What Wikipedia Reveals About AI Overviews and Web Traffic: The Definitive Informational Case Study

For years, digital marketers and publishing executives debated the true impact of generative AI summaries on organic search traffic. While anecdotal reports pointed to shrinking click-through rates, clean causal evidence remained difficult to isolate due to simultaneous algorithm updates and seasonal shifts. Now, an econometric working paper from the University of Washington provides one of the first rigorous empirical answers by measuring the impact of Google AI Overviews on the web's largest informational repository: Wikipedia.

Authored by researchers Mehrzad Khosravi and Hema Yoganarasimhan, the study estimates that the default rollout of Google AI Overviews in the United States reduced monthly external search referrals to English Wikipedia by approximately 5%. Across English Wikipedia as a whole, this percentage drop represents an estimated loss of 100.27 million search referrals per month, amounting to roughly 1.2 billion fewer visits annually.

Measuring the net effect of AI search features on a website as massive as Wikipedia required an innovative econometric model. Because Google rolled out AI Overviews by default in the United States in May 2024 while holding back default deployment in continental Europe during the sample period, the authors constructed a quasi-experiment using Wikimedia's public monthly clickstream dataset.

The researchers paired 499,927 matching English-German article topics and 530,873 English-French pairs across a 13-month panel from December 2023 to December 2024. Because roughly 40% of English Wikipedia traffic originates in the United States, while German and French Wikipedia readers were predominantly outside default AI Overview regions, any divergence in referral trends between language pairs for the exact same topic after May 2024 reflected the presence of generative search summaries.

Using a Poisson pseudo-maximum likelihood difference-in-differences regression, the model revealed a 5.45% decline in external search referrals for English Wikipedia relative to German controls, and a 4.82% decline relative to French controls. Directional checks on Japanese editions showed an even steeper decline of 16.53%.

The University of Washington findings align with independent browsing telemetry. A behavioral study conducted by the Pew Research Center examining 68,879 Google searches discovered that when an AI summary appeared on the SERP, users clicked on traditional organic results only 8% of the time, compared to 15% on standard search pages without an AI summary.

More critically, the assumption that source citation links within generative answers compensate for lost organic traffic was contradicted by user telemetry: users clicked on citation links inside AI summaries on only 1% of visits. While Google leadership maintains that AI Overviews drive traffic to a broader diversity of websites, user behavior reveals that comprehensive informational overviews answer transactional and definitional questions on the search engine results page itself, leaving little incentive for readers to visit underlying publisher pages.

While human visitors are clicking less, machine scraping of Wikipedia has exploded. According to Wikimedia Foundation infrastructure reports, media file downloads surged by 50% as automated scrapers harvested datasets for AI training. Automated bots now account for 65% of the foundation's most resource-intensive core datacenter bandwidth and roughly 35% of total pageviews, forcing engineering teams to throttle or block over 1.5 billion unauthorized automated requests per day.

This dynamic creates an asymmetrical relationship: AI platforms consume record compute bandwidth to scrape encyclopedia content to power real-time answer engines, while simultaneously diverting human traffic away from the open web.

For commercial publishers, a 5% to 15% traffic drop impairs advertising monetization, with the authors calculating a hypothetical loss between $10.8 million and $37.1 million annually for an ad-driven site of Wikipedia's scale. But for non-profit entities like Wikimedia, the implications are structural. Wikimedia's draft strategic plan acknowledges that almost 90% of its visitors historically arrived via Google search, and overall human pageviews have dropped by 8% globally.

Wikipedia relies on an organic funnel: casual searchers become regular readers, readers become volunteer editors, and editors become financial donors. When conversational AI tools extract knowledge without downstream clicks, the volunteer pipeline that maintains and fact-checks the world's knowledge is severed at its source. As commercial search engines evolve into closed answer environments, preserving the incentives for human creators to generate original, verified information remains the internet's most pressing unresolved dilemma.

• What Wikipedia Reveals About AI Overviews And Web Traffic — Search Engine Journal
• Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia — arXiv / University of Washington
• Creating Helpful, Reliable, People-First Content — Google Search Central