AI killing Wikipedia: traffic dropped dramatically

Wikipedia has seen a significant decline in direct traffic as more users access the world’s largest online encyclopedia through generative AI-powered chatbots.

The Wikimedia Foundation says that AI-powered chatbots and search engines that provide generalized answers, trained on Wikipedia articles, reduce the need to visit the site itself. This poses risks to the long-term sustainability of the project.

According to Marshall Miller, senior director of product at the Wikimedia Foundation, the decline in traffic is hurting the ability of volunteers and donors to support the work of developing and updating the encyclopedia’s content.

While generative AI models and search platforms are reducing Wikipedia’s direct traffic, its data remains critical to its operation. Wikipedia articles are among the most widely used datasets for training AI, and Google and other companies have been using them for years to build their own knowledge cards and answer snippets that distract users from the original site.

“Virtually all major language models are trained on Wikipedia data, and search engines and social networks rely on it to answer user queries. This means that people continue to read information created by Wikimedia volunteers without even visiting wikipedia.org. This human-generated knowledge has become even more important for the dissemination of reliable information on the Internet,” Miller said.

He also said that in May, Wikipedia recorded unusually high traffic, likely mostly from Brazil. This forced the Foundation to review its own bot detection systems.

“After the system update, we are seeing a decrease in real human traffic of about 8% compared to the same period in 2024. We believe this reflects the impact of generative AI and social networks on how people search for information. More and more often, search engines provide answers directly in the interface, often based on Wikipedia content,” he explained.

According to him, the Foundation imposes strict requirements on third-party bots that crawl Wikipedia: they must specify identifying information, adhere to robots.txt rules and limit the frequency of queries.

Recent studies confirm Wikimedia’s observations. For example, in July, the Pew Research Center found that only 1% of Google searches result in users clicking on a link in an AI-generated answer.

Search engines are increasingly using artificial generative intelligence to provide ready-made answers instead of redirecting to websites. Younger audiences are increasingly looking for information on video platforms rather than the traditional internet. This gradual shift is not limited to Wikipedia — many publishers and media platforms are also seeing a drop in traffic as users prefer to get answers directly from AI services and social networks.