Chapter 22: How Links Work — PageRank, Authority, and Why Links Are Still the Strongest Ranking Signal

"The citation (link) graph of the web is an important resource that has largely gone unused in existing web search engines." — Sergey Brin and Lawrence Page, The Anatomy of a Large-Scale Hypertextual Web Search Engine (1998)

Overview

Return, one more time, to the page stuck at #11 — Rivertown's guide to replacing a home water heater. We have already done real work on it. In Chapter 3 we diagnosed that its title answered the wrong question; in Chapter 9 we rewrote that title to match what searchers actually type; in Chapter 15 we linked to it from the plumbing hub and a cluster of related articles, so authority could finally flow to it from inside the site. Those were the right moves, and the page should have improved. And yet, for the competitive head term, it may still not reach the top of page one. Why?

Because we have not yet touched the one lever Chapter 15 deliberately left open. Look again at the honest gap analysis from Chapter 1: the result that outranks Rivertown is "referenced by local news and home-improvement sites," while Rivertown has "few external links to the site." We fixed how authority moves within the site. We have not addressed how much authority the site has to move in the first place — and that comes from other websites vouching for it. That vouch, on the open web, is a link. This chapter is about the signal those links carry: where it came from, how it works, and why, after a quarter-century of algorithm change, it remains the strongest signal Google has for judging a page's standing off the page itself.

Here is the real question this chapter answers: why should Google trust a page it has never seen a human vouch for — and how does one site vouch for another in a way a machine can count? The answer is the idea that built Google, and it is worth understanding at its root, because almost everything that goes wrong in "link building" goes wrong because someone learned the tactic without understanding the mechanism.

In this chapter, you will learn to:

• Explain PageRank — links as votes, weighted by the authority of the voter — and read its one honest formula for meaning, not computation.
• Trace how link-based ranking evolved past naive counting: the arms race, nofollow, Penguin, and machine learning.
• Judge a link's value along four dimensions: authority, relevance, anchor text, and placement.
• Tell apart the follow, nofollow, sponsored, and ugc attributes and what each signals.
• See the web as a link graph and understand how authority flows across it and into your pages.
• Recognize why buying links is a detectable, losing bet — and why Domain Authority and Domain Rating are useful estimates, not Google's verdict.

Learning Paths

Links matter to everyone, but the angle differs. 📊 Strategist and 📝 Content Creator: §22.1–22.3 and §22.7 are your core — the mechanism, what makes a link valuable, and how to read authority metrics honestly. 🏪 Local Business (Rivertown's path): weight §22.3 (relevance and local links), §22.6 (why the old "SEO guy" was a liability), and the Strategy File — your links are fewer but more winnable. 🛒 E-Commerce: §22.3 and §22.5 — authority flow into category pages is the whole game at scale. 🔧 Developer: §22.4 (the rel attributes, in the markup) and §22.5 (the graph) are yours; you implement the signals everyone else only talks about.

22.1 The original PageRank insight: a link is a vote

To understand why links matter, go back to the problem they solved. In the mid-1990s, search engines ranked pages mostly by their text: how many times the query words appeared, whether they were in the title, how prominent they were on the page. This worked until people noticed how it worked. If repeating a word ranked you, you repeated the word — a hundred times, in white text on a white background if necessary. The result was a search experience drowning in keyword-stuffed junk, because the ranking signal was entirely under the control of the person trying to rank. A page could say it was the best result, and there was no independent check.

Two Stanford graduate students, Larry Page and Sergey Brin, had a different idea, borrowed from an old practice in academia: citation analysis. In scholarship, you can estimate how important a paper is by counting how many other papers cite it — and, crucially, by weighing who cites it. A thousand citations from obscure, ignored papers mean less than a handful from the field's landmark works. Page and Brin saw that the web already had citations built into it: links. When one page links to another, it is, in effect, casting a vote — saying "this is worth looking at." And the web contained billions of these votes, made by millions of independent people, most of whom had never heard of each other. That is a signal you cannot fake by editing your own page, because it lives on everyone else's pages. This is the origin of PageRank, named (with a pun) after Larry Page: a way of scoring every page on the web by the quantity and quality of the links pointing at it.

The genius was the "quality" part. PageRank is recursive: a page's score depends not just on how many pages link to it, but on the scores of those pages. A link from a page that is itself heavily linked-to counts for more than a link from a page nobody points at. Importance flows through the network. A vote from an important page is a loud vote; a vote from a nobody is a whisper.

Here is the original formula, from the 1998 work. Do not be intimidated by it — we are going to read it, not solve it.

$$PR(A) = (1 - d) + d\left(\frac{PR(T_1)}{C(T_1)} + \frac{PR(T_2)}{C(T_2)} + \cdots + \frac{PR(T_n)}{C(T_n)}\right)$$

Read it left to right in plain English. $PR(A)$ is the PageRank of your page, A — the number we want. $T_1$ through $T_n$ are the pages that link to A. For each of those linking pages, we take its PageRank, $PR(T_i)$, and divide it by $C(T_i)$ — the number of links out of that page. That division is the most important idea in the whole equation: a page splits its vote among all the pages it links to. If an authoritative page links only to you, you get its full weight. If it links to five hundred pages, your share is one five-hundredth. A link from a strong page that links to almost nothing is a concentrated endorsement; a link from that same page buried among five hundred others is diluted almost to nothing. Then we sum all those incoming shares, multiply by $d$ — the damping factor, conventionally around 0.85 — and add the leftover $(1 - d)$.

That damping factor has a beautiful intuition behind it, the random surfer model. Imagine a person who starts on some page and keeps clicking links at random. About 85% of the time (that is $d$) they click another link on the current page; about 15% of the time they get bored and jump to a completely random page (that is the $(1-d)$ term). PageRank is, essentially, the probability that this endlessly wandering random surfer lands on your page. Pages that many roads lead to — especially roads from other well-traveled pages — get visited often, and score high. The 15% random jump keeps the math from getting stuck in loops and stops a cluster of pages that only link to each other from hoarding all the importance.

🔎 How Search Sees It The recursive nature is what makes PageRank hard to game and expensive to compute. Because your score depends on your linkers' scores, which depend on their linkers' scores, you cannot calculate it in one pass — Google computes it iteratively, guessing all the scores, then refining them over and over across the entire link graph until the numbers stop changing much. The practical upshot for you: you cannot inflate your own PageRank from your own site. It is conferred by others, weighted by their standing, all the way up. That is exactly why it was such a durable signal — and exactly why the entire manipulation industry that followed had to go get links from other people's sites, which is a much harder crime to commit invisibly.

Let us put a number on the intuition, clearly labeled as illustrative — the absolute values are made up; only the ratio teaches. Suppose an authoritative industry page carries a PageRank of 100 units and links out to just 4 pages. Each of those links passes 100 / 4 = 25 units of endorsement. Now suppose a weak, ignored page carries a PageRank of 4 units and links out to 40 pages; each of its links passes 4 / 40 = 0.1 units. The link from the strong, selective page is worth roughly 250 times the link from the weak, spraying one — before we even ask whether the two pages are relevant to your topic. If your page A were linked only by that one authoritative page, its score would land near 0.15 + 0.85 × 25 ≈ 21 units. This is why one great link can outweigh a thousand junk ones, a theme we will hammer for the rest of Part IV.

What PageRank can do: give every page a query-independent measure of importance derived from the independent judgment of the whole web — a signal that is genuinely hard to fake from your own property. What PageRank cannot do: tell whether your page is relevant to a particular search. PageRank is about importance, not aboutness; a hugely authoritative page about gardening should not win the query "furnace won't ignite" no matter how many links it has. Relevance (Chapters 3 and 4) and importance (this chapter) are different axes, and ranking needs both. And PageRank, in its raw 1998 form, had one fatal practical flaw: the moment it became known that links determined rankings, links became something to manufacture. That is the story of §22.2.

⚖️ Evidence Check Claim: "Links are the strongest off-page ranking signal." Where does that sit on our honesty scale? — Confirmed by Google (with caveats): Google has stated many times that links remain an important ranking signal; in a 2016 public Q&A a Google search-quality representative named content and links among the most important signals. Google's engineers have also publicly confirmed that a form of PageRank is still used internally, "among many, many other signals." — Strong correlation, not proven causal: large-scale studies from SEO tool vendors consistently find that pages and sites with more referring domains tend to rank higher. This is a correlation — better sites also tend to attract more links — and it is repeated so often it has hardened into "fact." We report the direction, not a fabricated percentage. — The author's professional experience: earning quality links reliably moves rankings for competitive terms in a way little else does; it is slow and it is real. — Speculation: the exact weight of links versus content versus everything else. Google has never published it, it changes, and anyone who quotes you a precise number is selling something.

22.2 How link-based ranking evolved beyond raw PageRank

If you understand §22.1, you can predict the next twenty years of SEO history, because it follows mechanically from one fact: the instant links determined rankings, links became a currency. And wherever there is a currency, there is counterfeiting.

The first counterfeit era was crude and effective. If a link was a vote, you simply manufactured votes: you spammed the same link into thousands of blog comments and forum signatures; you built "link farms," networks of empty sites whose only purpose was to link to each other and to paying customers; you traded links with anyone ("I'll link to you if you link to me"); you submitted your site to hundreds of low-value directories; you bought links outright from anyone with a website and a price. For a while, this worked, because early PageRank was relatively naive — it counted a link as a link. The web filled up with garbage engineered to rank, and Google, whose entire business depended on results being good, had to respond. The evolution of link-based ranking is the story of that response.

The first countermeasure was structural. In 2005, Google — with the other major engines — introduced the nofollow attribute (we cover it fully in §22.4), a way to mark a link as "don't count this as my endorsement," aimed initially at the epidemic of comment spam. But attributes only help where site owners apply them. The deeper shift was Google getting better at judging links itself: learning to recognize the fingerprints of manipulation and to simply discount the links that carried them.

That effort has a name you already met in Chapter 6: Penguin. Launched in 2012, the Penguin system specifically targeted manipulative link profiles — unnatural anchor text, link networks, paid links, the whole counterfeit economy. Sites that had climbed on bought and spammed links saw those links stop working, or worse. Over time Penguin's approach matured from penalizing sites for bad links toward mostly devaluing the bad links — ignoring them, so they simply count for nothing — which is closer to how Google handles link spam today through its machine-learning spam systems.

🔗 Connection The named update that reset link building is Penguin (2012), introduced in Chapter 6 (Google Updates). The modern machinery for detecting spam at scale — including link spam — is Google's SpamBrain system, also from Chapter 6. When links become an outright liability and you must decide whether to clean them up, that is Chapter 26 (Toxic Links, Disavow, and Manual Actions). This chapter explains the mechanism; those chapters handle the fallout.

Two more evolutions matter for how you think about links today. First, Google moved from treating all links on a page as equal toward weighing them by where they are and how likely they are to be used. A Google patent describes a "reasonable surfer" refinement of the original random-surfer model: instead of assuming a surfer clicks every link with equal probability, it assumes they are more likely to click a prominent, in-content link than a link buried in a footer or a site-wide boilerplate menu. A patent tells you what Google was thinking about, not what it definitely runs today — that is a Tier-2 signal, not a confirmed fact — but it matches everything practitioners observe: a link's placement changes its value (§22.3).

Second, and most importantly, links stopped being a standalone scoring system and became one input among hundreds, combined by machine-learned ranking systems (RankBrain, BERT, and successors from Chapter 2) that weigh signals contextually. This is why nobody, inside or outside Google, can give you "the value of a link" as a fixed number. The same link may help one page and barely move another, depending on the query, the competition, and every other signal in play.

🚫 SEO Myth: "PageRank is dead." You will hear this constantly, usually as a reason to stop caring about links. It is a confusion between a number you could see and a signal Google uses. What died is the public Toolbar PageRank — the little green 0-to-10 bar Google showed in a browser toolbar, which it stopped updating around 2013 and removed entirely in 2016. What did not die is PageRank the concept: Google engineers have publicly confirmed the company still uses PageRank internally as one of many signals. So "PageRank is dead" is true about the visible score and false about the underlying mechanism. Links still carry authority; you just can no longer read a tidy Google-blessed number for it — which, as §22.7 argues, is probably healthy.

What this evolution means for you: the naive game — accumulate as many links as possible — has been a losing game for over a decade. Google no longer counts links so much as it weighs them, discounts the obviously manipulated ones, and folds the rest into a much larger judgment. A thousand junk links can be worth literally nothing (devalued) or worse than nothing (a liability). The modern lever is not quantity; it is earning a smaller number of links that are genuinely good — which is precisely what §22.3 defines.

22.3 What makes a link valuable: authority, relevance, anchor text, placement

Not all links are created equal — that is the entire lesson of the last two sections, made practical. Before we grade links, two definitions you will use for the rest of the book.

A backlink (or inbound link) is any link from another website pointing to a page on yours. A referring domain is a unique website that links to you, counted once no matter how many individual backlinks it sends. The distinction is not pedantic; it is one of the most useful triage habits in link analysis. Ten thousand backlinks from a single spammy site is one referring domain and almost certainly worthless. Ten backlinks from ten different respected, relevant sites is ten referring domains and can be transformative. When someone brags about "50,000 backlinks," ask how many referring domains — the honest number is usually a small fraction, and the small number is the one that matters.

With that, here are the four dimensions along which a link's value is decided.

1. Authority — how much standing the linking page and site have. This is PageRank's core: a link from a site the whole web trusts (a major news outlet, a respected trade association, a well-known publication in your field) passes far more weight than a link from a brand-new blog nobody reads. Authority is not a single public number (§22.7), but the intuition is sound: whose vote is this?

2. Relevance — how topically related the linking source is to your content. Post-Penguin, this matters more than it used to. A link to Rivertown from a regional home-and-garden publication, an HVAC trade body, or a local contractors' association is worth more than a link from an unrelated site about, say, cryptocurrency — even if the crypto site has higher raw authority. A relevant link is a more credible vote, and an irrelevant one (especially in bulk) is a classic manipulation fingerprint. Relevance connects to the entity and topical-authority ideas from Chapter 4: Google is trying to understand who is an authority on what, and a link from within your topic is a stronger statement than a link from nowhere in particular.

3. Anchor text — the visible, clickable words of the link. Define this cleanly, because it is one of the terms this chapter owns. Anchor text is the text a link is wrapped around: in "read our furnace repair guide," the anchor text is "furnace repair guide." Anchor text is a relevance signal — it gives Google a human-written hint about what the linked page is about. Historically it was extremely powerful, so powerful it produced "Google bombing," where activists pointed thousands of links with a chosen phrase at a target page to make it rank for that phrase as a prank. That power made anchor text a prime manipulation target — buy a hundred links all reading "best HVAC repair Rivertown" and you have signed a confession. So Google grew cautious: a natural link profile has varied anchors (your brand name, your bare URL, "click here," the article's title, a descriptive phrase), and an unnaturally repetitive, exact-match commercial anchor profile is a red flag.

🔗 Connection You rarely control the anchor text of a link someone freely gives you — and that is healthy. The strategy of anchor-text distribution, and the over-optimization risk that comes from trying to control it, belongs to Chapter 24 (Link Building Tactics), §24.7. Here we only need the definition and the principle: natural, varied anchors good; engineered, repetitive money-phrase anchors dangerous.

4. Placement — where on the page the link sits, and whether a human chose to put it there. A link inside the main body of an article, surrounded by relevant text, placed because the author genuinely wanted to point readers to your resource, is the gold standard. A link in a footer, a sidebar widget, a sponsored author bio, or a site-wide template is worth far less — it is boilerplate, not an endorsement of this specific page. This is the "reasonable surfer" intuition from §22.2 in action.

Placement leads to the single most important quality distinction of all: the editorial link. An editorial link is one that a publisher chose to give because your content earned it — no payment, no exchange, no request that crossed a line. It is the kind of link PageRank was designed to count, and it is the only kind that is durably safe. Its opposite is the self-placed or acquired link — one you put there yourself (a forum signature, a directory submission, a comment) or paid or bartered for. Google's entire link-quality effort is, at bottom, an attempt to tell these two apart and count only the first.

📄 Read the Report

text FIGURE 22.1 — "Two links to the same page, graded" [constructed teaching example] THE PAGE Rivertown's water-heater replacement guide — the page stuck at #11. WHAT'S THERE LINK A: the regional newspaper's home-improvement columnist links to the guide from inside an article titled "When to repair vs. replace your water heater," anchor text "a local plumber's replacement checklist." One in-content link, freely given. LINK B: a batch of 60 backlinks from an overseas "web directory" network, all with the anchor "water heater replacement Rivertown cheap," all in footer link lists. WHAT IT SHOWS Link A is one referring domain but scores on every dimension: authority (a real local publication), relevance (home improvement), natural anchor, in-content placement, editorial. It is worth far more than Link B's 60. WHAT IT DOESN'T Link A does not, by itself, guarantee a #1 ranking — it is one vote among the whole contest. And we cannot read its exact "value"; there is no such number. THE MOVE Pursue Link-A-shaped opportunities (Chapters 23–24). Treat Link-B-shaped links as the liability they are (§22.6; cleanup in Chapter 26). THE LESSON Grade links by authority × relevance × anchor × placement × how they were acquired — never by raw count. One earned editorial link can outweigh sixty manufactured ones.

What a great link can do: meaningfully improve a page's odds in a competitive contest, and — because authority flows (§22.5) — lift the standing of the whole site a little. What it cannot do: guarantee a position, or work in isolation from relevance and content. A page has to deserve to rank on its merits first; links are how the web confirms that it does.

22.4 follow, nofollow, sponsored, and UGC: the link attributes

Not every link is meant as an endorsement, and the web needed a way to say so. That way is a small piece of markup — a rel attribute on the link — and understanding the four cases is essential, because they are how you both read other people's signals and correctly label your own.

A normal link with no special attribute is often called a follow or "dofollow" link (the word "dofollow" is informal — there is no such attribute; it just means "an ordinary link"). This is the default: it is crawlable and, in principle, passes authority. It is a full-strength vote.

The nofollow attribute, introduced in 2005, was the web's first way to say "I am linking to this, but do not treat it as my endorsement, and do not pass my authority to it." It was created to fight comment and forum spam — so that a spammer dumping links in your comments section gained nothing. Over the years it spread far beyond that: sites apply nofollow to paid links, to links they can't vouch for, and often to all outbound links as a blanket policy.

In 2019, Google refined this into three attributes and made an important conceptual change. The two new ones:

• sponsored — for links that are advertisements, paid placements, or affiliate links. This is how you correctly disclose a paid link (more on why that matters in §22.6).
• ugc — for user-generated content: links in comments, forum posts, user profiles — content your users created, not you.

<!-- An ordinary editorial link — a full endorsement -->
<a href="https://example.com/guide">a plumber's replacement checklist</a>

<!-- A paid or affiliate link — disclose it as sponsored -->
<a href="https://shop.example.com/heater" rel="sponsored">this water heater</a>

<!-- A link inside a user comment or forum post -->
<a href="https://example.com" rel="ugc">check out this site</a>

<!-- "I link to it but don't vouch for it" -->
<a href="https://example.com" rel="nofollow">an unverified source</a>

Here is the conceptual change, and it surprises people. As of March 2020, Google treats nofollow, sponsored, and ugc as hints, not directives. For years, nofollow was an absolute instruction: Google would not follow the link or pass any signal, period. Now Google says it may use these links for crawling and ranking purposes if it chooses, treating the attribute as a strong suggestion rather than an iron rule. In practice, the safe way to read all three is the same: they are ways of saying "this link is not my plain editorial endorsement," and you should use them honestly to label your own links.

🔎 How Search Sees It Why move from directive to hint? Because Google wants to understand the web, and sometimes a nofollow link is still informative — a widely-cited source that everyone happens to nofollow is still, obviously, an important source. By reserving the right to interpret the attribute, Google keeps the information while respecting the signal. For you, the operational rule is unchanged: do not rely on nofollow to pass or block authority precisely. It never was a scalpel, and it certainly isn't now.

That last point buries an old myth. Around 2009, some SEOs used nofollow for "PageRank sculpting" — nofollow-ing internal links to unimportant pages (a login, a cart) so that "more" PageRank would flow to the important ones. Google's Matt Cutts publicly explained that this no longer works: the PageRank that would have flowed through a nofollowed link doesn't get redistributed to the other links — it simply evaporates. Sculpting your internal authority with nofollow is a dead tactic; you shape internal flow with your architecture and which pages you link (Chapter 15), not with attributes.

🚫 SEO Myth: "nofollow links are worthless — don't bother getting them." Wrong, on several counts. First, nofollow links from big, trusted places (major publications, Wikipedia, many social platforms) drive real referral traffic and real brand awareness — value that has nothing to do with PageRank. Second, since 2020 Google may treat them as ranking hints anyway. Third, and most subtly, a natural link profile contains plenty of nofollow links — that is what an organically-linked site looks like. A profile that is 100% follow links is itself a manipulation fingerprint, because real mentions on the real web are a messy mix. Chasing only "follow" links is both a misunderstanding of value and a way to make your profile look engineered. Earn mentions from good places; let the attributes fall where they may.

🔄 Check Your Understanding Rivertown pays a regional podcast to mention its services, and the show links to Rivertown's site from its episode notes. Which attribute should that link carry, and why does labeling it correctly protect Rivertown?

Answer

It is a paid placement, so the correct attribute is rel="sponsored" (a nofollow would also communicate "not an editorial endorsement," but sponsored is the precise label for a paid link). Labeling it correctly matters because Google's guidelines require that links which are paid for not pass PageRank; a paid follow link that passes authority is a link scheme (§22.6), and disclosing it as sponsored is exactly how Rivertown stays on the right side of the line. The responsibility to label is on both parties, but the smart brand insists on it.

22.5 The link graph and how authority flows

Step back and picture the whole thing at once. Model the entire web as a link graph: every page (or site) is a node, and every link is a directed edge — an arrow pointing from the linking page to the linked one. Google builds and continuously updates a colossal map of this graph — recall from Chapter 1 that following links is how Googlebot discovers pages in the first place — and PageRank and its successors are computed over this map. When we talk about "authority," we are talking about a property of your position in this graph: how many arrows point at you, from where, and how much authority those sources have to give.

HOW AUTHORITY FLOWS THROUGH THE LINK GRAPH                   [schematic — not to scale]

  EXTERNAL WEB                        YOUR SITE
  ┌───────────────┐                   ┌──────────────────────────────────────┐
  │ Local news    │──(editorial)────▶ │  HOMEPAGE  ●───────┐                  │
  │ (high auth.)  │                   │     │  (internal links distribute      │
  └───────────────┘                   │     ▼   authority inward — Ch 15)      │
  ┌───────────────┐                   │  Plumbing HUB ●──┬──▶ ● Water-heater   │
  │ HVAC trade    │──(relevant)─────▶ │     │            └──▶ ● Drain-cleaning │
  │ association   │                   │     ▼                                  │
  └───────────────┘                   │  Service pages ● ● ●                   │
  ┌───────────────┐                   │                                        │
  │ Random blog   │┄┄(nofollow)┄┄┄▶  │  (a nofollow arrow carries little to   │
  │ (low auth.)   │                   │   no authority — dashed)               │
  └───────────────┘                   └──────────────────────────────────────┘
     ▲  authority ENTERS the site        ▲  and is then MOVED AROUND inside it
     └── mostly from external links       └── by internal links (architecture)

Two flows are happening, and keeping them distinct dissolves most confusion about links. External links bring authority into your site — usually landing on your homepage or whichever pages other people naturally link to. Then internal links distribute that authority around your site, from the pages that have it to the pages that need it. This second flow is what Chapter 15 called link equity and taught you to direct with hub-and-spoke architecture and strategic internal linking. The two are partners: external links fill the reservoir; internal links pipe the water where it is needed.

🔗 Connection The internal half of this picture — how authority moves between your own pages, the 3-click reachability rule, hubs and spokes, orphan pages — is the canonical subject of Chapter 15 (Site Architecture), especially §15.4–15.5. This chapter owns the external half: where the authority comes from in the first place. Together they explain the page stuck at #11 completely.

Now we can finally close the loop on that page. In Chapter 15, we gave the water-heater guide internal links from the plumbing hub — we built the pipes. But internal linking can only distribute the authority the site has. If the whole Rivertown site holds little external authority, then piping a small amount of it to the water-heater page still leaves that page under-powered against a competitor whose site is "referenced by local news and home-improvement sites." Architecture directs authority; it cannot manufacture it. The lever Chapter 15 left open — and the reason this page may still sit near the top of page two after every on-page and internal fix — is that the site needs more genuine external authority to distribute in the first place. That is the work of Chapters 23 and 24. The #11 page is not, in the end, an on-page problem or even an internal-linking problem; its last mile is an authority problem.

A few honest clarifications about how the flow behaves, because myths cluster here:

• Linking out does not "leak away" your rankings. Some people hoard links, terrified of passing authority to others. Linking to good, relevant sources is normal, useful for readers, and a signal of a genuine resource; the fear of "link leakage" is mostly folklore. You are not spending a finite battery.
• A link's authority is shared, not copied. Recall the $C(T_i)$ divisor from §22.1: a page passes its authority split among its outbound links. This is why a link from a page with few outbound links is worth more — not because linking out is bad, but because the vote is less divided.
• Authority is a flow at query time, not a stamp on your forehead. There is no single stored "site authority score" Google shows you (§22.7). There is a position in a graph, recomputed constantly.

What the graph view can do: make you think correctly about links — as a network of endorsements you sit inside, where both who points at you and how you route their authority internally matter. What it cannot do: give you a dial to turn. You influence your position in the graph only by earning real edges from real sites and arranging your own pages sensibly. There is no back door.

22.6 Why buying links is risky — and detectable

We now arrive at the tactic that tempts every business that understands how valuable links are and how slow they are to earn: why not just buy them? We cover this the way a fraud chapter in a finance textbook covers embezzlement — from the investigator's seat, so you can recognize it, understand why it fails, and never build your business on it. This is understand-and-avoid, not a how-to.

Start with the rule. Google's link-spam policy is explicit: buying or selling links that pass PageRank — or exchanging goods, services, or money for links meant to manipulate ranking — violates Google's guidelines. (This is why the sponsored attribute from §22.4 exists: a paid link is allowed as long as it is disclosed and does not pass PageRank. A paid link that passes authority is the violation.) The reason is obvious once you understand PageRank: the whole signal works because links are supposed to be independent editorial judgments. A bought link is a forged vote, and forged votes, at scale, would destroy the signal's value for everyone — including Google, whose results would degrade.

Why is it detectable? Because manipulation leaves fingerprints, and Google has spent two decades learning to read them:

• Unnatural velocity. A site that earns a handful of links a month suddenly gaining hundreds is a spike that stands out against every normal growth curve.
• Footprints. Paid-link sellers and link networks reuse infrastructure — the same sites linking to unrelated businesses, shared hosting and ownership, identical templates, the same "write for us" boilerplate. One customer's footprint exposes them all.
• Irrelevant sources in bulk. A local HVAC company suddenly linked from gambling, pharmaceutical, and overseas "general" sites is not a natural pattern; relevance (§22.3) is missing.
• Over-optimized anchors. Dozens of links all reading the exact commercial phrase you want to rank for is the confession from §22.3.
• Known marketplaces. Google actively investigates public and private link-selling networks; sites caught selling links have themselves been penalized (they lose the ability to pass PageRank).

Google applies two very different remedies, and the difference matters. The gentler one is algorithmic devaluation: the manipulative links simply stop counting — Google ignores them. You wasted the money, but your site is not otherwise harmed. The harsher one is a manual action: a human reviewer on Google's web spam team examines the site, confirms a link scheme, and applies a penalty that can crater a site's rankings overnight. (Manual actions, how to find them in Search Console, and how to recover are the subject of Chapter 26.)

🔎 How Search Sees It The most important thing to understand about detection is that it works retroactively. You may buy links and rise, and for a while nothing happens — which is exactly what makes it seductive and what sinks people. Google's spam systems and Penguin refreshes reassess link profiles continuously; a scheme that "works" today can be devalued or actioned in the next reassessment, months later, taking your rankings with it. You are not getting away with it; you are accumulating a liability with a delayed fuse. And when it goes off, you own a cleanup project (Chapter 26) on top of the lost rankings.

Two real, public cases make this concrete, and both are Tier-1 documented — no invented figures.

JCPenney, 2011. A New York Times investigation ("The Dirty Little Secrets of Search") revealed that the retailer was ranking #1 for an enormous range of lucrative terms — "dresses," "bedding," "area rugs" — on the back of thousands of paid links from unrelated sites. When Google was shown the evidence, it took manual action, and JCPenney's rankings for those terms fell sharply. A brand large enough to know better learned that scale does not buy immunity; if anything, it buys scrutiny.

Interflora, 2013. The UK flower-delivery brand was widely reported to have been penalized after a campaign of paid advertorial links in regional newspapers. Its visibility dropped, and it recovered only after a cleanup and a reconsideration process. The lesson repeated: even "respectable" paid links, dressed as editorial, are still paid links, and Google treats them accordingly.

⚖️ Evidence Check Claim: "Buying links is detectable and risky." — Confirmed by Google: the link-spam policy explicitly prohibits link schemes, and Google publicly documents both algorithmic and manual enforcement against them. Tier-1 public record: the JCPenney and Interflora cases are documented in major press and were confirmed as enforcement events. The honest limit: we cannot quote you a probability of getting caught or a timeline, and some manipulation undoubtedly goes undetected for a while — but "sometimes undetected for a while" is a description of a delayed liability, not a safe strategy. The expected value is negative, and the downside is catastrophic and slow to repair.

What buying links can do: produce a short-term rise, waste real money, and create a hidden liability that requires a painful cleanup when it is caught. What it cannot do: build durable authority. This is theme 5 of the whole book, stated at its sharpest: the only links that last are the ones you earn. Which is exactly why Chapters 23 and 24 exist — and exactly why Rivertown's inherited bought links are a problem we must now put on the table.

22.7 Third-party "authority" metrics: DA and DR are not Google

If Google won't give you a public authority score (§22.2's dead Toolbar PageRank), the market will sell you one. Several SEO tool vendors publish their own site-level "authority" metrics, and you must understand exactly what they are before you let one anywhere near your goals or your reports.

The best known is Domain Authority (DA), a metric from the tool vendor Moz: a proprietary score, usually on a 0-to-100 scale, that predicts how well a site might rank based on its link profile. Its close cousins are Ahrefs' Domain Rating (DR), Semrush's Authority Score, and Majestic's Trust Flow / Citation Flow. Each vendor crawls the web, builds its own link graph, and runs its own model. These are useful inventions. But here is the fact that must anchor everything you do with them:

These are third-party estimates. None of them is a Google metric, and Google does not use any of them to rank your site.

🚫 SEO Myth: "Raise your Domain Authority and you'll rank higher." This confuses a thermometer with the weather. DA and DR are outside estimates of ranking ability built from link data; they are not inputs Google reads. Google's own John Mueller has stated plainly and repeatedly that Google does not have a single "domain authority" or website-authority score that it uses in ranking. Moz itself says DA is its own metric, designed to predict ranking, not to influence it. So "improving your DA" changes a vendor's estimate of you; it does nothing directly to your Google rankings. Worse, because these scores are driven by links, chasing the number invites exactly the link-buying behavior §22.6 warned against — optimizing the thermometer by holding a match under it. When a measure becomes a target, it stops being a good measure.

So are they useless? No — and this is where honesty cuts both ways. Used as relative comparison tools, DA and DR are genuinely helpful:

• Triage of link prospects. Comparing two potential link sources, the one with higher DR (and real, relevant traffic) is usually the better bet. It is a fast filter, not a verdict.
• Spotting obvious junk. A "site" with a DR near zero and thousands of outbound links is a fair thing to distrust.
• Tracking trajectory. Watching your own referring-domain count and estimated authority trend upward over time, relative to named competitors, is a reasonable directional signal.

The danger is only in treating a third-party estimate as Google's opinion or as a KPI to maximize. It is neither. (We return to DA-as-vanity-metric when we separate real KPIs from vanity numbers in Chapter 29.)

🛠️ Try It on Your Site Skip the third-party estimate for a moment and look at Google's own view of your links — for free. Open Google Search Console and go to the Links report. You will see your top linking sites (your referring domains, in Google's own data), your most-linked pages, and your most common anchor text. This is not an estimate from a vendor's crawl; it is what Google actually knows about who links to you. Two questions to ask: Do the top linking sites look like real, relevant places or like junk? Does the top anchor text read naturally (your brand, your URLs, descriptive phrases) or suspiciously repetitive? You have just run the first pass of a link audit — the exact thing we are about to do for Rivertown.

What DA/DR can do: help you quickly compare and prioritize link opportunities and watch a rough trend. What they cannot do: tell you what Google thinks of your site, predict any specific ranking, or serve as a goal worth optimizing directly. Read them the way you would read a weather forecast from a private app: a useful estimate, not the sky itself.

📈 The Strategy File

Rivertown Home Services — the family firm Ray Delgado founded in 1984, now run by his children Marisa and Tony Delgado — has never thought about its links. That is normal; almost no small business has. But two things force the issue now. First, everything we have learned says links are the strongest off-page signal and the last unfixed lever on the page stuck at #11. Second, Rivertown did not arrive with a blank link profile — it arrived with a contaminated one, because the contractor the Delgados hired years ago (the "SEO guy" whose retainer they are finally ending) built exactly the kind of manufactured links §22.6 describes. Our job this chapter is diagnosis only: baseline what Rivertown's link profile actually contains, and flag the inherited liability. We do not decide what to do about the bad links yet — that decision belongs to Chapter 26, on purpose.

Here is the baseline, assembled from Google Search Console's Links report and a backlink tool for a second opinion.

FIGURE 22.2 — "Rivertown's link profile: a first honest baseline"        [the Strategy File]
  WHAT WE'RE LOOKING AT   Referring domains pointing at rivertownhome.example, sorted by kind.
  WHAT'S THERE
    THE LEGITIMATE (keep, and build on):
      • Branded/navigational: its own social profiles; a few customers and a local blogger who linked
        naturally to "Rivertown Home Services." Small, real, healthy.
      • Local citations: chamber of commerce, a suppliers' "find a certified dealer" page, standard
        business directories (Yelp, YellowPages). Many are nofollow — still valuable for local trust
        and consistency. (These belong to the NAP/citation work in Chapter 25.)
      • A genuine editorial mention: the regional paper covered Rivertown sponsoring a youth sports team
        two years ago and linked to the site. One good, relevant, editorial link — the shape to want more of.
    THE INHERITED LIABILITY (flag now, decide in Ch 26):
      • A batch of low-quality directory and article-spin links from the old contractor.
      • A cluster of overseas "guest post" links on unrelated sites, several with the over-optimized anchor
        "best HVAC repair Rivertown."
      • Blog-comment and forum-signature links dropped across irrelevant sites.
      These carry every manipulation fingerprint from §22.6: irrelevance, repeated exact-match anchors,
      link-network footprints.
  WHAT IT SHOWS          Rivertown's real authority is thin (few strong, relevant referring domains) — which
                         is precisely why the water-heater page can't push past #11 on internal links alone.
                         The manufactured links add no durable value and are a latent risk.
  WHAT IT DOESN'T        This baseline does NOT tell us whether the spammy links are actively hurting the site
                         (Google may already be ignoring them) or whether to disavow them. That assessment —
                         disavow or ignore, and why — is deliberately deferred to Chapter 26.
  THE MOVE (this chapter) Record the baseline: count of referring domains, the good vs. the risky, the anchor
                         profile. Stop paying for any new manufactured links immediately. Do NOT rush to
                         disavow — modern guidance is that it is rarely needed, and a hasty disavow can remove
                         good links by mistake. Diagnose now; decide in Ch 26.
  THE LESSON             You cannot plan to earn authority until you know what you start with — including the
                         debts someone else ran up in your name. (All figures are a constructed teaching example.)

What this increment settles: Rivertown now has an honest starting map — a small stock of legitimate, mostly-local authority, and a pile of inherited manufactured links flagged as a liability. It confirms, from the link side, why the #11 page and the location pages struggle: the site simply hasn't earned enough genuine external authority for its architecture to distribute.

What it does not settle: what to do about the bad links (Chapter 26), and how to earn good ones — the linkable-asset idea in Chapter 23 (a "Cost of AC repair in the Rivertown metro" data piece) and the white-hat plan in Chapter 24 (local sponsorships, suppliers, trade relationships). Marisa Delgado's instinct to "just disavow all the junk today" is exactly the instinct we are going to hold — because acting before we have assessed whether the links are hurting anything is how people accidentally remove the good with the bad.

Your Strategy-File task: open your own Search Console Links report, list your referring domains, and sort them into legitimate and questionable. Note your top anchor text. Do not disavow anything. You are building the baseline every later authority decision will stand on.

Conclusion

We started with a page stuck at the top of page two and ended by naming the last thing holding it there: authority, the off-page signal that other websites confer when they link to you. We traced that signal to its root — PageRank, Page and Brin's insight that a link is a vote, weighted by the standing of the voter and divided among everything the voter points at — and we read the original formula not to compute it but to understand it. We watched link-based ranking evolve past that naive beginning through a twenty-year arms race: the spam, the nofollow attribute, the Penguin update, the shift from counting links to weighing them and folding them into machine-learned systems that no longer hand anyone a tidy number.

We built the practitioner's judgment that replaces the number: a link's value comes from authority, relevance, anchor text, and placement, and above all from whether it was editorially given or manufactured. We learned to read the follow, nofollow, sponsored, and ugc attributes; to see the web as a link graph where external links fill the reservoir and internal links (Chapter 15) route the water; and to recognize why buying links is a detectable, retroactively-punished bet that builds a liability rather than an asset. Finally, we put Domain Authority and Domain Rating in their place — useful third-party thermometers, emphatically not Google's weather.

What remains genuinely uncertain, and we said so plainly, is the exact weight Google places on links versus everything else. Google has never published it; it changes; and precise-sounding claims about it are folklore. What is not uncertain is the strategic conclusion: if links are this valuable and this impossible to safely fake, the only durable path is to earn them — to build things worth referencing and reach the real people who might reference them. That is theme 5 of this book, and it is the whole subject of the next chapter.

→ Continue to Chapter 23: Earning Links — Creating Content People Want to Reference.

Key Terms

• PageRank — Google's original link-analysis algorithm (Page & Brin, 1998): a page scores higher when more, and more authoritative, pages link to it; each linking page splits its vote among its outbound links. Still used internally as one signal; the public "Toolbar PageRank" score was retired in 2016.
• Backlink — any inbound link from another website to a page on yours; the raw unit of off-page authority.
• Referring domain — a unique website that links to you, counted once regardless of how many individual backlinks it sends; a far more meaningful measure than raw backlink count.
• Anchor text — the visible, clickable words a link is wrapped around; a human-written relevance hint about the linked page, and, when unnaturally repetitive, a manipulation signal.
• Editorial link — a link a publisher freely chose to give because your content earned it, with no payment or exchange; the gold-standard, durably safe link PageRank was designed to count.
• nofollow — a rel attribute (2005) marking a link as "not my editorial endorsement," so it should not pass authority; since 2020 treated by Google as a hint rather than an absolute directive.
• sponsored (attribute) — a rel value labeling a link as a paid, advertising, or affiliate placement; the correct way to disclose a paid link so it does not pass PageRank.
• UGC (attribute) — a rel value (ugc, for user-generated content) labeling links created by users, such as in comments and forum posts.
• Link graph — the web modeled as a network of nodes (pages/sites) connected by directed edges (links); the structure over which PageRank and its successors are computed, and the map Googlebot builds by following links.
• Domain Authority (DA) — a third-party 0–100 score from the vendor Moz that predicts ranking ability from link data; not a Google metric and not used by Google to rank (Ahrefs' Domain Rating and Semrush's Authority Score are analogous third-party estimates).

Spaced Review

Retrieval practice. Mix of this chapter and earlier ones. Try each before revealing the answer.

1. In the PageRank formula, a linking page's authority is divided by $C(T_i)$, its number of outbound links. In one sentence, why does that division mean a link from a page with few outbound links is worth more than one from a page with many?
2. "nofollow links are worthless." Give two reasons this is a myth.
3. A site brags about "80,000 backlinks." What single follow-up question most quickly cuts through the boast, and why?
4. (From Chapter 1.) Following links is how Googlebot performs one specific stage of the pipeline before any ranking happens. Which stage, and what is a page called that no links point to?
5. (From Chapter 2.) Sort this claim onto the evidence tiers: "Sites with more referring domains rank higher." Is it confirmed causation, a labeled correlation, or something else — and what is the honest way to state it?