Remember When Wikipedia Was the Biggest Danger to Students?

A history of twenty years of panic

Photo by Stephen Harlan on Unsplash

In winter of 2007, a professor at Middlebury College sat grading final papers for a course on early Japanese history and kept reading the same wrong fact. He found it in six different essays. Neil Waters traced every instance to one Wikipedia article, and his department decided to act.

That January, Middlebury’s historians voted to bar students from citing Wikipedia in papers and on exams, and they ordered the policy printed on every syllabus. The chair, Don Wyatt, explained the reasoning in plain terms. Educators, he said, are “in the business of reducing the dissemination of misinformation.”

The decision made national news. Waters later said the coverage caricatured him as “the dinosaur trying to kill Wikipedia,” though the department had banned only citation, not use.

Cathy Davidson, a Duke scholar of digital learning, wrote to Middlebury’s president the moment she heard, half-alarmed that a national trend was forming. “Is Wikipedia really that bad?” she asked, then answered her own question by proposing that studying what Wikipedia does well and badly belonged inside the research-methods curriculum rather than outside the gates.

Two years earlier, journalist John Seigenthaler had discovered a Wikipedia biography that falsely implied he had taken part in the assassinations of John and Robert Kennedy. The hoax sat on the site for months before anyone caught it. If that could happen to a public figure with the means to notice and object, what was happening on the ten thousand pages no one was watching?

In that climate the encyclopedia looked like a membrane between students and real knowledge, and the responsible thing was to keep the two apart. Yet the panic was never really about the error rate. Underneath it sat a deeper disagreement about what makes knowledge trustworthy in the first place, and the next twenty years would be spent working that disagreement out.

The first reckoning

In December 2005, only weeks after the Seigenthaler affair, the journal Nature published a bombshell by reporter Jim Giles. Giles took Jimmy Wales at his word. Wales, who co-founded Wikipedia in 2001, had set the project a goal of “Britannica-or-better quality,” so Giles arranged a test.

He sent 42 matched pairs of science articles, one from Wikipedia and one from Encyclopaedia Britannica, to subject experts who were not told which was which, and asked them to mark the errors.

Wikipedia averaged about four errors an article. Britannica averaged three. On serious errors the two tied, four apiece. The headline finding was that the amateur upstart had come within reach of the professionals. The equally remarkable finding, that actually rearranged the furniture, was about Britannica. The gold standard was not made of gold.

Michael Twidale, an information scientist at the University of Illinois, captured the surprise for Nature: readers would be startled, he said, “to see how many errors there are in Britannica.”

Britannica did not take it lightly. It issued a twenty-page rebuttal titled “Fatally Flawed,” accusing Nature of grading it against a children’s edition and of counting mere omissions as errors. Nature published its methodology and held its ground. The posture of Britannica’s spokesman was its own small comedy: Tom Panelas allowed, sardonically, that the open encyclopedia had admirably thorough coverage of subjects like extreme ironing.

The episode looks like the first time the Wikipedia question was argued with evidence rather than intuition, and it exposed a premise everyone had been carrying without noticing. Critics had not been comparing Wikipedia to a real reference work with its real flaws. They had been comparing it to an imaginary Britannica that never existed.

Once the phantom was retired, the question changed from whether Wikipedia was perfect to whether it was good enough, and good enough for what.

The historian’s verdict

The most thoughtful early answer came from inside the academy that was busy banning the thing. In 2006, the historian Roy Rosenzweig, who founded the Center for History and New Media at George Mason University, published a long essay in the Journal of American History asking whether history could be open source.

He approached Wikipedia the way a historian approaches any artifact, curious about how it was made. The scale impressed him: by his account the most-read history project on the web, produced almost entirely by volunteers. As of late 2005, he noted, the entire operation ran on two paid employees.

His verdict on accuracy was measured and, for the time, generous. He found Wikipedia often comparable to traditional references and frequently ahead of Britannica on sheer coverage, with biographies especially strong and thematic essays weaker. The most revealing part of the essay, though, was diagnostic rather than evaluative.

History, Rosenzweig observed, is “a deeply individualistic craft.” The single-authored work was the disciplinary standard; collaboration of the kind that was routine in the sciences was rare in history, and the profession was built around attributing this idea to that named scholar. Wikipedia offended all of it. It had no byline, no author to praise or blame, no version that stayed still long enough to cite.

Rosenzweig had put his finger on the real source of the unease. The objection to Wikipedia was not only that it might be wrong. It was that it produced knowledge in a way the academy did not recognize as legitimate, by anonymous consensus instead of credentialed authorship.

The panic was a collision between two models of trust. One located authority in the qualifications of the author. The other located it in the transparency of the process, in the visible edit history and the citations a reader could follow for herself.

Using Wikipedia well would come to mean reading the second kind of trust, learning to judge a process rather than a pedigree. That was a genuinely new skill, and in 2007 almost no one was teaching it.

From abstinence to literacy

While the historians debated authority, librarians did something more practical. They watched what students actually did, and they tried to help them. It was a slow and good-natured professional surrender to reality.

Sook Lim, an information scientist, surveyed undergraduates in 2009 for the Journal of the American Society for Information Science and Technology. Her students used Wikipedia constantly, mostly to check a fact or get oriented on a new topic, and they held oddly cautious opinions about its quality even as they leaned on it daily.

Lim’s conclusion pointed away from prohibition: educators and librarians, she argued, should teach students to use Wikipedia well rather than forbid it outright.

Her contemporaries reached the same place. Diane Murley, a law librarian, wrote a piece for the Law Library Journal whose title captured the change in mood, “In Defense of Wikipedia,” and showed colleagues how the site’s own edit histories and talk pages could be turned into a lesson in evaluating sources.

Surveys of faculty, including across the California State University system, found professors warming to the encyclopedia and beginning to build assignments around it.

By the middle of the next decade the scattered findings had become a literature, and the literature had a verdict. In 2015, Mostafa Mesgari and colleagues published a systematic review in the same journal, synthesizing 110 peer-reviewed studies of Wikipedia’s content.

Their summary was that the articles were, on balance, fair-minded and largely accurate, even where some stayed shallow or clumsily written. The stigma, by then, had outrun the evidence by a full decade.

The live question was no longer whether students used Wikipedia, since the answer was everyone, including the faculty who warned against it. The question was what universal use obligated teachers to do, and the librarians kept arriving at the same answer the encyclopedia’s own spokespeople had given during the Middlebury affair, when Sandra Ordonez described the site: the ideal place to begin research and a poor place to end it.

From reading to building

Then came the move that turned the whole anxiety inside out. If Wikipedia was a serviceable place to read, it proved to be an extraordinary place to write, and the act of writing it changed students in a way nobody had quite predicted.

Brenna Clarke Gray, an English instructor at Douglas College in British Columbia, tried an experiment that many teachers would go on to repeat. Instead of assigning essays that only she would read, she had her students write Wikipedia articles about Canadian writers.

The result astonished her. Students who normally turned in work with no citations were suddenly up until two in the morning, sourcing every sentence and rewriting for clarity, because their readers, the Wikipedia community, were harder graders than she was.

An essay for a professor is performed once and filed. An edit to a public encyclopedia is inspected by strangers who can revise or revert it within the hour. The audience was real, and a real audience demanded real evidence.

This is, on reflection, the most profound result in the entire literature. The thing educators feared would make students lazy turned out, once students wrote it rather than merely read it, to make them rigorous. The care a grade had failed to produce, an audience supplied for free.

Writing for the public gave people a stake in getting it right, along with a reason to ask the questions information literacy is forever trying to instill: where did this claim come from, and how would I know if it were false.

Institutions noticed and built on it. The Wiki Education Foundation, spun out of the Wikimedia Foundation, has supported college instructors running Wikipedia assignments since 2010. By mid-2023 its students had added more than 100 million words to English Wikipedia across roughly 6,200 courses, about twice the content of the last print edition of Britannica.

The program now supplies close to a fifth of all new active editors on English Wikipedia, and because its participants include more women and are more racially diverse than the existing volunteer base, has started to repair one of the project’s most-studied blind spots.

The scholarship registered a reversal. Benjamin Mako Hill and Aaron Shaw, surveying two decades of Wikipedia research, note that the academic literature on teaching with the site no longer asks whether students use it or how to stop them. It asks how to get them contributing. The encyclopedia had gone from contraband to curriculum.

The world responds

The final turn in the evidence came from a discipline with no particular stake in libraries or pedagogy. Economists had noticed that Wikipedia sat near the top of nearly every search result on earth, which made it a rare thing: a lever long enough to move the real world, and one a researcher could actually pull. They stopped asking whether Wikipedia was accurate and began measuring what it caused.

The boldest experiment belonged to Neil Thompson of MIT and Douglas Hanley of the University of Pittsburgh. They commissioned chemistry graduate students to write brand-new Wikipedia articles on scientific topics the encyclopedia was missing, then flipped a coin. Half the articles went live and half stayed in a drawer. Over the following months the scientific papers being published began to echo the language of the articles that had been posted, while the held-back articles left no trace in the literature at all.

Thompson’s summary was bracing in its directness: Wikipedia, he said, was “actually helping to shape science.” By his estimate, a third of the correlation between Wikipedia and the published literature was genuinely causal, and the effect ran strongest for scientists with the least access to expensive journals.

A neighboring study explained why that mattered for the open knowledge movement. Misha Teplitskiy and colleagues found that a journal’s openness sharply raised the odds of its being cited on Wikipedia, which suggested that one of the chief payoffs of open-access publishing is the distance it lets research travel, through an intermediary like Wikipedia, into the hands of a general public.

The influence did not stop at the edge of science. In a field experiment eventually published in the Journal of Economics & Management Strategy in 2023, Marit Hinnosaar and colleagues added a couple of paragraphs and a photograph to the Wikipedia pages of randomly chosen Spanish cities, mostly translated from existing Spanish articles.

Overnight hotel stays in the treated cities rose about 9 percent in tourist season, with the largest gains on the pages that had started out thinnest. A few paragraphs of volunteer prose moved real money through local economies.

Notice how far the conversation had traveled. It began with a professor worried that a free website would corrupt six term papers. It arrived at peer-reviewed evidence that the same website was steering the direction of scientific research and the flow of tourist euros.

The question was no longer whether Wikipedia could be trusted. It was that Wikipedia had become infrastructure, and infrastructure has consequences whether or not we approve of it.

Where the voices leave us

Two decades on, the verdict is in the record, written in Wikipedia’s characteristic style. The encyclopedia’s own article on its reliability traces the arc without ceremony: heavy criticism in the 2000s, steady improvement, and general praise for the English edition by the late 2010s and early 2020s. Studies that once felt provocative are now the baseline.

Quality is still uneven and the measurements still quarrel. This is not a coronation, and an honest account keeps the caveats in view.

Jonathan Jarry, writing for McGill’s Office for Science and Society, points out that scholars assessing Wikipedia’s health articles often “want to have it both ways,” faulting an entry for leaving things out while also faulting it for being too technical, and that a 2023 study which found chemotherapy-drug articles thin on basics must be read against earlier work that found drug coverage accurate next to textbooks.

What a study concludes depends heavily on which articles it samples and on what its authors want the encyclopedia to be. The fair summary is also the oldest one: Wikipedia is reliable enough to begin with and uneven enough that it can never be where you stop.

The scale, though, has passed beyond argument. On its 25th birthday in January 2026, the Pew Research Center reported that web crawlers and other automated agents pulled more than 88 billion views from Wikipedia in 2025 alone, even as human readership slipped about 8 percent, a decline the Wikimedia Foundation attributes to AI summaries that answer the question before anyone clicks.

The reference work once kept away from students now trains the machines those students use amidst a new era of panic, with a new villain. Anusha Alikhan, the foundation’s communications chief, puts it plainly: Wikipedia plays an essential role in “training almost every large language model.”

The Middlebury historians were not foolish to worry about a student who treats any encyclopedia as the last word. They were arguing about the wrong thing. The error was never reaching for Wikipedia; it was stopping there.

And the cure was the one its volunteers had been repeating from the beginning: treat it as the best-stocked front door in the history of research, then read it critically and follow its citations back to the source. Two decades of scholarship, from a single Nature reporter to a roomful of economists, converged on that unglamorous instruction, and it taught information literacy better than any prohibition could have.

There is a lesson in the shape of the whole episode, and the next panic will test whether we learned it. Every era greets a new technology for spreading knowledge with the same certainty that it will make the young credulous and lazy.

The encyclopedia that was going to rot students’ minds became the most trusted layer of the world’s information supply, and it now checks the work of the machines we have decided to fear next. The tool always demanded judgment. The Wikipedia story is the proof that judgment meant learning to use it well.