PageRank Explained: How Google Values Links

In 1996, Larry Page was a twenty-three-year-old PhD student at Stanford, and he had a problem that was bothering him in the way that only academic problems can — endlessly, quietly, in the background of everything else. And the web was growing. Fast. Search engines existed, sure, but they were terrible. They matched keywords and returned results based on how many times a word appeared on a page. You'd search for "jaguar" and get a mishmash of car dealerships, animal facts, and someone's personal homepage where they'd mentioned the word seventeen times in white text on a white background because they'd figured out that trick. The web was information without organization, a library where every book was dumped on the floor.

Key Takeaways

• The Origins of PageRank
• How PageRank Actually Works
• PageRank in Modern SEO
• Why PageRank Still Matters Today

Page's insight — and this is the part that changed everything — came from academia itself. In the world of scholarly research, a paper's importance is partly determined by how many other papers cite it. A paper cited by hundreds of other researchers is probably more significant than one cited by two. But it's not just the count that matters. A citation from a landmark paper in your field carries more weight than a citation from an obscure thesis nobody read. Quality of the citation matters as much as quantity. Maybe more.

He looked at the web and saw the same structure. Web pages link to other web pages. Those links are, in a sense, citations. Endorsements. One page pointing to another is saying, "this is worth your attention." And just like academic citations, not all links should count equally. A link from a major news organization's homepage should mean more than a link from a random page buried six levels deep in someone's forgotten GeoCities site. So how do you quantify that intuition? How to turn a vague sense of "some links matter more than others" into math.

Working with Sergey Brin, his fellow Stanford grad student, Page developed what they initially called "BackRub" — a system for analyzing the web's link structure. BackRub didn't stick as a name, thankfully. What did stick was the algorithm at its heart, which they named PageRank. PageRank was a pun, or at least a happy coincidence — it referred both to web pages and to Larry Page himself. The paper they published in 1998, "The Anatomy of a Large-Scale Hypertextual Web Search Engine," laid out the math and the vision. It's still readable today, which is more than you can say for most academic papers from that era. If you want to go further, Understanding Link Equity: A Complete Guide has you covered.

Mathematically, PageRank is elegant in a way that might surprise people who think of Google as just a big advertising company. It models the web as a directed graph — pages are nodes, links are edges. Then it imagines a "random surfer" who starts on a random page and follows links, clicking from one page to the next without any particular goal. At each step, the surfer either follows a link on the current page (chosen randomly from all available links) or gets bored and jumps to a completely random page somewhere on the web. This boredom factor — technically called the damping factor — was set at 0.85 in the original paper, meaning that 85% of the time the surfer follows a link, and 15% of the time they jump somewhere random.

The Origins of PageRank

For any given page, its PageRank is essentially the probability that this random surfer ends up on that page at any given moment. Pages with lots of incoming links from important pages have a higher probability. Pages with few links pointing to them, or links only from unimportant pages, have a lower probability. Solving it involves iterating this calculation across the entire web graph until the values converge — until each page's score stabilizes and stops changing significantly with additional iterations. It's an eigenvector problem, if you want to get technical about it. Put technically, the PageRank values are the principal eigenvector of the web's link matrix, normalized so that all values sum to one.

What made this brilliant wasn't just the math. It was the insight that the web's own structure contains information about quality. You don't need a team of human editors deciding which pages are good — though Yahoo was trying that approach and struggling to keep up with the web's growth. You don't need to trust what page authors say about their own content — because of course everyone claims their page is the most relevant result. Instead, you look at what everyone else on the web implicitly says by choosing to link or not link. Collectively, the judgment of millions of webmasters, expressed through their linking decisions, becomes a quality signal. It's crowdsourced expertise, extracted automatically from the architecture of the web itself.

Think of it like recommendation letters. If you're hiring someone and they hand you three letters of reference, those letters matter. But a letter from a well-known leader in the field carries more weight than a letter from someone you've never heard of. And a letter from someone who writes recommendations for everyone who asks is less meaningful than one from someone who rarely endorses anybody. PageRank captures both of these intuitions. Links from authoritative pages count more. And a page that links to thousands of other pages dilutes its endorsement across all of them, so each individual link from that page carries less weight.

A voting analogy works too, to a point. Every link is a vote. But it's not a democracy where every vote counts equally — it's more like a weighted election where some voters' opinions matter more based on how much trust they've accumulated. And the trust itself is determined recursively. Page A is trustworthy because trustworthy pages link to it. But how do we know those pages are trustworthy? Because other trustworthy pages link to them. It's circular reasoning, technically, but the iterative mathematical process resolves the circularity. After enough iterations, the scores settle into stable values that reflect the overall link structure of the web. For the full picture, read How Many Backlinks Do You Need to Rank on Google?.

When Google launched in 1998, PageRank was its secret weapon. Other search engines were matching keywords and counting word frequency. Google was doing that too, but layered on top was this understanding of link-based authority. Search results were noticeably better. People tried Google once and stopped going back to AltaVista. Word spread through dorm rooms and offices. Within a few years, "google" had become a verb, and the company had gone from a Stanford research project to one of the most valuable entities on earth.

How PageRank Actually Works

Looking back, the early days of PageRank were almost innocent. Webmasters didn't know about it. They linked to things they genuinely found useful or interesting. The link graph was a relatively honest map of how people on the web felt about different content. It was messy and incomplete, but it was authentic. Google's results benefited from that authenticity, feeding off a web where links meant something real.

That didn't last, of course. Once people figured out that links influenced rankings, the manipulation started. Link farms — networks of sites created solely to link to each other and boost PageRank — appeared almost immediately. People bought and sold links. They created automated tools that would blast links across forums, blog comments, guestbooks, and anywhere else that accepted user-generated content with a hyperlink. Soon the web became polluted with links that existed not to help anyone find or discover content, but purely to game the algorithm.

Google adapted. They had to. Originally, the PageRank formula was clean and beautiful, but the real world is neither of those things. Over the years, Google layered on additional signals and filters. They began discounting certain types of links — links from known spam sites, links in blog comments, links with suspicious anchor text patterns. They introduced the "nofollow" attribute in 2005, giving webmasters a way to indicate that a link shouldn't pass PageRank. They launched algorithm updates — most famously Penguin in 2012 — that specifically targeted manipulative link building practices.

There's a question that comes up a lot: does Google still use PageRank? The answer is yes, but with heavy caveats. Google confirmed as recently as 2020 that PageRank is still one of many signals in their ranking system. But the PageRank of today is not the PageRank of 1998. It's been modified, extended, supplemented, and in some cases overridden by hundreds of other ranking factors. Modern Google uses machine learning systems — BERT, MUM, RankBrain — that understand content semantically in ways that pure link analysis never could. PageRank is one instrument in a very large orchestra. See also our post on Backlink Quality vs Quantity: What Matters More for SEO for more on this.

Remember the toolbar PageRank that webmasters used to obsess over — that little green bar that went from 0 to 10 — was discontinued in 2016. Google stopped updating it publicly years before that. They realized, probably correctly, that making the metric visible was causing more harm than good. People were gaming it, stressing over it, and making bad decisions based on an oversimplified representation of a complex internal score. Sound familiar? It's the same pattern that plays out with every visible SEO metric. Once people can see a number, they start trying to manipulate it.

PageRank in Modern SEO

But the underlying concept — that links are signals of trust and quality, that some signals are stronger than others, that the web's structure itself encodes information about what deserves attention — that hasn't gone away. It can't, really. It's too fundamental to how interconnected information works. Even if Google replaced PageRank entirely with some other system, that system would still need to account for the fact that links exist and that they mean something.

What's fascinating is how the ideas behind PageRank have spread far beyond web search. Social network analysis uses similar math to identify influential users. Academic citation networks are now analyzed with PageRank-inspired algorithms to measure the impact of papers and researchers. Twitter — or whatever it's called now — uses variations of the concept to rank tweets and recommend accounts. Recognizing that importance can be derived from network structure, that who connects to whom tells you something about relative significance, has become one of the foundational ideas of the information age.

And it's worth sitting with the strangeness of that for a moment. An algorithm designed to rank web pages in the late 1990s ended up reshaping how we think about influence, trust, and authority across almost every networked system. Epidemiologists have used PageRank-style analysis to model how diseases spread through contact networks. Economists have applied it to trade networks between countries. Biologists have used it to study protein interaction networks, trying to identify which proteins are most critical to cellular function. And the core idea — that a node's importance depends on the importance of the nodes connected to it — turns out to be one of those rare mathematical insights that keeps finding new applications in places its creators never imagined.

There's something almost philosophical about it. The web, in those early days, was often compared to a wild frontier. No rules, no hierarchy, no central authority deciding what mattered. PageRank imposed a kind of emergent order on that chaos. Not top-down order — nobody was making editorial decisions — but bottom-up order, arising organically from the choices people made when they decided to link one page to another. It was democratic in theory, meritocratic in aspiration, and deeply imperfect in practice. But it worked well enough to build one of the most powerful companies in history on top of it. There's more to explore, and What Is Domain Authority and How to Improve It is a great place to start.

The people who study network science sometimes talk about "preferential attachment" — the idea that nodes with more connections tend to attract even more connections. Popular pages get linked to more often, which makes them rank higher, which makes them more visible, which gets them linked to even more. It's a rich-get-richer dynamic, and PageRank both reflects and amplifies it. Critics have pointed out that this creates a kind of entrenched hierarchy on the web, where established sites maintain their dominance partly because the algorithm rewards their existing dominance. New sites, no matter how good their content, face an uphill battle. Whether that's a flaw or a feature depends on your perspective.

Why PageRank Still Matters Today

Larry Page probably didn't foresee all of that when he was sketching out equations in a Stanford dorm room. Or maybe he did. He was, by most accounts, thinking big even then. The original PageRank paper talks about scaling to the entire web, which at the time was maybe 150 million pages. Today's web has billions. The math still works. The principle still holds. The implementation has changed beyond recognition, but the seed — the idea that a link is a vote, and some votes count more than others — remains planted at the foundation of how we find information online.

There's a certain irony in the fact that PageRank, which was designed to surface the best content on the web, inadvertently created an entire industry devoted to gaming it. SEO exists, at least in part, because PageRank exists. Link building as a practice — the outreach, the guest posting, the digital PR campaigns — all traces back to the simple insight that links influence rankings. Every cold email asking for a backlink is, in some distant way, a consequence of a math paper written by two grad students who thought academic citation networks might teach us something about the internet.

And the arms race continues. Google gets better at detecting manipulation. SEOs get more sophisticated in their approaches. The definition of a "natural" link keeps shifting as both sides evolve. Some people see this as adversarial, and in a sense it is. But there's another way to look at it. PageRank aligned Google's incentives with the web's incentives — at least partially. If you create something genuinely valuable, people link to it. Those links help you rank. Ranking brings you traffic. The algorithm, despite all the gaming and manipulation, still roughly rewards quality in the long run. Not perfectly. Not always. But roughly.

The original paper is worth reading if you haven't. It's technical but accessible, and there's a clarity of vision in it that's rare in academic writing. Page and Brin weren't just describing an algorithm. They were describing a philosophy of how information should be organized. The web, they argued, contains within its own structure the answer to the question of what's important. You just have to know how to read it. To understand this better, take a look at How Links Affect Your Google Rankings.

Whether Google still reads it that way — whether the company that grew out of that philosophy still holds to it — is a different question. One that probably doesn't have a clean answer. The algorithm has gotten so complex, so layered with machine learning and behavioral signals and entity recognition and a thousand other things, that the original elegance of PageRank is buried under years of engineering. It's still there. Somewhere. Like the foundation of a building that's been renovated so many times you can't see the original walls anymore, but the structure still rests on them.

I think about Larry Page sometimes, walking across the Stanford campus in the mid-90s, thinking about links as citations, citations as votes, votes as a map of what matters. The web was small enough then that you could almost hold it in your head. Now it's unimaginably vast, and the algorithm that helped organize it has become one of the most consequential pieces of software ever written. Not because the math was particularly novel — eigenvector computations were well understood — but because the application was perfect. The right idea at the right moment, applied to the right problem.

That kind of thing doesn't happen on purpose most of the time. It happens because someone is curious about something specific and follows the thread far enough to find something universal. Page was curious about backlinks. He followed the thread. And the rest of us are still living with what he found