfall clearly this is not a sponsored post

Low value: In the blogroll area of your friend’s blog footer, there’s a simple anchor text link Chongxi’s blog

The former’s context clearly explains why it’s recommending you; that’s a strong editorial endorsement signal

3. User behavior around the link User Behavior How the modern model thinks: does anyone actually look at this link? After clicking, do users stay a long time or bounce straight back

NOTE

EXAMPLE

Google can obtain anonymized user behavior data through Chrome, Android and various other channels. If a link pointing to your maimai DX Chart Constants Table is clicked by large numbers of users, and they stay on your page for an average of 5 minutes, that effectively verifies the link’s real value

Today’s link authority is a complex score combining relevance, context, editorial value and real user feedback. It is far harder to manipulate than PageRank, and therefore fairer to creators who genuinely produce high-quality, professional content. I hope that after reading this, you’ll stop trying to outsmart SEO through every possible channel; good wine needs no bush, and writing quality content comes first

3.5 Neural Re-ranking#

This is one of the most advanced techniques, using deep learning models (such as Transformer, BERT) to perform finer semantic matching that goes beyond simple keyword matching

NOTE

EXAMPLE

Query: How do I quickly raise my Rating in maimai DX

How neural re-ranking understands it: Traditional methods: might look for articles containing the words “maimai DX”, “quickly”, “raise” and “Rating”

Neural models: they can understand that the deep intent of this query is “finding efficient score-farming tricks”
So it will not only return articles whose titles match, but will also rate highly an article titled Recommended Easy Songs for Farming Score, even if the words “quickly raise” never appear in it. Because at the semantic level it understands that “farming easy songs” is an excellent answer to the question of “quickly raising Rating”

4.1 Classic Algorithms: BM25 & TF-IDF#

Covered in detail in section 3.2; TF-IDF is the foundation of BM25, and its core idea is to evaluate word importance through term frequency and inverse document frequency

4.2 HITS#

Hyperlink-Induced Topic Search, unlike PageRank, is query-dependent. It divides pages into two types:

• Hub Hub: a high-quality hub page is a resource list page that links to multiple related authority pages
• Authority Authority: a high-quality authority page is an official or definitive content page that is linked to by multiple related hub pages

NOTE

EXAMPLE

Query: OpenWrt firmware download
Authority page: the OpenWrt official firmware download page
Hub page (Hub): a blog post titled OpenWrt from Beginner to Expert that lists and links the official firmware download page, the LEDE firmware download page, and the GitHub pages of several well-known third-party firmware builds; that article is a high-quality hub page

4.3 LTR#

Covered in detail in section 3.4 as the core of modern ranking

4.4 Deep Learning#

Covered in detail in section 3.5 as a frontier semantic matching technique; its core is using models such as BERT for deep semantic understanding

5.1 Title#

NOTE

EXAMPLE

Bad: OpenWrt

Awesome: [2025] The Ultra-Detailed Guide to Configuring an OpenWrt Router

5.2 Snippet#

The snippet is the short description shown below the title. Search engines automatically extract the part they consider most relevant.

5.3 URL and Breadcrumb Navigation Breadcrumb Navigation#

A clear URL structure and breadcrumb navigation help users and search engines understand a page’s position within the site

NOTE

EXAMPLE

• URL: .../guides/openwrt/openclash/
• Breadcrumb: Home > Guides > OpenWrt > OpenClash Configuration Tutorial

I suggest using BreadcrumbList Schema.org structured data to mark up your breadcrumb navigation, which gives it a chance to appear directly in search results. json { "@context": "https://schema.org", "@type": "BreadcrumbList", "itemListElement": [{ "@type": "ListItem", "position": 1, "name": "指南", "item": "/guides" },{ "@type": "ListItem", "position": 2, "name": "OpenWrt", "item": "/guides/openwrt" }] }

6. Technical SEO#

Technical SEO aims to optimize a site’s technical structure, ensuring search engines can smoothly crawl, render and index it, and improving user experience

6.1 Crawling and Indexing Optimization Guide#

6.2 Website Performance Optimization Core Web Vitals#

6.3 Structured Data Schema Markup#

Using the Schema.org standard can help search engines better understand your page content

• Basic deployment: add markup for articles, blogs, organization info and so on
• Aim for rich results

NOTE

EXAMPLE

At the end of the OpenWrt Configuration Guide, there’s an FAQ section json { "@context": "https://schema.org", "@type": "FAQPage", "mainEntity": [{ "@type": "Question", "name": "旁路由模式下如何避免DHCP冲突？", "acceptedAnswer": { "@type": "Answer", "text": "核心是关闭旁路由的DHCP功能，并将其网关指向主路由..." } }] }

After deployment, this Q&A may appear directly below your search result, drawing users’ clicks

6.4 Mastering nofollow, sponsored, ugc#

Driving traffic is of course important, but equally important is having a valve to precisely control how your site’s weight flows out, and to show search engines the nature of your links to other sites
That’s what the three values nofollow, sponsored and ugc in the rel attribute are for

7.1 Keyword Research and Topic Modeling#

7.2 E-E-A-T#

E-E-A-T stands for

Experience
Expertise
Authoritativeness
Trustworthiness

8.1 Modern Link Building Strategies#

8.2 Off-site Reputation and Brand Signal Building#

9.1 CSR & SSR & SSG#

First, we need to understand how Google’s crawler Googlebot handles JavaScript. It has a component called the Web Rendering Service, essentially a headless browser. The crawling process has two waves:

1. First wave: Googlebot crawls your raw HTML; if the content is all in the HTML, it’s sent straight to indexing
2. Second wave: if the page needs JS to generate content, the page is placed in a rendering queue. Days or even weeks later, the WRS executes the JS, renders the final DOM, and sends the rendered HTML to indexing

9.2 Dynamic Rendering#

At the server level, decide what content to return by identifying the request’s User-Agent

• If the User-Agent is an ordinary user’s browser, return the regular CSR application
• If the User-Agent is Googlebot or another crawler, the server uses middleware (such as Rendertron) to render your JS app into static HTML, then returns that HTML to the crawler

WARNING

Why it was abandoned

This was once Google’s recommended approach, but it is now considered a stopgap. It increases system complexity and maintenance cost, and risks being misjudged as cloaking (i.e. showing crawlers and users different content)

Unless you have a massive legacy CSR project you can’t refactor, you should prefer SSR or SSG

9.4 Writing crawler-friendly JS code#

Even if you choose CSR, you can reduce the risk of rendering failure through some best practices

1
// ×××××爆赞×××××
2
<a href="/openwrt-guide" onClick={handleNavigate}>OpenWrt 指南</a>
3
```

10.1. Image SEO#

Search engines aren’t human; they can’t truly see what’s in an image (AI image recognition is improving, but it’s nowhere near something you can fully rely on). They depend on the metadata you provide to understand images

10.2 Video SEO#

Video results usually take up huge space on the search page and are very eye-catching. A well-optimized video is a powerful tool for earning a high click-through rate

11.1 What is an entity#

• String: “composer xi”

• Entity: an ID (such as /m/114_514), and this ID is associated with a set of attributes and relationships:

1
- Name: xi
2
- Occupation: composer
3
- Nationality: Japan
4
- Affiliation: Diverse System
5
- Representative works: Freedom Dive, Ascention to Heaven
6

7
- Attended: BOF
8
- Works included in: iidx, sdvx, maimai...

When Google can understand xi like this, it can answer more complex user queries, such as “which xi songs are included in iidx?”.

Our goal is to help Google build such an entity profile for you or your internet assets as well

11.2 How to build your brand entity#

12.1 Choosing a URL structure#

Your first step toward going global, and the most important architectural decision. You have three main ways to organize multilingual content, each with clear technical pros and cons and signal strength

WARNING

ccTLDs

Structure: chongxi.jp (Japanese), chongxi.us (English)

Pros:

• Strongest geolocation signal: the .jp domain itself strongly tells Google and users: “this site is specifically for the Japanese market.” This is a huge advantage for local rankings
• Flexible server location: you can put the .jp server in Tokyo and the .de server in Frankfurt, giving local users the fastest access speed

Cons:

• High cost: you need to register and manage multiple domains.
• Diluted authority: each domain is an independent site; you need to build backlinks and authority separately for .jp and .de, and they do not automatically inherit from each other

Suitable for: large multinational brands with ample budget and local operations teams. For personal blogs, this is usually not the first choice

WARNING

Subdirectories/Subfolders

Structure: chongxi.us/ja/ (Japanese), chongxi.us/en/ (English)

Pros:

• Concentrated authority: all language versions live under the same main domain; they share and jointly contribute to the overall authority of chongxi.us. Backlinks you earn for the English version also pass weight to the Japanese version, and vice versa. This is the biggest advantage.
• Low maintenance cost: only one site, one hosting account and one SSL certificate to manage
• Easy setup: the easiest to implement technically.

Cons:

• Geolocation signals are relatively weak (but this can be compensated for with GSC settings)

Suitable for: for the vast majority of sites, this is the most recommended and most powerful option

WARNING

Subdomains

Structure: ja.chongxi.us, en.chongxi.us

Somewhere between the first two. Google officially says it now handles weight transfer between subdomains well, but the SEO industry generally believes its authority concentration is still slightly worse than subdirectories

Suitable for: when different language versions differ hugely in content, design or even functionality and need to be fully isolated technically. For example, a main site that’s an e-commerce store and a Japanese site that’s a pure content blog

12.2 Core technical implementation#

hreflang is an HTML <link> tag attribute whose job is to precisely declare to Google: the German version of this page is here, the Japanese version is there, and this version is the default English one

WARNING

Characteristics

1. Solves the duplicate content problem:
   Prevents Google from treating the English and Japanese translations of your Arch Linux Installation Guide as plagiarism
2. Provides the right search results:
   Ensures Japanese users searching on google.co.jp see your Japanese article rather than the English one

CAUTION

Implementation rules

The implementation is extremely strict; not one mistake is allowed

1. Reciprocal:
   If page A links to page B, then page B must link back to page A
2. Self-referential:
   Every page must have an hreflang tag pointing to itself
3. Include x-default:
   You should designate a default or fallback version, to which users are directed when their language or region doesn’t match any version you provide

NOTE

EXAMPLE

Suppose an article has three versions:

• English (default): /en/music/xi
• Japanese: /ja/music/xi
• Simplified Chinese: /zh-hans/music/xi

Then, in the <head> of all three pages, you must include the exactly identical block below:

PRTCL // PLAINTEXT

1
```html
2
<link rel="alternate" hreflang="en" href="/en/music/xi" />
3
<link rel="alternate" hreflang="ja" href="/ja/music/xi" />
4
<link rel="alternate" hreflang="zh-Hans" href="/zh-hans/music/xi" />
5
<link rel="alternate" hreflang="x-default" href="/en/music/xi" />
6
```

COPY

Language and region codes:

• You must use ISO 639-1 format for languages (e.g. en, ja, zh)
• You can append ISO 3166-1 Alpha 2 format for regions (e.g. en-GB for British English, en-US for American English)
• Note: zh-Hans and zh-Hant are exceptions; they are script codes

Deployment methods:

• HTML <head>: most common, as in the example above
• HTTP response headers: suitable for non-HTML files (such as PDF)
• XML sitemaps: when the number of pages is huge, this is the most efficient and most recommended management method

12.3 Content and servers#

Technical tags are only the foundation; content and experience are what decide success or failure

Localize content, don’t machine-translate it:

• Don’t simply use Google Translate on your OpenWrt Bypass Router Tutorial. You need someone who truly understands both the technology and the target language to do the localization

Localization means:
Replacing the UI screenshots in the article with versions in the corresponding language, using the customary terminology of the local tech scene, and even offering specific advice based on local users’ network environment (such as ISP characteristics)

Server location:

• Although CDNs have largely solved the problem of global access speed, hosting your server in the region where your main target users are is still a positive ranking signal

Google Search Console geolocation targeting:

• For sites using a gTLD (such as .com, .net, .org) with a subdirectory or subdomain setup, you can find International Targeting in GSC’s legacy tools to report
• You can explicitly tell Google: content under the /ja/ subdirectory is specifically aimed at Japanese users, which can compensate for the weakness of gTLD geolocation signals

International SEO is very precise: it requires you to make strategic decisions about URL architecture, strictly follow the hreflang specification in implementation, and genuinely consider local users at the content level. Done right, you can truly look global; done wrong, it can lead to a chaotic SEO disaster — for example, Google treating all your translated versions as plagiarized content and wiping them out

13.1 The three cores of local SEO#

13.3 On-site localization optimization#

You need to fill your site content with local signals too

Local landing pages

• If you offer services in multiple cities, you should create an independent landing page with unique content for each city

NOTE

EXAMPLE

chongxi.us/openwrt-support/Tokyo This page should include:

• The city name explicitly in the title and H1 (e.g. Tokyo OpenWrt on-site configuration service)
• A detailed description of your service features in the Tokyo area in the page body
• An embedded Google Map of the regional office
• Testimonials from local Tokyo customers.
• A local phone number

LocalBusiness Schema

• On your site’s contact page or footer, deploy LocalBusiness Schema to tell search engines your NAP, business hours and other information in a structured way

13.4 Local citations#

This means your business’s NAP information being mentioned anywhere on the internet, even without linking to your site

Sources are usually: local chamber of commerce sites, industry association directories, online maps (such as Amap, Baidu Maps), Dianping and so on
As with GBP, keep NAP information 100% consistent across all citation sources

I recommend aggressively searching for your own brand to check and correct inconsistent citations

14.1 Understand the characteristics of voice search queries#

1. Longer and more conversational: voice queries are longer on average and are usually complete questions
2. Clearer intent: usually seeking a direct, concise answer
3. Strongly local: many voice searches are about “nearby…”

14.2 Featured snippets#

A featured snippet is a special information box at the very top of Google search results that directly answers the user’s question. Voice assistants (such as Google Assistant) usually read out the featured snippet when answering questions. So winning the featured snippet is winning voice search

NOTE

EXAMPLE

1. Identify question-type queries: use GSC or third-party tools to find queries that already rank on the first page of your site and start with “how”, “what” or “why”
2. Provide concise, direct answers: in your article, right after the question heading (H2/H3), answer the question directly and clearly in one paragraph (usually between 40 and 60 words)

For example

H2: <h2>什么是OpenWrt的旁路由模式？</h2>
The paragraph right after it: <p>OpenWrt的旁路由模式，指的是在不改变现有主网络结构的前提下，将一台运行OpenWrt系统的设备作为旁路网关接入局域网。它不负责主要的路由和DHCP功能，而是专门处理特定任务，如广告过滤、科学上网或网络加速，实现对全屋设备的透明代理
This paragraph’s structure and length make it a perfect featured snippet candidate

3. Use ordered/unordered lists: for step-by-step or list-type questions, use <ol> or <ul> tags to organize the content
4. Use FAQPage Schema: deploying FAQPage Schema for Q&A content on a page can increase your chances of being selected as a featured snippet

14.3 Make content more conversational#

Writing style: when writing, imagine that you’re answering a friend’s question directly. Use more natural, colloquial language
Content structure: use Q&A structures more often to organize your articles

Of course, mind your audience: if your site doesn’t need conversational AI, there’s no need to do this part of the optimization

15.1 Google Search Console#

GSC is a free tool provided directly by Google, the starting point and endpoint of all SEO work. It tells you how Google views your site

15.2 Google Analytics 4#

GA4 tells you what users do after they enter your site. It helps you judge whether your content truly meets user needs

WARNING

Practical analysis I

Identifying content that gets applause but no engagement

1. Path: Reports > Engagement > Landing pages

2. This report lists the first page users land on when entering your site. By default it’s sorted by “Users”. Now click the “Average engagement time” column header to sort ascending

3. You find that your maimai DX Full Song List page brings in many users (near the top in user count), but its average engagement time is only 15 seconds

4. Diagnosis: users arrive, then leave immediately

This usually means:

• Your page loads too slowly
• The page design is confusing and users can’t find the information they want
• The content is comprehensive, but presented poorly (for example, it’s just a giant HTML table with no way to search or filter)

5. Optimize the page experience: add client-side search, sorting and filtering to this song list. Let users quickly find the songs they want by difficulty, version and composer (such as xi)

6. Improve loading speed: if the list contains lots of images (covers), make sure they’re all compressed and lazy-loaded

WARNING

Practical analysis II

Tracking conversions to find your cash cow

0. You need to set up Conversions in GA4 first. A conversion doesn’t have to mean a monetary transaction. For a blog, a conversion can be defined as “a user successfully copies a code block”, “a user downloads your OpenWrt config file” or “a user stays on a tutorial page for more than 5 minutes”

1. Path: Traffic acquisition > Traffic acquisition

2. Reading the data: look at the report, set the primary dimension to “Landing Page”, and check the “Conversions” each landing page generates

3. A case study: you find that your article The Top Ten Must-Have Apps After Installing Arch Linux leads by a wide margin on “code block copied” conversions, far ahead of other articles

4. Diagnosis: this article precisely captures users’ action intent. After reading it, users immediately start doing things. It’s a highly valuable action-guide homework-copying type of content

5. Around this successful pattern, create more similar content. For example The 5 Best Terminal Emulator Configurations on Arch Linux

6. Push this high-converting article to a more prominent spot on your site (such as a homepage recommendation), and link to it from more articles

16.1 meta name="keywords"#

This is a meta tag that used to live in the HTML <head>, allowing site owners to proactively tell search engines which keywords a page is related to

1
        <meta name="keywords" content="舞萌DX, maimai DX, 谱面, 攻略, xi, 白系">

In the wild early days of search engines, algorithms couldn’t accurately determine a page’s topic by analyzing its content. So they needed site owners to provide this kind of self-introduction

But this tag had a fatal design flaw: it relied entirely on the site owner’s integrity. Soon black-hat SEOs began abusing it wildly, stuffing it with large numbers of popular but irrelevant keywords to steal traffic (for example, adding free movie downloads to an Arch Linux tutorial page)

This behavior severely polluted search results, causing user experience to plummet. To save themselves, search engines were forced into an algorithmic revolution, learning to rely on their own ability to understand content instead of trusting the self-reports of sites

This is a classic case of a defense against the honest but not the dishonest

WARNING

Death sentence

Google publicly announced back in 2009 that it completely ignores the meta name="keywords" tag in its ranking algorithm

Bing and other mainstream search engines have also made clear that this tag has almost no value

• For new site owners, there is absolutely no need to use this tag
• For veteran site owners, if your site still has this tag, there’s no need to waste time deleting it. It’s harmless and basically useless. Just let it lie there quietly as a fossil of internet history

This is why I haven’t mentioned meta keywords at all throughout this article. If a 2025 SEO tutorial is still teaching you how to optimize meta keywords, you can laugh with relief; the author’s knowledge system is at least 15 years behind

16.2 Traditional PageRank scores#

This is an integer score from 0 to 10 that used to be displayed in the Google Toolbar, representing a page’s link authority

For more than a decade, the PR value was the most intuitive, most quantifiable authority metric in the entire SEO industry, directly spawning a huge link trading market

Why did it die? (We broke this down in detail in section 3.3.) Its publicness led to large-scale algorithmic manipulation, destroying the fairness of the search ecosystem

WARNING

Death sentence

In 2013, PageRank was hidden In 2016, it was completely removed by Google

Any third-party tool claiming to check your site’s “PR value” is displaying meaningless, outdated data, or a metric they invented themselves that has nothing to do with Google (such as Ahrefs’ DR or Moz’s DA)

Forget the concept of PR value. Stop asking me what my site’s PR value is
Put your energy into acquiring topically relevant, contextually appropriate, genuinely traffic-driving high-quality links, rather than chasing a phantom score that no longer exists. That is the modern Rank

17.1 Paid backlinks#

Scammer scripts usually go: “We have thousands of high-authority site resources and can publish articles with links for you” “For just XXX yuan, you get a dofollow backlink from a DA 50+ site (or another made-up metric)”

Deconstructed, this is:

1. Private Blog Network: the most common kind. Scammers buy a bunch of expired domains that once had some authority, then use templated programs to quickly build lots of blogs that look “normal.” These blogs exist for one purpose only — selling links. Their content is low-quality, their topics are scrambled, and they link to each other, forming a fake “authority network”

2. Hacked sites: even worse, they exploit vulnerabilities to hack into normal sites, then secretly add links pointing to the client on those sites

3. Low-quality guest posts: they publish an AI-generated, badly written spam article on low-quality news sites or blogs that let anyone pay to post, and insert your link into it

NOTE

How Google sends them to the gallows

Google’s anti-spam team has extremely complex algorithms and manual review processes to identify these unnatural links

1. Link Graph Analysis: Google can analyze the link topology of the entire internet. Sites in a PBN network usually have highly unnatural link patterns (e.g. only outgoing links and no incoming links, or linking only within a specific small circle), and are easily identified by algorithms as link farms

2. Anchor text abuse detection: if a new site suddenly gains a large number of backlinks pointing to it with completely identical anchor text (e.g. all “OpenWrt router recommendation”) within a short period, that’s an extremely dangerous signal. Natural links should have diverse anchor text

3. Site quality assessment: Google assesses the quality of the site publishing the link itself. A normal site should have topical content, regular updates and real user visits. PBN sites usually have jumbled content, stagnated updates and no organic traffic

4. Manual reports and review: your competitors, or anyone who understands SEO, can report suspicious link trading to Google through its “Report paid links” tool

• Once Google’s algorithm or a manual reviewer confirms that your site has participated in large-scale link manipulation, your Google Search Console will receive an “Unnatural outbound links” manual action notice
• Your site’s rankings will fall off a cliff, and may even be completely removed from Google’s index
• You’ll need to manually find all the spam links you bought, contact the site owners to request removal (which is usually ignored or met with blackmail), then use GSC’s “Disavow Tool” to submit to Google a list of spam links you disown, and write a Reconsideration Request begging for Google’s forgiveness. The whole process can take months, and success is not guaranteed

Site owners would do best to let go of speculative ideas too; good wine needs no bush. Looking back at this article, real authority comes from creating linkable assets worth linking to, and through digital PR and community contributions, letting high-quality links happen naturally. It’s hard and slow, but it’s the only correct, sustainable path

17.2 Keyword stuffing#

Scammer script: “Your page’s keyword density isn’t enough; we need to optimize it to 5% for you”

As we said in chapter 16, this is a Stone Age idea. Modern NLP algorithms will consider it as reducing content readability, and may even give negative evaluations

17.3 “Submit to XXX search engines” services#

Scammer script: “We can submit your site to 500 search engines worldwide in one go”

99.99% of search traffic comes from a few giants: Google, Bing, Baidu and so on. These giants don’t need you to submit manually; their crawlers discover your site automatically. What this service submits to is hundreds of garbage search engines you’ve never heard of with zero traffic: pure psychological comfort, completely worthless

17.4 Buying social media signals#

Scammer script: “Buying 10,000 Twitter retweets will boost your SEO rankings”

Google has made clear that the number of likes and shares on social media is not a direct ranking factor. Although widespread social sharing may bring indirect benefits (such as brand exposure and referral traffic, which may in turn bring organic backlinks), buying fake signals from bot accounts is completely ineffective and will be easily seen through by algorithms