A study of 65,000 URLs found that 52% of newly published web articles were AI-generated. In the same analysis, only 14% of Google search results contained AI content.
The gap between those two numbers is the most important fact about the modern internet. Generation has been solved. Discovery has not.
Jorge Luis Borges published "The Library of Babel" in 1941. The story describes a universe composed entirely of hexagonal galleries, each containing the same number of shelves, each shelf holding the same number of books, each book the same number of pages. Every possible combination of 25 orthographic symbols exists somewhere in the collection. Every book that could be written is already written. Every truth, every lie, every contradiction of every truth. The problem is finding which is which.
The librarians can't. The ratio of gibberish to meaning is so vast that a librarian could walk the galleries for a lifetime and never encounter a single coherent sentence. The story is four pages long and contains no plot. It is an existence proof: here is what a universe of total information looks like from the inside.
Borges describes three responses. The Purifiers travel the hexagons destroying books they judge to be without value, hoping to narrow the collection toward meaning. The pilgrims wander in search of the Vindications, books that are supposed to contain a faithful summary of each person's life. And a persistent rumor circulates about the Catalog of Catalogs, a master index that would let you find any specific volume.
The Catalog, if it exists, is itself a book in the library. It sits on a shelf in one of the hexagons, indistinguishable from the nonsense volumes surrounding it. To find the Catalog, you would need a catalog of where to find it. The recursion is the point.
Eighty-five years later, Google employs 16,000 human contractors as Search Quality Raters. Their job is to flag low-quality content so the ranking algorithms can learn what to suppress. Google also built SpamBrain, an AI system that detects synthetic spam at scale. YouTube's CEO Neal Mohan declared "managing AI slop" a top priority for 2026. In 2025, Google updated its Quality Rater Guidelines to specifically target mass-produced AI pages as "Lowest" quality regardless of how they were created.
These are Purifiers. They're building filters for a library that grows faster than any filter can sort. The 52% figure from the Graphite study describes only text articles. It doesn't count AI-generated images, code, product listings, reviews, social media posts, or video. The real ratio of machine-generated to human-generated content is higher, and it moves in one direction. The Purifiers in the story are described with pity. They destroy real books along with the garbage, because at the ratios involved, any filter aggressive enough to remove the noise will inevitably remove signal too. Borges understood the precision-recall tradeoff before information retrieval existed as a field.
A paper in Nature describes what the researchers call model collapse. AI systems trained on AI-generated output degrade progressively, losing the nuance and variety of the human-created content they originally learned from. Each generation of model trained on the previous generation's output drifts further from the original distribution. The library is generating copies of its own books, and the copies are worse than the originals.
This is Borges's Catalog problem restated as mathematics. A system that indexes the library using the library's own contents converges on noise. The Catalog can't be built from books already on the shelves. It requires something from outside the system. In practice, that something is still human judgment, applied at a scale (16,000 contractors, manual review guidelines, editorial policies) that doesn't match the rate of generation.
The 52%/14% gap measures how well the current Catalog works. More than half of new content is synthetic. Fewer than one in seven search results are. Google's filters catch most of it. For now. The question is whether the filter can keep pace with the flood, or whether the ratio eventually overwhelms it the way the library overwhelms the Purifiers in the story.
Borges died on June 14, 1986, in Geneva. He wrote "The Library of Babel" in Buenos Aires during the Second World War, in a country watching Europe destroy itself from across an ocean. He never saw the internet. He saw what the internet would become.
The story ends with a note of qualified hope. The library, the narrator observes, is "unlimited but periodic." If you walk far enough in any direction, the arrangement of books repeats. The universe of information isn't infinite. It's just too large to distinguish from infinite by anyone standing inside it.
The same might be said of the internet in 2026. It is not infinite. It just looks that way from inside the hexagon to anyone trying to find something true.