Wikipedia volunteer stress is showing up in one of the encyclopedia’s most basic enforcement tools: the block button. Ryan McGrady, a senior research fellow at the Initiative for Digital Public Infrastructure at UMass Amherst, analyzed two decades of English-language Wikipedia moderation records and found patterns suggesting administrators are trading some openness for speed, prevention and article quality.
McGrady’s study examined roughly 20 million blocks, the public actions that stop a user from editing while still allowing them to read the site. Unlike YouTube, TikTok and other large platforms with mostly closed moderation systems, Wikipedia logs nearly every administrative action in public, which makes the machinery unusually inspectable.
The findings land as Wikipedia turns 25 and carries more weight than its early editors could reasonably have expected. The English-language site had 1 million articles in 2006 and passed 7 million in May 2025. The Conversation reported that it had more than 7.2 million English articles in June 2026. Its active editor count, however, has stayed roughly flat since 2010 at about 40,000.
What do Wikipedia blocks show about volunteers?
They show a system trying to protect a larger, more important encyclopedia with fewer people in key enforcement roles. Administrators, the trusted volunteers elected by the community to delete pages and block users, number slightly above 800, down from almost 1,800 in 2011, and not all of them are active.
McGrady found that the total number of blocks has climbed sharply in recent years, largely because administrators use bots to block proxies before they can be used for abuse. Proxies, including VPN services, can hide a user’s identity and are often used on Wikipedia to support manipulation or harassment. One bot, ST47ProxyBot, became the most prolific blocker in the site’s history, according to McGrady.
That kind of preemptive blocking can stop bad edits before they hit an article. It can also catch legitimate contributors who happen to be using the same privacy tools. McGrady wrote that the rise of AI agents and their possible disruption makes more preemptive blocking likely.
Wikipedia is also dealing with low-quality text generated by large language models, the systems behind chatbots and AI search tools. Those systems are trained to produce likely text responses from patterns in data, as explained in Kernel’s guide to how LLMs work. McGrady noted that AI companies have relied heavily on Wikipedia as training material, while chatbots and AI search can answer users directly and reduce visits to Wikipedia itself.
Why are Wikipedia administrators blocking people?
After removing proxy-related blocks from the analysis, McGrady found a shift in the reasons administrators give. In Wikipedia’s earlier years, most blocks were for vandalism, meaning deliberately bad or nonsensical edits. Today, vandalism accounts for about a quarter of blocks.
Blocks for promotional editing and sockpuppetry have increased. Sockpuppetry is Wikipedia’s term for one person using multiple accounts to influence content. McGrady interpreted those changes as evidence that Wikipedia’s value as an information source has made it a larger target for coordinated influence and reputation work.
The study also found more use of broad block rationales such as “disruption.” Wikipedia defines disruptive editing as a pattern that impedes article improvement or encyclopedia-building. McGrady said the trend is partly explained by boilerplate menu options that let administrators choose a standard reason instead of writing a custom one.
That efficiency has a cost. McGrady cited research showing that blocked users are more likely to leave or resist correction when sanctions seem vague or unfair. Longer average block durations, more preemptive blocking and more generalized explanations all point in the same direction: the system is favoring prevention and quality control over teaching new contributors how to stay.
The labor is uneven, too. For many years, the most active 10% of administrators made about 80% of blocks. That share fell to 37% in 2024, McGrady said, mostly because one prolific administrator changed activity.
McGrady’s conclusion is not that Wikipedia has failed. It is that the reader-facing encyclopedia is being protected by a smaller volunteer enforcement layer that is absorbing more of the strain. The open question is how long that bargain can hold for a site now used by readers, platforms and machine-learning systems as basic information infrastructure.
This story draws on original reporting from Techdirt.
