At 25, Wikipedia faces a double threat: the rise of AI and the decline of local media

When Wikipedia first appeared in 2001, it was a time when most had to be patient to find information – waiting for the loud scree and its answering beep as the computer connected, in earnest, to the Internet by dialing.

And the idea of ​​an open-source encyclopedia that could be updated by anyone in real time – or its equivalent in those pre-fibre-optic days – has prompted many questions and criticisms about how accurate that information can be.

Fast forward 25 years and Wikipedia is now the ninth most visited site on the Internet, with nearly 15 billion visitors each month, researching and editing its more than 65 million articles.

But despite its rapid rise in the early years and steady growth thereafter, Wikipedia is no longer as visible as it once was. Now, when Google asks a question, the top search result will be a Wiki link, but its AI will also compile an answer for you on top of it. And ChatGPT? That completely breaks Wikipedia.

Now, human visitors to the site have dwindled, dropping by about eight percent in parts of 2025, while large-scale linguistic models (LLMs) — chatbots or other types of AI that can condense words and information — crowd Wikipedia’s servers and use it as a training ground.

If these trends continue, they will coincide with the decline of Wikipedia’s leading local news outlets The future is “much worse than you think,” said Zachary McDowell, an associate professor of communication studies at the University of Illinois at Chicago and the author of the book. Wikipedia and the Representation of Truth.

Look at it as a pyramid of information accessibility, with LLMs at the top, Wikipedia in the middle and traditional media at the bottom, he said.

“Like you destroy all the secondary sources below and start destroying Wikipedia, what you have is something that will definitely come across,” he said.

“It has been shown many times that if you feed these [AI] programming data, when you feed them things created by other AI sources, they end up with what they call model collapse. ”

In layman’s terms, it’s considered digital reproduction – where AI-generated information is fed back to itself over and over again, increasing the number of errors and inaccuracies.

Man fails

The founder of Wikipedia, Jimmy Wales, expresses further concern about the financial implications of the growing demand that LLMs place on the online encyclopedia. He notes the need for more data and servers to support that additional traffic from “AI crawlers” was the reason behind the deals he announced with several AI partners in January, including Amazon, Meta and Microsoft.

“A typical donation to Wikipedia is $10 [US],” he said. “People don’t donate to fund OpenAI.”

But McDowell’s concern about those AI creeps making it more difficult to access neutral, accurate information? Wales said he does not share them when it comes to Wikipedia.

“We don’t listen to AI; Wikipedia is written by people and one of our strongest principles is that everything on Wikipedia needs to … have a quality source,” he said. “That’s the way to get into Wikipedia … human-created, human-reviewed information.”

The first draft of history

But McDowell and Wales agree that the focus of the media – especially on small local newspapers and news channels – affects Wikipedia, but in a larger sense, it also affects the ability to accurately capture the historical record.

The merger destroys the “neutrality” Wikipedia and traditional media are fighting for, McDowell said.

“These conglomerates, many of which have a very political bent, are now pushing a certain point of view and agenda.”

WATCH | How AI can be used to disrupt elections:

Can you spot a deepfake? How AI threatens elections

Fake AI-generated videos are used by scammers and cybercriminals, but what happens when they are created to disrupt elections? CBC’s Catharine Tunney reveals how technology can be used and looks at whether Canada is ready for a serious election.

Canada, more than 250 local news publication or broadcast closed between 2008 and Oct. 1, 2025, according to advocacy group News Media Canada.

““You know, it’s somewhat easier to write the history of a small town from 30 years ago than three years ago when, as happens in many places, the local newspaper has died,” said Wales in an interview with CBC Radio.

“That first draft of history hasn’t been taken in the first place. So there’s no doubt that’s a problem.”

‘Wikipedia detour’

AI, however, is accelerating what McDowell calls the “Wikipedia detour” — something that began a decade ago, as Google began summarizing answers on the search results page itself.

Taking Wikipedia out of the equation not only affects its ability to hire editors or contributors, it undermines digital literacy and knowledge, because people don’t see the citations that form the basis of these articles.

And they are encouraged to dig deeper, as what may begin as a search for black holes eventually brings you to the days of the upcoming lunar eclipse. Wikipedia can be a rabbit hole, but in a good way.

It’s how Jess Wade helped raise the profile of female scientists. The British physicist and assistant professor at Imperial College London in the UK has written more than 2,200 biographies on Wikipedia of women and other marginalized groups working in science over the past eight years, saying that most of his articles are visited as people investigate the concept of science and stumble upon the fact that it was founded by a woman.

And that improves their visibility in real time. During the COVID-19 pandemic, Wade and her colleagues added history about women and people of color on the front lines of the public health crisis. As time went on, he said, the “white old men” who appeared in the newspapers or on television began to be replaced by the experts he had installed.

“I was very impressed by how many broadcasters or teachers or lawyers use Wikipedia as the first point of call when looking for information.”

How AI can help

There are ways, however, that Wikipedia is exploring how to use AI to improve, including its search experience, since the interface has not changed much in recent years. That could include using a chatbot, Wales said.

And while the site’s 250,000 volunteer editors will still be the ones who fix it for us in the future, he said he could see AI doing some simple automation — fixing a dead link in an article, for example, by finding something else that a human could verify and decide to include.

“Automating some of the operational concerns of Wikipedia would be very useful and make it of higher quality.”