Before the artificial intelligence wave truly broke over Wikipedia, the people who tended it had already sensed the weather. The Wikimedia Foundation, the nonprofit that hosts the encyclopedia, was not caught entirely by surprise. There had been years of conversation about machine learning, about automation, about the possibility that software might someday do some of the drudgery of keeping the world’s largest reference work alive. Wikipedians, those countless volunteers who write, edit, argue, and refine articles in their spare time, braced themselves with a kind of qualified optimism. They made confident predictions that the storm would break and then wash over them, leaving the project intact. Some even said large language models might improve Wikipedia. Perhaps AI could take over the less glamorous housekeeping tasks—fixing formatting, cleaning up citations, adding missing links—and free human beings to do the work that actually mattered. The creative work. The thoughtful work. The work that made Wikipedia feel like a living human endeavor rather than a cold database. It was not so simple, of course. In 2019, Katherine Maher, then the head of the Wikimedia Foundation, published an op-ed in The New York Times under the warning headline “Without Humans, A.I. Can Wreak Havoc.” She wrote that artificial intelligence was too often presented as an all-powerful solution, a scalable replacement for people. The issue she named was not abstract. Wikipedia was facing the same question that would soon confront the rest of the world: how do we keep machines from taking over the things that make us human? How do we preserve human control over the tools we create? Maher saw clearly that the question of who controls the encyclopedia was really a question about who controls knowledge itself.
That question became urgent faster than most people expected. As early as 2018, Google was already training artificial intelligence to write Wikipedia-like articles. By October 2024, a study from Princeton University found that around 5 percent of the nearly three thousand newly created articles on English Wikipedia in a single month—August 2024—contained a substantial amount of AI-generated content. Five percent may not sound enormous, but in a project of Wikipedia’s scale, it represented thousands of pages written by machines. The signs were subtle, and yet they were there if you knew where to look. AI loved em dashes, those long parenthetical dashes that interrupt a sentence with an aside—like this one. It favored curly quotation marks in a way human writers rarely did. Sometimes, startlingly, the machine left behind scraps of its own creation, revealing the prompt that had been used to compose the article. “Here is your Wikipedia article,” a page might say, as if a ghost were confessing. At first, Wikipedians tried to shape and control the technology. They wanted to make AI output “acceptable and constructive,” to teach it the norms of the community, to fix its mistakes when it failed. Generic filler—praising a village for its fertile farmlands, livestock, and scenic countryside—was weeded out. Fake sources were flagged and removed. But the problem only grew. By March 2026, the violation of Wikipedia’s core principles had become so frequent that the community made a radical decision: writing with large language models was banned outright, and their role was restricted to translation and minor proofreading. It was a defensive maneuver, a line drawn in the sand, and no one knew whether it would hold.
The fear behind that ban was articulated in a widely read essay called “Death of Wikipedia.” The essay was not about a sudden collapse but about a slow, insidious decay. It described a ripple effect. If contributors began to reduce their participation because AI had made the platform feel alien, there would be a temptation to fill the gaps with even more AI-generated content. That would drive away more human contributors, which would create more gaps, which would invite more machines. A downward spiral. This dread struck a chord far beyond Wikipedia. A viral post on X captured the same frustration with artificial intelligence more broadly: “I want AI to do my laundry and dishes so that I can do art and writing, not for AI to do my art and writing so that I can do my laundry and dishes.” It was a simple, devastating reversal of priorities. People did not want to be freed from creativity to do chores; they wanted to be freed from chores so they could be creative. But the logic of AI was pushing in the opposite direction, taking over the noble, expressive, human activities and leaving people with the tedious, mechanical work. “Death of Wikipedia” warned that Wikipedia could suffer the same fate. Instead of AI performing the housekeeping roles while humans wrote the articles, AI would gradually compose the articles themselves, and humans would be relegated to formatting and cleanup. The result would be a strange inversion. Editors would become custodians of a machine’s output, trapped in an endless curatorial project, polishing sentences that no human had ever truly written, preserving ideas that no human had ever truly thought. The priorities that shaped the encyclopedia would derive not from human needs or human passions, but from the frostbitten harvests of machines—cold, repetitive, and strangely lifeless.
The most painful concern was not just about the quality of articles, but about the fate of the people who had built Wikipedia. One Wikipedian wrote that the potential whittling down of the pool of talented writers, year after year, was the greatest threat Wikipedia faced. The project needed to increase the number of editors, especially expert editors, not watch them drift away. Without them, no amount of AI-generated content could fix the sprawling hellscape of weak, inaccurate, and incomplete articles that would drag the project down. The image is striking: a hellscape, but sprawling, the kind of landscape that emerges when you try to build a cathedral by microwave. A Wikipedia written and read by machines would be a deracinated Wikipedia, stripped not only of its context but of its animating passion. It would be artless, and soulless. The Wikimedia Foundation often says that the care and commitment of Wikipedians to reliable encyclopedic knowledge is something AI cannot replace. That is a comforting phrase. But it is not the same as saying that it will not be replaced. There is a difference between value and survival. Something can be irreplaceable in principle and still be lost in practice. The threat was not that AI could do what Wikipedians did better, but that it would do it just well enough, at a scale no human community could match, and then the human community would quietly fade away. And once the people were gone, the spirit that held the whole thing together would be gone too. The encyclopedia might still look like Wikipedia on the outside, but it would be a hollow edition, a library where all the librarians had left and the books were still being shelved by robots.
The arrival of Google’s AI Overviews in search only deepened the uncertainty. These AI-generated summaries quickly became as central to the search experience as Wikipedia itself had been until recently. This does not mean Wikipedia stopped being relevant. On the contrary, it continues to serve a crucial role as a source, feeding information to the AI systems that feed the rest of us—only now at one step removed. But that distance matters. Anyone who has used AI agents for long enough knows how they work. They give plausible answers, almost intuitive, that can pass a superficial review of their sourcing. They sound confident. They sound complete. And they are often wrong. When an AI system draws on Wikipedia and then garbles the facts, or when it invents a citation out of thin air, there is usually no way for the reader to check the original context. The machine has absorbed the source and regurgitated a smooth, confident version of it, with no indication of where it went wrong. This was always the deeper worry. If Wikipedia itself becomes corrupted by artificial intelligence, by what means will we prove when the machines are wrong? In the past, when a dispute arose, someone could point to an article, a conversation, a source. There was a community that could be appealed to. There were human hands on the text, human memories behind it. But if the knowledge base itself is poisoned, the chain of accountability is broken. There is no one left to ask. No one to challenge the error. No one to say, with authority, that the machine has misunderstood. We would be left with systems that cite other systems, and engines that trust their own output, and an endless loop of plausible falsehoods with no human anchor.
Implicit in the idea of a single, machine-edited encyclopedia is a promise of finality. This is the dream of the one true reference work, the ultimate authority, a digital King Solomon rendered in calm and unbiased tones. The appeal of that dream, for the kind of people who find it appealing, is that it will be coolly rational. It will not have pet topics or political axes to grind. It will be a disinterested arbiter, standing above the noise of human conflict. But there is no reason to think such cold logic would not be cruel instead. There is no reason to believe a perfectly rational encyclopedia would be kind, or wise, or just. Its aesthetic failings are a clue. The word “slop” has become a useful name for the product of artificial intelligence, and it carries a lot of meaning. Slop is unvariegated and monotonous. It fills you up without nourishing you. It lacks not only nutritional value but also the flavor that makes nutrition a pleasure. A machine encyclopedia might be comprehensive in a way Wikipedia never was, but comprehensiveness is not the same as connection. It might contain all the facts and none of the love. Wikipedia is imperfect. It is full of arguments, dueling sources, quirky edit histories, inside jokes, and stubborn volunteers who will not stop tinkering with an article until it is just right. It is, at its best, an approximation of the “sum of all human knowledge,” not because it holds every fact, but because it holds the traces of so many human efforts to understand and share what matters. When smartphones became widespread after the iPhone’s launch in 2007, Wikipedia became costless, portable, immediate, electric. It was enormous and omnipresent. It really did become a kind of oracle. But so what? What did access to this oracle mean for human flourishing? What good has Wikipedia done? Maybe the answer is not in the article count, not in the statistics, not in the number of facts verified or citations added. Maybe the good it has done is showing that a vast, shared, human-created knowledge project is possible. That people will contribute their time and their care to something they will never own and never be paid for. That we can build a cathedral together, desperately and beautifully and imperfectly, without any single architect ordering us around. And that is something no machine can replace—though the machines may try.
