Wikipedia Is Losing Readers to AI Overviews, and the Volunteer Cycle Is at Risk

Wikipedia editors long viewed generative AI as just another hype wave that would peak and then subside, as blockchain and the metaverse did. In a new piece, Wired describes how that confidence proved wrong. Search summaries such as Google AI Overviews, which launched in the US in May 2024, answer the question directly on the results page. The Wikipedia article that the summary largely draws on goes unread. Wired argues that the loss is far more serious than a drop in page views: the model on which the encyclopedia rests is in question.

Fewer Readers, More Load

The publicly available figures add up to an unpleasant picture. In October 2025, the Wikimedia Foundation reported that, after it revised its bot-detection algorithms, human page views came out roughly 8% lower than a year earlier. The foundation cited generative AI and social media as causes, since younger audiences increasingly get answers without visiting the site.

Pew Research Center data from 2025 shows how the effect works. When an AI summary appears in the results, users click a regular link in about 8% of cases. Without one, they do so about 15% of the time. Links inside the summary itself get clicked in about 1% of cases.

Infrastructure load is rising at the same time. Wikimedia previously reported that traffic for downloading multimedia has grown 50% since January 2024, and a significant share of that growth comes from automated scrapers collecting data to train models.

• Project scale: more than 65 million articles in 300+ languages.
• Human traffic: about −8% year over year, according to the Wikimedia Foundation (2025).
• Clicks when an AI summary is present: about 8% versus 15% without one (Pew).
• Multimedia download load: +50% since January 2024.

A Cycle That Depends on People

Wikipedia works like a conveyor belt. A reader spots an error and fixes it, gets drawn in, and over time becomes an editor. Some readers donate money, which funds the foundation, and the foundation does not run ads. Every link in the chain depends on a person seeing the page itself, not a paraphrase of it.

An AI summary cuts that path. The model takes text that volunteers wrote and proofread and hands it to the user with no attribution, or with a small link. The reader does not know that a community stands behind the answer, and gets no reason to join or pay. Yet the encyclopedia remains the main source of facts for the models themselves, so they depend on a project that they are depleting.

The foundation is trying to negotiate in a businesslike way. Through its commercial arm, Wikimedia Enterprise, it sells large companies structured access to its data, and in January 2026 Amazon, Meta, Microsoft, Mistral AI and Perplexity were reportedly added to its client list. That brings in money and takes some load off the servers. But a payment from a company does not replace a living editor who was once a casual reader.

What to Do

For those building products on top of LLMs, the takeaway is simple: a knowledge source needs to be sustained, not just cited. Prominent attribution with a link, use of the paid API instead of mass scraping, and honest disclosure of where data comes from all cost less than the degradation of the dataset that answer quality rests on.

For publishers and site owners, Wikipedia serves as an early warning. If the encyclopedia, with its reputation, is losing click-throughs, niche resources should track the share of their direct and subscriber audience, not just search traffic. For ordinary users, it is enough to get into the habit of opening the source under an AI answer, especially for dates, figures and contested topics. Those clicks are what now determine whether the next version of the encyclopedia will have anyone to write it.