What PageRank actually measured
PageRank scores a web page's importance using nothing but the links between pages. It ignores the words on the page entirely, and asks a question about the shape of the web itself: if somebody browsed the web forever by clicking links at random, occasionally getting bored and jumping to a completely random page, what proportion of their time would they spend on this particular page? That proportion is the page's PageRank.
Two consequences follow, and almost every misunderstanding comes from missing one of them.
PageRank is recursive. A link is not one vote; it is a share of the voting page's own score, so a single link from a heavily linked page can be worth more than a thousand links from pages nobody links to. That is what made the idea interesting rather than trivial: counting inbound links was obvious in 1996, but weighting each link by the importance of its source, and solving the resulting circular definition, was not.
PageRank is also query-independent. It is computed over the whole link graph before anyone types anything, and the same number applies to every search. It contains no information about what a page is about, so on its own it cannot rank anything — it cannot tell you whether a page is relevant, only that the web points at it.
The 1998 paper and the patent that belonged to a university
The public description of PageRank appears in The Anatomy of a Large-Scale Hypertextual Web Search Engine by Sergey Brin and Lawrence Page, presented at the Seventh International World Wide Web Conference in Brisbane in April 1998 and published in Computer Networks and ISDN Systems, volume 30, pages 107–117. A companion Stanford technical report, The PageRank Citation Ranking: Bringing Order to the Web — credited to Page, Brin, Rajeev Motwani and Terry Winograd — sets out the ranking mathematics in more detail.
The paper is not really about PageRank. It is a systems paper describing a working prototype search engine — crawler, repository, inverted index, lexicon, sorter — of which PageRank is one component. It also contains an appendix arguing that advertising-funded search engines are structurally biased against their users' interests — written by the founders of what became the largest advertising business in history.
The intellectual property did not belong to Google. United States patent 6,285,999, Method for node ranking in a linked database, names a single inventor, Lawrence Page, and was originally assigned to the Leland Stanford Junior University. Its priority date is 10 January 1997, it was filed on 9 January 1998 and granted on 4 September 2001 — three years after Google was incorporated. Google held an exclusive licence; Stanford took Google shares, recorded by Wikipedia as 1.8 million sold in 2005 for $336 million. The patent has since expired: Google Patents lists it as "Expired – Lifetime", assigned to Google LLC.
The formula, and what the damping factor is for
In the form given in the 1998 paper, the PageRank of a page A with pages T1…Tn linking to it is:
PR(A) = (1 - d) + d ( PR(T1)/C(T1) + … + PR(Tn)/C(Tn) )
where C(T) is the number of links leaving page T, and d is the damping factor, conventionally 0.85. Read in English: each page hands its own score out in equal portions to every page it links to, and a page's score is the sum of the portions it receives. Some papers normalise the constant term as (1-d)/N for a web of N pages, which changes the absolute numbers but not the ordering.
Because the definition is circular, it is solved not algebraically but by iteration: give every page an arbitrary starting value, apply the formula to every page, repeat until the values stop moving. In linear-algebra terms the answer is the principal eigenvector of a matrix over the link graph, and it converges in a few dozen passes even at web scale. That tractability was the real engineering claim.
The damping factor does two jobs. It models the bored surfer, and it rescues the mathematics: without it, score drains into "sinks" — pages with no outbound links, or clusters that only link to each other — and accumulates there forever. The 15% teleport probability guarantees every page keeps a small share and that the iteration converges to a unique answer. It is a chosen constant, not a measured one.
What PageRank could not do
PageRank never ranked search results by itself, and Google never claimed it did. The 1998 system combined it with conventional retrieval over the page text and — the underrated half of the paper — with anchor text: the words inside a link, attributed to the page being linked to rather than the page doing the linking. Anchor text let the prototype return sensible results for pages it had never crawled, and for pages with no useful text at all. A great deal of what people remember as "PageRank magic" was anchor text.
The deeper limitation is that PageRank has no notion of topic. A page about marine diesel engines linked from a popular unrelated site inherits authority it did not earn, because the model treats every link as an endorsement of equal kind, differing only in the endorser's weight.
That assumption held for as long as it took to become public knowledge. Once links were widely understood to determine rankings, they stopped being purely editorial and started being manufactured, traded and sold. Much of Google's visible ranking history since is a response: the rel="nofollow" attribute introduced in 2005, the Penguin link-spam system of April 2012, and the link-spam systems that replaced it.
Where the idea came from, and the rival that shipped
Ranking documents by who cites them is not a web idea. It comes from bibliometrics, where journal and article influence had been scored from citation counts since the 1950s, and Brin and Page were explicit that they were treating hyperlinks as citations.
Nor was PageRank alone in 1997. Robin Li's RankDex used link analysis for search ranking, with a patent filed on 5 February 1997 and granted on 6 July 1999; Li went on to co-found Baidu. Massimo Marchiori's Hyper Search propagated relevance across hyperlinks. Jon Kleinberg's HITS algorithm split the problem into "hubs" and "authorities" and computed scores per query rather than once globally.
The HITS line of thinking is what shipped as a consumer alternative. Teoma, built out of the DiscoWeb project at Rutgers University led by Apostolos Gerasoulis and in beta from April 2001, ranked by subject-specific popularity: it identified the cluster of pages forming a topical community around the query and counted link votes only from inside it. A link from a leading cardiology site should count for a cardiology query; a link from a generically famous site should not. Ask Jeeves bought Teoma in September 2001 for a reported sum a little over $1.5 million, rebranded it ExpertRank, and shut the index down in late 2010 — ending the only serious alternative to global link popularity ever to run at consumer scale.
The green bar: how a private number became public
Google released the Google Toolbar for Internet Explorer in December 2000. Among its features was a small horizontal bar displaying the PageRank of whatever page you were looking at, as an integer from 0 to 10.
Three properties of that number explain what happened next. It was coarse: eleven buckets covering the entire web, on a scale understood to be roughly logarithmic, so the gap between 5 and 6 was vastly larger than the gap between 2 and 3. It was stale: values were refreshed in batches every few months. And it was a projection — a heavily rounded export of one signal — not the internal value.
None of that stopped it becoming a currency. Because it was the only number Google ever published about an individual page, it was treated as the score of record, and a secondary market grew up in which links were priced by the displayed PageRank of the page selling them. Google spent years explaining that the toolbar figure was not the thing itself. The explanation did not take: the bar was visible and the explanation was not.
Why the public score was retired in 2016
Google withdrew the number in stages over seven years.
- 15 October 2009 — removed from Webmaster Tools, Google's reporting console for site owners, on the stated grounds that people paid it far more attention than it deserved.
- Late 2013 — the last refresh of toolbar values. Nothing was announced; the figures froze and stayed frozen for over two years, so by 2015 the bar described a web that no longer existed.
- April 2016 — the display was removed, dated by Wikipedia to 15 April. The data feed it called was switched off, and every third-party service scraping it went dark at the same moment.
- 12 December 2021 — the Google Toolbar itself was quietly discontinued.
The stated reasoning was consistent: a single stale integer, published without context, was not useful to searchers, and its only committed audience was a trade in links the number itself had created. Google was equally consistent that the internal signal was not being retired — only its publication. In 2016 the public lost a number. Google lost nothing.
What survives of PageRank today
Google still names PageRank in its own documentation. Its ranking systems guide, which lists the systems the company acknowledges using, includes an entry headed "Link analysis systems and PageRank" that reads: "We have various systems that understand how pages link to each other as a way to determine what pages are about and which might be most helpful in response to a query. Among these is PageRank, one of our core ranking systems used when Google first launched… How PageRank works has evolved a lot since then, and it continues to be part of our core ranking systems."
That confirms three things and denies a fourth. Link analysis is in use; something called PageRank is still part of it; and it is one of several systems, on a page listing well over a dozen others. What it does not say is that the 1998 formula is still running. "Has evolved a lot" is doing a great deal of work, and Google has published nothing about the current implementation.
Two practical points follow. There is no public PageRank — any site or extension claiming to show a page's current Google PageRank is showing something else. The third-party authority metrics that filled the vacuum, such as Moz's Domain Authority, Majestic's Trust Flow and Ahrefs' Domain Rating, are each vendor's own model computed over that vendor's own crawl; they are not Google's number and they routinely disagree. And PageRank is not "how Google ranks pages": it is one query-independent signal in a system that also includes text matching, neural systems such as RankBrain and BERT, freshness, locality, deduplication and spam detection.
The mathematics outlived the search engine that made it famous: ranking graph nodes by the stationary distribution of a random walk is now standard practice well outside the web.
Frequently asked questions
What is PageRank?
PageRank is an algorithm that scores a web page's importance from the link structure of the web alone, ignoring the page's content. It models a person browsing at random, following links and occasionally jumping to a random page, and scores each page by how often that random surfer would land on it. A link counts as a share of the linking page's own score, so links from important pages are worth more.
Who invented PageRank and when?
Larry Page and Sergey Brin developed it at Stanford University from 1996, publishing it in the 1998 paper "The Anatomy of a Large-Scale Hypertextual Web Search Engine". The patent, US 6,285,999, names Lawrence Page as sole inventor, has a priority date of 10 January 1997 and was granted on 4 September 2001 to Stanford University, not to Google. Google held an exclusive licence to it.
Is PageRank named after web pages or after Larry Page?
After Larry Page. The pun on "page" is deliberate and the coincidence is genuine, which is why the name reads as generic. The patent's formal title avoids the word entirely: it is "Method for node ranking in a linked database", describing the technique in graph-theory language rather than web language, because nothing in the mathematics is specific to web pages.
Does Google still use PageRank?
Yes, in some form. Google's published ranking systems guide lists "Link analysis systems and PageRank" among its active systems and states that PageRank "continues to be part of our core ranking systems" while noting that how it works "has evolved a lot" since 1998. Google has never described the current implementation, so what remains is the name and the general approach rather than the original formula.
Why did Google remove the PageRank toolbar score?
Because the published number was coarse, months out of date, and useful mainly to a trade in bought links that it had itself created. Google dropped PageRank from Webmaster Tools on 15 October 2009, stopped refreshing toolbar values after late 2013, and switched the display off in April 2016. Only the public figure was retired; the internal link-analysis signal was not.
Can you check a website's PageRank today?
No. Google has published no PageRank value for any page since the toolbar display was switched off in April 2016, and the data feed those tools relied on was turned off with it. Anything advertising a live PageRank check is showing an unrelated third-party metric such as Domain Authority, Domain Rating or Trust Flow — each computed by a private company from its own crawl.
What is the damping factor in PageRank?
The damping factor, conventionally 0.85, is the probability that the modelled random surfer follows another link rather than jumping to a random page. It has a practical purpose beyond realism: without the 15% chance of teleporting, score would drain permanently into pages with no outward links or into clusters that only link to each other, and the iterative calculation would not converge to a single answer.
Was PageRank the first link-based ranking algorithm?
No. Ranking documents by citations comes from bibliometrics and predates the web by decades. Robin Li's RankDex, patented from a February 1997 filing, used link analysis for search; Jon Kleinberg's HITS algorithm scored hubs and authorities per query; Massimo Marchiori's Hyper Search propagated relevance across links. PageRank's distinction was scale, tractability and the search engine built around it.