Wikipedia Has Lost a Quarter of Its Readers Since 2022

By Marcus Bennett

Wikipedia has lost almost a quarter of its readership since 2022 — more than 1.1 billion visits per month — an audience that presumably now reads AI summaries instead. That figure comes from a "wake-up call" published by the long-time Wikipedian Christophe Henner in the days before the encyclopedia's 25th anniversary in January 2026. It frames a harder question than the project's birthday celebrations could absorb: whether the encyclopedic enterprise itself survives the AI era.

The excerpt, from Richard Cooke's The Last Best Place on the Internet: A Human History of Wikipedia (W. W. Norton, 2026), documents how the community's early confidence about AI gave way to prohibition and then to numerical decline. The stakes extend past one website. Wikipedia remains the reference layer beneath search engines and AI agents; if machine-generated text corrupts the source, the reasoning goes, no independent check remains on the agents' errors.

What did the Princeton study find?

A Princeton University study in October 2024 found that about 5 percent of nearly 3,000 newly created articles on English Wikipedia in August 2024 contained a substantial amount of AI-generated content. The telltale signs were stylistic. ChatGPT favored em dashes and curly quotation marks; sometimes prompt residue survived in the published text, such as "Here is your Wikipedia article."

Google had been training AI to write "Wikipedia-like articles" as early as 2018. At first, Wikipedians tried to shape the output — making it "acceptable and constructive," or deleting it when it proved otherwise. Editors weeded out generic fluff, such as praising a village for "its fertile farmlands, livestock, and scenic countryside," along with fabricated sources.

The cleanup failed. By March 2026, AI content violated the project's core principles so often that writing with large language models was banned outright, with machine use restricted to translation and minor proofreading.

Who warned about human control first?

In 2019, then-Wikimedia Foundation head Katherine Maher wrote an op-ed in The New York Times headlined "Without Humans, A.I. Can Wreak Havoc." Her diagnosis has aged well. "Too often," she wrote, "artificial intelligence is presented as an all-powerful solution to our problems, a scalable replacement for people."

A widely read essay titled "Death of Wikipedia" extended the argument into a spiral: as contributors reduce their output in response to AI pervading the platform, the temptation grows to replace their work with more AI. The inversion is the danger — not AI performing Wikipedia's housekeeping, but AI composing its articles while humans handle formatting and cleanup.

"The potential 'whittling down' each year of our pool of talented writers is the greatest threat to Wikipedia," another Wikipedian wrote, "as we must increase the number of editors, especially expert editors, to be able to fix the sprawling hellscape of weak, inaccurate and incomplete articles that drag down the project."

How bad are the traffic numbers?

The Pew Research Center quantified the encyclopedia's reach for the anniversary: between 2015 and 2025, Wikipedia was viewed 1.9 trillion times, an average of 508 million views per day. Those immense numbers contain an unmistakable arc of decline.

The click-through data is the sharpest evidence. One figure showed that the click-through rate from AI chatbots was only 0.33 percent; for Google search, it had been closer to 9 percent. For a model whose lifeblood was readers becoming editors, Cooke writes, the gap was a disaster.

Henner's essay called AI an existential risk and warned that Wikipedia could be remembered as a "historical artifact that AI once trained on … a footnote in the training data of models that have moved on without us." Fewer editors now perform more labor — "a smaller club working harder to maintain something fewer people see," as he put it. Once-sacrilegious ideas, such as paying editors, must now be considered.

Coverage elsewhere matched the tone. In Nature, Dariusz Jemielniak accused agentic AI of exploiting a "parasitic relationship" that "endangers the last bastion of freely accessible, human-curated knowledge." In WIRED, Stephen Harrison wrote that "economic precarity [has] trained many Gen Zers to see any unpaid labor as a scam" — "thankless work that AI companies strip-mine for a profit."

What happened to the cofounders?

Cooke frames the crisis against a longer intellectual history, and the divergent paths of Wikipedia's cofounders carry part of the argument. Twenty-five years on, Jimmy Wales lives in London, criticizes the MAGA movement, and remains, in Cooke's phrase, a Davos Man. Larry Sanger never left Ohio, homeschooled his children, became a vaccine skeptic during the pandemic, voted for Donald Trump, and in 2025 announced he had abandoned agnosticism for Christianity while writing a book-length philosophical proof of God's existence.

The pairing maps onto the philosopher Alasdair MacIntyre's 1990 taxonomy of "rival versions of moral enquiry" — encyclopaedia, tradition, genealogy. Wales, Cooke argues, stayed with the Enlightenment rationalism of "encyclopaedia"; Sanger returned to "tradition." Wikipedia's text-heavy format descended from that older encyclopedic lineage, created by two men who believed that discovering facts and learning from them was the central task of a life well lived.

Is this the end of the encyclopedic age?

Cooke's larger claim is that more than Wikipedia is at stake. What is losing relevance is the encyclopedic project itself — the idea, embodied in the "Tree of Diderot and d'Alembert" diagram of the French Encyclopédie, that ordering and connecting all knowledge reveals One Big Thing about everything. That ambition, he argues, was resuscitated in the computer age and culminates in the concept of AGI, artificial general intelligence, the point at which machines surpass every form of human understanding. "Human-written encyclopedias will still exist after AGI," he writes, "but only as secular scriptures in the mind of a digital god."

The longest-standing critique of total knowledge, he notes, is religious — from Pandora and Prometheus to the Tower of Babel — and it found strange modern form in the bio-artist Joe Davis's 2010s project Malus ecclesia: a plan to copy English Wikipedia into the DNA of a 4,000-year-old apple strain, breeding a literal Tree of Knowledge. Legal barriers around genetic modification kept it from fruition.

What is Wikipedia's actual dividend?

Cooke's closing counterweight is not content but process. At the annual Wikimania conference, Wikipedians have sustained a tradition, initiated by a Japanese Wikipedian, of wearing national dress at the opening ceremony — hanboks, kaftans, kimonos, serimpak headdresses shaped like ships from Borneo's seafaring tribes, chapan robes from Central Asia, ponchos from the Andes. When almost three-quarters of the nations on earth attend, the effect resembles the old hopeful canvases of the "Peoples of the World." The shared mission, he writes, is humble, bounded by human fallibility — "the same thing that ennobles them." The phrase the scene calls to mind is not Malus ecclesia but E Pluribus Unum.

Even that dividend depends on people showing up. With a 0.33 percent click-through rate from chatbots, a shrinking editor pool, and Henner's warning that the "AI summaries" explanation may undersell the change, the question for Wikipedia's next 25 years is whether unpaid human curation remains viable at all — and if not, what replaces it as the reference layer machines and people both trust.