How Search Algorithms Killed the Weird Web by Making It Invisible

On April 21, 2015, Google rolled out an algorithm update internally called "Mobilegeddon," and it did something to the web that no amount of nostalgia blogging has ever fully reversed. The update was simple in its logic: sites that failed Google's mobile-friendliness test would see their search rankings drop. The test checked for things like readable text without zooming, tap targets spaced far enough apart, and pages that didn't require horizontal scrolling. A site built in 1998 with nested tables, fixed-width layouts, and a visitor counter at the bottom had no chance of passing it. And when a site lost its ranking, it lost its audience. When it lost its audience, the person who maintained it—often someone who had been updating a niche fan page for fifteen years out of pure obsession—stopped paying for hosting. The common story about this moment is that algorithms got too smart. That predictive engagement models learned what we wanted and fed it to us until the weird edges of the web got smoothed into irrelevance. But that story misses the mechanism. The algorithm didn't decide that Jar Jar Binks fan fiction from 1999 was unworthy of being read. The algorithm decided it was unworthy of being indexed in a way that made it findable. And those are two different things. A page built by a fan in the late 1990s was, by the standards of 2015, a compliance nightmare. It was not mobile-friendly. It was not accessible. It was not structured in a way that made it easy for an ad network to place a unit next to the content. It was, in many cases, a copyright ambiguity: fan fiction, image manipulations, and character shrines that existed in the gray space between tribute and infringement, tolerated by rights holders when the content was invisible and ignored, but a liability when it was surfaced by a search engine that had to answer to advertisers and legal departments. The economic infrastructure of indexing had no room for that kind of page, not because the page was bad, but because the page was unmonetizable, legally ambiguous, and structurally non-compliant with the standards that had become table stakes for any site that wanted to be found. The GeoCities archival project, The GeoCities Gallery, is the forensic evidence for what was lost. GeoCities hosted millions of pages, organized into neighborhoods by topic. A neighborhood called Area51 was for science fiction and fantasy. A neighborhood called Hollywood was for film and television. Inside those neighborhoods were pages that had no reason to exist except that someone wanted to make them: a shrine to a minor Star Wars character, a page collecting every appearance of a specific background actor, a fan theory about a plot hole that no one else had noticed. When GeoCities shut down in 2009, a group of archivists scrambled to save as much of it as they could. The resulting archive is a snapshot of a web where the cost of publishing was so low and the discoverability infrastructure was so indifferent to monetization that people built things simply because they wanted to. What the archive also shows is that the content was not, by any commercial standard, good. The pages were visually chaotic. They used default browser fonts. They had broken image links. They were, in the language of the Mobilegeddon test, not mobile-friendly and not accessible and not structured for readability. But they were also, in a way that a modern content management system cannot replicate, specific. A page about a single Star Wars character, maintained for five years by someone who had no audience and no analytics and no reason to keep updating it except that they cared about the thing, is a kind of content that the modern web's infrastructure does not produce. Not because people stopped caring about things. Because the infrastructure stopped making that kind of caring legible to a search engine. The economic logic here is not a conspiracy. It is a set of incentives that played out across millions of individual decisions. A search engine that makes money from ads needs pages that can host ads. A page that can host ads needs to be structured in a way that makes the ad unit visible and clickable. A page that is structured for ads is, by definition, not structured like a GeoCities fan page from 1998. The flattening of the web into a lukewarm sludge of content was not a thermodynamic inevitability; it was a series of compliance thresholds that filtered out everything that couldn't meet them. The weird stuff didn't stop being made. It stopped being findable. And when something stops being findable, for all practical purposes, it stops existing. The takeaway is not that Google is evil. It is that the infrastructure of discoverability has a set of requirements—mobile-friendliness, accessibility, ad compatibility, legal clarity—and those requirements are not neutral. They select for a certain kind of content and a certain kind of creator, and they deselect for the kind of person who built a Jar Jar Binks fan page in 1999 and updated it for a decade because no one told them they couldn't. The web didn't get less weird because people got less weird. It got less findable for weirdness. And that distinction matters, because it tells you what you would have to change if you wanted the weirdness back.