Wikipedia

Wikipedia is a free online encyclopedia that anyone can edit, and millions of people already have¹. It is written and maintained by human volunteers known as Wikipedians, who collaborate with almost no assistance from artificial intelligence¹. The encyclopedia is hosted by the nonprofit Wikimedia Foundation, the organization responsible for hosting and protecting Wikipedia and its fellow Wikimedia projects². Unlike most other high-traffic websites, it does not run advertisements and is free to access³. Its fundamental principles are summarized in five pillars¹.

Since its creation on 15 January 2001, Wikipedia has grown into the world's largest reference website, attracting over a billion visitors each month¹. As of September 2026 it comprises 364 language editions holding 68,348,064 articles in total, 77 of them with more than 100,000 articles⁴. The English Wikipedia alone contains 7,240,970 articles and 262,745 active contributors in the past month, on the project's 2026 figures¹. It is among the top-ten most-visited global websites and the only one of them run by a nonprofit⁵.

The Wikimedia Foundation describes Wikipedia's human-created and curated knowledge as more valuable than ever in the AI era⁵, and the encyclopedia is one of the highest-quality datasets used in training large language models⁵. Pew Research Center records Wikipedia among the top sources mentioned in Google search results and notes that it helps train the large language models behind many artificial intelligence technologies³. On the Foundation's 2026 anniversary account, all of Wikipedia's knowledge is created and maintained by a global community of nearly 250,000 volunteer editors who write, edit and fact-check it to rigorous standards on neutrality and reliability⁵.

Wikipedia grew out of Nupedia, a more structured free encyclopedia conceived as a way to allow easier and faster drafting of articles and translations⁶. Nupedia was founded by Jimmy Wales, underwritten by Bomis and led by Larry Sanger as editor-in-chief; it operated from March 2000 until September 2003 and is best known today as the predecessor of Wikipedia⁷. Its articles were written by experts and passed through a seven-step approval process before posting, rather than live wiki-based updating⁷. Production was nevertheless extremely slow: despite a mailing list of over 2000 interested editors and Sanger serving full-time as editor-in-chief, only 12 articles were written during the first year⁶.

Wikis had already been used elsewhere on the web to organize knowledge, and the idea of a wiki-based companion to Nupedia was seeded by two conversations — one between Sanger and Ben Kovitz, another between Wales and Jeremy Rosenfeld⁶. Wales put a wiki online on 10 January 2001 under the nupedia.com domain, and it moved to a new wiki under the wikipedia.com domain on 15 January⁶. In 2001 the licence for Nupedia was changed to the GFDL, and Wales and Sanger launched Wikipedia as a complementary project, using an online wiki as a collaborative drafting tool⁶. The project formally opened on 15 January, the day known as Wikipedia Day, with the first international editions — German, Catalan, French, Swedish and Italian — created between March and May⁶.

Pew Research Center records that the earliest edit found on Wikipedia's homepage announced the new WikiPedia³. The contrast with Nupedia was striking: 21 articles in its first year against 200 articles in Wikipedia's first month and 18,000 in its first year⁷. The first subdomain for a non-English Wikipedia, deutsche.wikipedia.com, was created on 16 March 2001, followed within hours by Catalan⁶. By 2003 the English Wikipedia had passed 100,000 articles, while the next largest edition, the German Wikipedia, passed 10,000⁶. On 28 March 2008 the project exceeded 10 million articles across all 251 language editions⁶.

Wikipedia runs on openness: anyone can edit, and its fundamental principles are summarized in its five pillars¹. Editors are human volunteers who collaborate with almost no assistance from artificial intelligence, and the quality of the content matters more than the expertise of whoever contributes it, provided it conforms with policies including verifiability by published reliable sources¹. Content must also follow the neutral point of view policy, which requires representing fairly, proportionately and, as far as possible, without editorial bias all the significant views that reliable sources have published on a topic⁸.

Wikipedia's content became licensed under a Creative Commons Attribution-ShareAlike licence in 2009⁶. The change followed an invitation from the Wikimedia Foundation's Board of Trustees, which in 2009 asked the community to vote on licensing Wikimedia material under CC-BY-SA while retaining dual licensing with the GNU Free Documentation Licence (GFDL)¹⁰. The vote ran from 12 April to 3 May 2009 and was open to all users, excluding bots, who had made at least 25 edits to any Wikimedia project before 15 March 2009¹⁰.

The motivation was practical: the GFDL includes potentially onerous provisions, such as the requirement to include the full licence text with each copy, and those requirements impede re-use of both text and multimedia¹⁰. Under the Terms of Use in force today, contributors license the text they submit under the Creative Commons Attribution-ShareAlike 4.0 International licence and the GFDL, and reusers may comply with either licence or both¹¹. Dual licensing with the GFDL was retained, so Wikimedia material remains available under both licences¹⁰.

The Wikimedia Foundation is the nonprofit responsible for hosting and protecting Wikipedia and the other Wikimedia projects, and it is a United States 501(c)(3) tax-exempt organization². It was founded on 20 June 2003, the same day the sister project Wikiquote was created⁶. Since 2003 it has supported the hundreds of thousands of volunteer editors who edit, expand and curate the Wikimedia projects². Governance accordingly rests with the volunteer community: editors are expected to seek consensus, avoid personal attacks and edit wars, and work under principles summarized in the five pillars⁹.

For serious conduct disputes that the community cannot resolve, English Wikipedia relies on its Arbitration Committee, the panel of editors responsible for conducting its arbitration process, with the authority to impose binding solutions to disputes between editors¹². Arbitrators are volunteer users — usually experienced editors and administrators — whom the community of editors at large elects to resolve the most complex or intractable disputes and to oversee the few areas where access to non-public information is a prerequisite¹². Beyond dispute resolution, the Committee determines which editors have access to CheckUser and Oversight permissions, and it considers certain matters where exceptional factors such as privacy preclude a public hearing¹². Formal elections for a board of the Foundation and for an Arbitration Committee on the English Wikipedia began in 2004⁶.

The table below sets out Wikipedia's scale as dated snapshots, reflecting the as-of dates of the underlying sources⁴. It combines September 2026 counts from Meta-Wiki, 2026 figures from the Wikimedia Foundation and the English Wikipedia project page, and 2024 editing statistics from Pew Research Center³. The English Wikipedia passed 500,000 articles in March 2005, one million in March 2006 and three million in August 2009, and by 9 March 2019 the project exceeded 50 million articles across all 309 language editions⁶. The German Wikipedia became the second edition to exceed one million articles, on 27 December 2009⁶. The Arabic Wikipedia, launched on 9 July 2003, has 1,332,681 articles, 2,992,083 registered users and 56,632 files as of September 2026, ranking 15th among Wikipedias by article count and 4th by depth¹³.

The first major Wikipedia scandal, the Seigenthaler incident, occurred in 2005, when a well-known figure was found to have a vandalized biography that had gone unnoticed for months⁶. In its wake, the project's first policy and system changes designed to counter that form of abuse were established: a Checkuser privilege policy update to assist sock-puppetry investigations, a new semi-protection feature, a stricter policy on biographies of living people, and the tagging of such articles for stricter review⁶.

In December 2005, Nature published an investigation finding that Wikipedia came close to Britannica in terms of the accuracy of its science entries¹⁴. Forty-two expert reviewers carried out the comparison, and after tallying their results Nature recorded errors — the great majority of them minor — at a rate of about three per online Britannica item and about four per Wikipedia item¹⁵.

Britannica's formal response, published in March 2006 under the title Fatally Flawed, disputed the study and demanded a retraction¹⁶. Britannica argued that the numbers buried deep in the article told the opposite story: Wikipedia in fact had a third more inaccuracies than Britannica¹⁶. Nature rejected the critique, stating that its comparison was unbiased and that it stood by the story¹⁵. The exchange became the defining early episode in the debate over Wikipedia's reliability¹⁶.

Peer-reviewed research in EPJ Data Science notes that although everyone can participate, Wikipedia's editor community is known to have narrow diversity, with a majority of white male editors¹⁷. The same study found that women in Wikipedia are more notable than men, an outcome its authors interpret as a subtle glass ceiling effect¹⁷. It also documents linguistic bias, with abstract terms tending to describe positive aspects in the biographies of men and negative aspects in the biographies of women¹⁷.

Article counts can mislead: Lsjbot, an automated program built by linguist Sverker Johansson, creates articles in Cebuano and Swedish³. Cebuano — a language spoken in the southern Philippines — now has the second-most Wikipedia articles of any language, despite having relatively few active users³.

Editor retention offers another signal: the number of regularly active editors on the English Wikipedia peaked in 2007 at more than 51,000 and has been declining since⁶. Pew Research Center reports that Wikipedia's summary statistics page lists more than 600,000 active users — account holders who have made at least one edit in the last 30 days — with just under half of them, 45%, contributing to English Wikipedia³. Nearly 40% of the edits made in 2024 came from Wikipedia-approved bots that operate alongside human editors³.

After revising its bot-detection systems, the Wikimedia Foundation reported in 2025 that human pageviews on Wikipedia had declined by roughly 8% compared with the same months in 2024¹⁸. The Foundation attributes the decline to the impact of generative AI and social media on how people seek information, especially with search engines providing answers directly to searchers, often based on Wikipedia content¹⁸. It warns that with fewer visits to Wikipedia, fewer volunteers may grow and enrich the content, and fewer individual donors may support the work¹⁸.

The same technology depends heavily on the encyclopedia: Wikipedia is one of the highest-quality datasets used in training large language models⁵, and its knowledge powers generative AI chatbots, search engines and voice assistants⁵. Pew Research Center records Wikipedia as one of the top sources mentioned in Google search results and as a training source for the large language models behind many artificial intelligence technologies³. In the Foundation's account, Wikipedia and its human-created and curated knowledge have never been more valuable than in the AI era⁵.

Both the Foundation and the editors have responded to these pressures⁵¹⁹. The Foundation's AI strategy directs future investments and developments in AI to support the human contributors at the core of Wikipedia⁵. Wikimedia Enterprise, its commercial product for large-scale reusers and distributors of content from Wikimedia projects, has gained new partners including Ecosia, Microsoft, Mistral AI, Perplexity, Pleias and ProRata, joining existing partners such as Amazon, Google and Meta⁵. Wikipedia's editors maintain an AI Cleanup project, which records that since 2022 large language models have become a convenient tool for creating text at scale but virtually always fail to properly source claims and often introduce errors¹⁹. Since March 2026, the Wikipedia guideline WP:LLM has prohibited the use of LLMs to generate or rewrite article content, with limited exceptions for translation and basic copyediting¹⁹.

Wikipedia's usefulness to a reader depends on its sourcing: the encyclopedia requires that content conform with its policies, including being verifiable by published reliable sources¹. Its verifiability policy holds that even if an editor is sure something is true, it must have been published in a reliable source before it can be added²⁰.