How Accurate Is Wikipedia? What the Studies Found

Ask how accurate Wikipedia is and you get two confident answers: that anyone can edit it, so it can't be trusted, and that it beat Britannica in a study, so it can. Both are half-remembered versions of real research. This post goes through what the studies measured, what they found, and what they can't tell you. The short version: on general science and history topics, Wikipedia's error rate is in the same range as the encyclopedias it replaced; on medicine, its facts are mostly right but incomplete and sometimes out of date; and across every study, what a reader most needs to know is what the article leaves out.

The comparison everyone half-remembers

In December 2005 the journal Nature sent 50 pairs of science articles, one from Wikipedia and one from Encyclopaedia Britannica, to subject experts who did not know which was which. Forty-two usable reviews came back. The experts found 162 errors, omissions or misleading statements in the Wikipedia articles and 123 in Britannica's: about 3.9 per article against 2.9. Eight errors were judged serious, four on each side. Wikipedia's own summary of the study put it as coming "close to Britannica in terms of the accuracy of its science entries".

Britannica published a twenty-page rebuttal in 2006, objecting that some of the reviewed text came from its yearbooks or from a student edition for younger readers rather than the encyclopedia proper, that omissions had been counted after reviewers were sent the 350-word introduction of a 6,000-word article, that differences of opinion over phrasing had been counted as inaccuracies, and that Nature had refused to release the full reviewer reports. Nature stood by its findings. Two things survive the argument. Both encyclopedias contained errors, at a rate any reader would find sobering. And the gap between them was far smaller than the reputation gap suggested.

Where Wikipedia scored well

A 2012 pilot study run by the e-learning company Epic with the University of Oxford, and funded by the Wikimedia Foundation, compared 22 article pairs in English, Spanish and Arabic, across the humanities, social sciences, natural sciences and medicine. Thirty-three academics, all with doctorates, reviewed the pairs blind, with the articles stripped of anything that gave away their origin. Across the whole sample the Wikipedia articles scored higher than the comparison encyclopedias (Britannica, Enciclonet, Mawsoah and the Arab Encyclopaedia) on accuracy, references, readability and overall judgment. Language by language it was closer: against Britannica and Enciclonet, Wikipedia did better on accuracy, references and overall judgment but no better on style or overall quality, and the two Arabic encyclopedias beat it on style and matched it on everything else. The authors are careful to call it a pilot: 22 pairs cannot be generalised, and none of the encyclopedias, Wikipedia included, was rated highly on whether academics would cite it.

The largest study of Wikipedia's scientific sourcing took a different angle. Nicholson and colleagues analysed 1.9 million Wikipedia articles citing 824,298 scientific papers and asked whether those papers had been supported, contradicted or simply never tested by later research. Most of the cited papers, 57%, had never been tested either way. But that is a better record than the scientific literature as a whole, where 85% of papers are untested, which suggests Wikipedia's editors tend to cite work that others have gone on to check.

The people best placed to judge a citation are the authors of the work cited. In a 2025 survey of researchers whose publications were cited on Wikipedia (750 answered, of 21,854 asked), the authors rated the accuracy of the citation at 7.65 out of 10 on average and how well the article represented their subject at 7.03. Scholars also cite Wikipedia back: a 2014 review in the BMJ found 2,049 citations of Wikipedia in 1,433 peer-reviewed papers across 1,008 health science journals, more than half of them for definitions or descriptions, and not only in journals with low impact factors. The authors asked editors and reviewers to use caution, since permanent evidence-based sources are available.

Where it scored badly

The counter-evidence comes mostly from medicine and from close reading. A 2008 study by Lucy Holman Rector compared nine Wikipedia history articles with Britannica, the Dictionary of American History and American National Biography Online. Wikipedia's accuracy came out at 80% against 95% to 96% for the others, with inaccuracies in eight of the nine articles. Nine articles is a small sample, but the gap is not.

In 2014, Hasty and colleagues took the Wikipedia articles on the ten most costly medical conditions in the United States, had two blinded reviewers list every assertion in each, and checked each assertion against the peer-reviewed literature. Nine of the ten articles showed statistically significant disagreement with the literature: coronary artery disease, lung cancer, major depressive disorder, osteoarthritis, chronic obstructive pulmonary disease, hypertension, diabetes, back pain and hyperlipidemia. The authors' conclusion was blunt: caution when using Wikipedia to answer questions about patient care.

Currency is a separate problem from accuracy. When the US Food and Drug Administration issued new safety warnings for 22 drugs in 2011 and 2012, Hwang, Bourgeois and Seeger checked how quickly the Wikipedia pages caught up. Forty-one percent were updated within two weeks. For drugs treating common conditions the figure was 58%; for rarer conditions, 20%. A year later, 36% of the pages had not changed at all.

Accurate is not the same as complete

The most useful single study for a reader is the one that measured both. Kräenbring and colleagues checked German Wikipedia's articles on 100 drugs, and English Wikipedia's on 50, against standard pharmacology textbooks. The German articles were 99.7% accurate: what they said was almost always right. They were 83.8% complete: what they left out was substantial, and it was uneven, with 91% of the textbook's information on indications present but only 68% on pharmacokinetics. The English articles were more complete, at 93.1%. The authors judged 62.5% of the missing material not relevant for teaching, and concluded that Wikipedia was an accurate and comprehensive enough source for medical undergraduates.

That pattern, right about what it says and silent about the rest, runs through the whole literature, and it is the pattern that error counts cannot see. A reviewer counting mistakes gives an article that omits half a subject a clean score. The omission problem is also where the language editions come in. English Wikipedia is not the fullest version of most widely read topics: in a measurement of the most-read articles in 20 language editions, 60% had a fuller article in another language and 32% had one at least twice as full. An English reader checking an English article for errors would not find the missing half, because it isn't there to check.

The editions also differ in how they source what they do include. A 2023 study of more than five million articles found that sources flagged as unreliable on English Wikipedia persist in other language editions, particularly smaller ones, and a case study of the Chinese, Russian and Swedish editions found that the reliability of their references differs from edition to edition.

What the studies can and can't tell you

Every study here is a sample, most of them small: 42 pairs, 22 pairs, nine articles, ten conditions, 22 drugs. Most are a decade or more old, and the articles they measured have been edited thousands of times since. A study of Wikipedia is a photograph of a moving object. What the photographs agree on:

• On broad, heavily read topics, Wikipedia's error rate is comparable to the reference works it replaced, and its sourcing is better than its reputation.
• On specialist medical detail, the facts present are usually correct, but the articles are incomplete and can lag the evidence by months or years.
• Quality follows attention. Articles with many readers and editors are updated quickly and checked often; articles on rarer subjects are not.
• The largest gaps are omissions, and the largest omissions are often filled in another language.

How to read Wikipedia well

Treat the citations as the product. A claim with an inline reference to a source you can open is worth checking; a claim without one is an unknown. On medical topics, look at when the article was last substantially edited and whether it reflects the current guideline. On any topic tied to a country or a language, open the Languages panel and compare the size of the other editions, since the fuller article is often not the English one.

That last habit is the reason SuperCharged Wiki exists. Each of its articles is written from every language edition of the topic, marks each paragraph with the editions it comes from, and keeps the citations pointing at the sources those editions cite, so the checking habits above work the same way. Whether that makes it more accurate than Wikipedia is a question with a longer answer; what it does make it is more complete, which is the gap these studies keep finding.