Artificial intelligence in Wikimedia projects

• 1. Automated Tools And Bots

  ›

  In 2002, the rambot began transforming census data into short new articles about towns in the United States. This automated tool started the vast majority of town, city, and county articles on Wikipedia. The project required human approval before running, establishing a rule that bots must be supervised by people. Fighting vandalism became a major focus for machine learning tools over the following years. The ClueBot launched in 2007 to identify likely vandalism using simple heuristics. Its successor, ClueBot NG, arrived in 2010 and used an artificial neural network through machine learning. Machine translation software also entered the toolkit, helping contributors translate content across languages.

  12

• 2. Generative AI Controversies

  ›

  The public release of ChatGPT in 2022 sparked immediate experimentation with writing Wikipedia articles. A contributor named Pharos created the article Artwork title on the 6th of December 2022 using the initial draft from the model. Another editor noted that while the overview was decent, the citations were fabricated. In early 2023, the Wiki Education Foundation reported that some experienced editors found AI useful for starting drafts. They warned that ChatGPT tended to use promotional language and generate prose not encyclopedic in tone. By October 2024, Princeton University found about 5% of 3,000 newly created English Wikipedia articles came from AI. Ilyas Lebleu, founder of WikiProject AI Cleanup, observed unnatural writing patterns connected to the technology. He stated that AI could mass-produce content sounding real while being completely fake.

  378914

• 3. Policy Enforcement And Deletion

  ›

  August 2025 marked a turning point when the community adopted a policy allowing speedy deletion of suspected AI-generated articles. Editors recognize these texts by checking if citations relate to the subject or are entirely fabricated. The wording often reads like an LLM response, such as phrases stating Here is your Wikipedia article on. Other signs include excessive em dashes and overuse of the word moreover. Promotional material describing something as breathtaking also triggers flags. Curly quotation marks instead of straight versions appear frequently in suspect drafts. One article reviewer described being flooded non-stop with horrendous drafts created using AI. A guide titled How to Spot Signs of AI-Generated Writing emerged in August 2025 to assist editors. Jimmy Wales proposed using AI for customized feedback when drafts were rejected during discussions.

  1516171819

• 4. Wikipedia As Training Data

  ›

  A 2017 paper described Wikipedia as the mother lode for human-generated text available for machine learning. Subsets of the corpus became one of the largest well-curated datasets used to train every large language model through 2023 according to Stephen Harrison. The Wikimedia Foundation worried that attribution was missing from models like ChatGPT and Siri. While licensing allows use, credit must be given, implying violations occur without sourcing clarification. An 8% decrease in visitors appeared in 2025, attributed partly to generative AI popularity. Companies scraping data increased costs significantly for the foundation. They began looking at special licensing deals and paid APIs to recover expenses. The foundation expressed concern that people would not visit or donate if they did not know when benefiting from the project.

  526323334

• 5. Community Reactions And Future

  ›

  Longtime Wikipedians Richard Knipel and Andrew Lih warned about losing the original bold spirit in 2023. Stephen Harrison argued that AI should be embraced with guidelines like transparency and human supervision. In June 2025, testing a Simple Article Summaries feature met harsh criticism from editors calling it a ghastly idea. The rollout halted that same month after questioning the necessity of the feature. Ilyas Lebleu believed speedy deletion remained a band-aid while some articles stayed up for a week during discussions. Proposals included using AI to spot inconsistencies and summarize editor debates for newcomers. Volunteers feared the project might lose its original spirit despite productivity benefits. The community continues balancing innovation with the need to maintain trust and accuracy in their work.

  1011121320