This article was AI-generated as part of an experimental historical-content project. The date reflects the period being analyzed rather than the date the article was originally written.
Wikipedia turns 25 today. The Wikimedia Foundation marked it with a Wikipedia 25 campaign: a video docuseries about eight volunteer editors, a time capsule with Jimmy Wales recalling the first servers, and a birthday message that “knowledge is human, and knowledge needs humans.” The numbers are large. More than 65 million articles in over 300 languages, nearly 15 billion views a month, and nearly 250,000 editors making at least one edit each month.
Further down the announcement is a list that deserves as much attention as the celebration. Over the past year, the Foundation says, Ecosia, Microsoft, Mistral AI, Perplexity, Pleias and ProRata became partners of Wikimedia Enterprise, its commercial product for large-scale reusers, joining Amazon, Google and Meta. The Foundation describes Wikipedia as “one of the highest-quality datasets used in training Large Language Models.” Maryana Iskander, in one of her last weeks as chief executive before Bernadette Meehan joins on January 20, called it “integral to the architecture of the entire internet.”
For a company with a Wikipedia article, that list raises a practical question. Who is the article actually written for?
Two audiences, one text
The first audience is people: the investor, journalist or job candidate who reads the page. The second is machines: search engines, assistants and models that take the text, sometimes under contract, and turn it into Knowledge Panels, summaries and answers. Microsoft’s quote in the announcement adds a third, describing Wikipedia as knowledge that “people, and the agents working on their behalf” can rely on.
It is tempting to conclude that companies should start shaping their articles for the second and third audiences. Make the lead more extractable, the infobox more complete, the facts easier for a model to lift.
I think that is the wrong lesson, for two reasons.
Why the answer is still readers
The first is Wikipedia’s own rules. Articles are written for readers under neutrality and verifiability, by editors who have no obligation to the subject. A company that tries to optimize its article for machines is still trying to shape content it does not control, and the community has spent 25 years getting good at noticing that.
The second reason is less obvious. The machine audience values Wikipedia because it is written for readers by people who are not paid by the subject. Partners pay for access to that quality. If articles drifted toward what companies want models to say, the reason to license them would weaken. Companies benefit from the independence they are sometimes tempted to erode.
What companies can sensibly do
Accept that more of the article’s influence now comes from reuse than from visits, and read it with that in mind. The sentences most likely to be lifted are the lead, the infobox and dated facts such as leadership, headquarters and major events. Check those against reliable published sources. Twenty-five years of accumulation means some company articles still carry passages written under looser sourcing habits, or describe a business that has since changed. Where something is wrong or outdated, bring sources to the talk page and disclose your affiliation.
Two years ago I asked who would audit the sentences AI will repeat if Wikipedia’s editors could not agree on how to use language models. The anniversary answers part of that. The volunteers still do most of the checking, and the partner list shows how far their work now travels.