How an Auction for Eight Words Became the Toll Booth of the Internet

A link is a vote: PageRank before there was a business

Larry Page and Sergey Brin were Stanford PhD students when they began building the algorithm that would become Google, starting in 1996 under the working name BackRub. The insight was not about words on a page. It was about the structure of the web around it. Page and Brin treated a hyperlink from one page to another as something closer to a citation than a decoration: a vote, cast by whoever built the linking page, that the linked page was worth pointing to. Count those votes, weight them by the authority of the page casting them, and a search engine could rank pages by something closer to reputation than by how many times a word appeared on them.

That distinction mattered because the search engines Page and Brin were competing against in the late 1990s, Yahoo and Excite among them, ranked results largely by matching and counting keywords on a page, a method easy to game by repeating a term until a page ranked first regardless of whether it deserved to. PageRank's link-counting logic sidestepped that weakness. A page stuffed with keywords but ignored by the rest of the web earned no votes and sank; a page other sites chose to link to, whatever its keyword density, rose. The method let Google out-rank the keyword-stuffed directories of the era on the strength of a structural signal no single page owner could fully manufacture alone.

Google, the company, was founded in September 1998, two years after the algorithm behind it had already been built. For its first two years the business had a genuinely good search engine and no established way to charge for using it. The gap between having the better ranking mechanism and having a way to price it is worth sitting with, because it is the gap the rest of this history closes, and it was not Google that closed it first.

The auction Google did not invent

Two years before AdWords existed, an entrepreneur named Bill Gross launched a search engine called GoTo.com. GoTo's idea was structurally different from PageRank's: rather than ranking pages by an algorithm reading the wider web, GoTo let advertisers bid, openly, for the right to appear at the top of a results page for a given search term, and charged them each time a searcher clicked through. It was the first search engine built around selling ranked placement on a pay-per-click basis, and it is the direct ancestor of the mechanism that now funds most of the commercial internet.

GoTo was renamed Overture Services in 2001 and acquired by Yahoo in 2003, and for a company whose invention now underwrites a trillion-dollar industry, it is a strikingly minor line in most histories of the web. Overture's obscurity relative to Google's AdWords is itself a data point about this period: the company that built the auction did not end up owning the search engine large enough to make that auction valuable at scale. Google supplied the audience. Bill Gross's idea supplied the payment mechanism. Neither piece alone was worth much; combined, they were worth tens of billions of dollars within a decade.

The two mechanisms stayed conceptually separate inside Google even after they were combined. PageRank kept deciding which pages appeared in the organic results, the ones no one paid for. The auction Google adapted from GoTo decided which paid listings appeared alongside them, priced not by an editor's judgment of relevance but by what an advertiser was willing to bid for that search term at that moment. The two systems answered different questions, who deserves to be found and who is willing to pay to be seen, and running them side by side, on the same page, in front of the same searcher, is the specific architecture that turned search into a business.

The toll booth opens, October 2000

Google launched AdWords in October 2000 with roughly 350 advertisers and earned about $70 million in advertising revenue that first year. The mechanics were simple by later standards: an advertiser chose the words a searcher might type, named a price they were willing to pay for each resulting click, and their ad appeared when that search happened, provided their bid cleared whatever threshold the auction that moment required. No salesperson had to answer the phone. No insertion order had to be signed. A business anywhere with a credit card could enter the auction for a single search term and start buying clicks within minutes.

What made this a genuinely new kind of transaction, not just a new sales channel for advertising, was the timing. A newspaper sold a fixed block of space to whoever paid for it in advance, regardless of who read the paper that day. A billboard sold visibility to everyone who drove past it, whether or not any of them wanted what it advertised. AdWords sold neither space nor an audience. It sold a single moment: the instant a specific person typed a specific phrase that signaled they wanted something, and only that instant. The bid an advertiser placed was a bid on intent itself, measured and priced at the exact second it appeared, a kind of transaction no earlier medium had been built to sell.

That is the mechanism worth naming plainly, because it is the economic argument of this history and not an aside to it. Search had been a free utility since the first web directories, the way a card catalogue in a library was free. AdWords did not charge searchers for using it and never has. It charged the businesses standing on the other side of that search, for the right to be shown at the precise moment a customer's want became visible to the system reading it. Information retrieval had been free for years. The moment of commercial intent inside it had never been priced before, and after October 2000, it always would be.

The self-serve part of the design mattered as much as the auction part. A newspaper's advertising desk decided which businesses were worth the trouble of a sales call; a small hardware store or a single-location dentist rarely made that list. AdWords needed no sales call. Any business with a credit card could open an account, name a search term, and enter the same auction as a national brand, bidding against it directly for the same searcher's attention. That did not make the auction fair in the sense of giving every bidder an equal chance, since a well-funded advertiser could simply outbid a small one, but it did remove the gatekeeper who had decided, in every earlier advertising medium, who was even allowed to compete for the placement in the first place.

A patent suit and 2.7 million shares

Overture, the renamed GoTo, did not watch Google build a business on its auction model without a response. In April 2002, Overture sued Google for patent infringement, arguing that AdWords' pay-for-placement bidding system violated a patent Overture held, US Patent 6,269,361, covering the method it had pioneered in 1998. The suit sat unresolved for more than two years while AdWords kept growing and Google kept preparing for a very different kind of event.

On August 19, 2004, Google went public, pricing its shares at $85 through a Dutch auction, a method that let the market rather than a syndicate of bankers set the offering price. The IPO raised $1.9 billion and valued the company at roughly $23 billion; the stock closed its first day of trading up 18 percent, at $100.34. Twenty days later, on August 9, 2004, Google settled the Overture suit by issuing 2.7 million shares of Class A stock to Yahoo, which had acquired Overture the previous year, worth roughly $229.5 million at the $85 IPO price, in exchange for a perpetual license to keep running the auction model it had built its revenue on.

It is worth stating what that settlement actually shows. Days before becoming one of the most valuable public companies in the world on the strength of an advertising auction, Google paid the equivalent of over two hundred million dollars for the legal right to keep using a mechanism someone else had invented first. Ownership of the underlying idea and the profit made from running it at scale were not the same thing, and by 2004 it was already clear which one mattered more.

From $70 million to $29.3 billion

The revenue trajectory that followed is the plainest evidence for how completely the auction reshaped the business. AdWords earned Google roughly $70 million in its first year, 2000. By its IPO year, 2004, Google's annual revenue had reached $3.2 billion. By 2010, six years later, that figure stood at $29.3 billion, a roughly ninefold increase in six years, financed overwhelmingly by AdWords' search-auction revenue rather than by selling ad space in the traditional sense or by charging anyone a subscription to search.

That distinction is the structural break from every prior advertising-funded medium. A newspaper or a broadcaster sold a defined unit, a page or a minute, at a price the publisher set, adjusted occasionally by circulation or ratings. Google sold nothing it had to define in advance. Every unit up for sale, each specific search term at each specific moment, was priced fresh by whichever advertisers chose to bid on it, continuously, at global scale, without a sales department setting a rate card. The auction did the pricing work a media company's commercial department used to do by hand, and it did it for every one of the billions of searches Google was, by 2010, running each year.

The growth also mapped onto a market no earlier advertising medium had reached at that speed. A newspaper's advertising rates were bounded by its print run and its city. AdWords' auction had no comparable ceiling, because a search engine has no physical edition to run off a press and no broadcast signal that stops at a state line. Once the mechanism worked, scaling it was a matter of more searches, more advertisers, and more languages, not more presses or more transmission towers, and each of those constraints fell away faster for Google in its first decade than it ever had for a newspaper or a television network in a comparable span.

A global toll, and who counts as authoritative

The auction Google built is not a national marketplace with export customers. It is a single, dollar-denominated market that any advertiser anywhere can enter, whether they are bidding from Ohio, Lagos, or Bengaluru, and it is cleared, priced, and paid out through one company's infrastructure regardless of where the advertiser or the searcher sits. By 2025, Google Search held roughly a 90 percent share of the global search-engine market, with the United States supplying only about 24 percent of its traffic. The remaining three quarters of the searches feeding that market, and a comparable share of the advertisers bidding into it, sit outside the country where the auction is run and the revenue is booked. That is the geopolitical shape of the toll booth: a mechanism built and governed inside one jurisdiction became the default point through which most of the world's commercial search intent has to pass, and be priced, before a buyer reaches a seller.

In August 2024, the US District Court for the District of Columbia ruled in United States v. Google that the company held an illegal monopoly over internet search and search advertising, finding it had paid billions of dollars to device makers and browser developers to remain the default search engine on their products. The ruling is a domestic antitrust finding, decided under US law inside a US court, but the market it concerns is not domestic. A monopoly ruling issued in Washington reaches a marketplace that clears bids from every country that can pay in dollars, which is most of them, an asymmetry with no obvious precedent in the advertising history that came before it.

PageRank's original mechanism carries a quieter version of the same asymmetry, and this reading of it is an interpretation of how the mechanism behaves, not a separately sourced statistic. A system that ranks authority by counting inbound links will, structurally, favor whichever parts of the web are most densely linked to each other, and the web's earliest and most heavily interlinked regions were disproportionately English-language and disproportionately built by institutions in wealthier, more internet-connected countries. A page written in a less-linked language, or hosted in a market where fewer sites existed to link to it, had structurally fewer votes available to earn, whatever the quality of the page itself. PageRank never set out to decide whose language or whose country counted as authoritative. Its link-counting logic did that anyway, as a side effect of measuring the web as it happened to have been built, and the effect compounded every time a highly linked page earned more links for already being highly linked.

None of this argues that the auction or the ranking algorithm was built with either intent. Bill Gross wanted a working ad business, not a foreign-exchange chokepoint. Larry Page and Sergey Brin wanted a better way to rank web pages, not an arbiter of national authority. What both mechanisms share with the answer engines reading the web today, the ones now deciding which businesses get named when a buyer asks a question rather than clicks a link, is the same lesson: whoever builds the ranking and pricing infrastructure for a medium ends up governing, whether deliberately or not, who gets found inside it, and on whose terms.

It is also worth holding the two-sided reading of the toll booth itself. The self-serve auction that AdWords opened in October 2000 genuinely lowered the barrier to advertising for millions of small businesses that no newspaper or television station would ever have taken a sales call from, and the search results PageRank ranked did, for years, reward genuinely useful pages over merely keyword-stuffed ones. The same mechanisms that opened access also concentrated the toll collection in a single company's ledger, denominated in dollars, cleared through servers in the United States, and reachable by an antitrust court in only one jurisdiction. Both of those things are true about the same twenty-five years, and neither cancels the other out.