{"id":255135,"date":"2026-09-11T15:11:37","date_gmt":"2026-09-11T12:11:37","guid":{"rendered":"https:\/\/digitrendz.blog\/?p=255135"},"modified":"2026-09-11T15:11:37","modified_gmt":"2026-09-11T12:11:37","slug":"wikipedia-data-shows-ai-overviews-impact-web-traffic","status":"publish","type":"post","link":"https:\/\/digitrendz.blog\/z\/quick-reads\/255135\/wikipedia-data-shows-ai-overviews-impact-web-traffic\/","title":{"rendered":"Wikipedia Data Shows AI Overviews Impact Web Traffic"},"content":{"rendered":"
\u25bc Summary<\/summary>

– A revised University of Washington working paper estimates that Google’s AI Overviews reduced monthly search referrals to English Wikipedia by approximately 5% after becoming the default in the U.S.
– The study utilized a difference-in-differences model comparing traffic to English, German, and French Wikipedia editions to isolate the impact of the feature rollout.
– Researchers calculated that the decline amounts to roughly 100 million fewer monthly referrals, totaling about 1.2 billion annually for direct search traffic.
– Earlier versions of the paper reported a steeper 15% decline using daily pageviews, but the updated analysis relies on monthly referral data for greater accuracy.
– Google disputes the methodology’s ability to isolate its specific feature, while Wikimedia notes that many users access content via AI tools without visiting the main site.
<\/p><\/details>\n\n

<\/p>\n\n

New Estimates on AI Overview Traffic Impact<\/h2>\n\n

A recent working paper from the University of Washington<\/a> suggests that Google<\/a>\u2019s AI Overviews<\/strong> have significantly altered search traffic patterns for English Wikipedia<\/a><\/strong>. The study, authored by Mehrzad Khosravi<\/a> and Hema Yoganarasimhan<\/a>, estimates that monthly search referrals to the site dropped by approximately 5%<\/strong> after the feature became the default in the United States during 2024. This figure represents a revision from earlier reports that suggested a steeper decline. The paper was last updated on September 2 and remains unpeer reviewed.<\/p>\n\n

The authors refined their methodology to isolate the impact of the AI feature more precisely. While Google disputes whether the combined referral metric can accurately attribute changes solely to its own product, the researchers argue that the timing of the U. S. rollout provides a clear natural experiment. They note that other search engines account for a minimal share of traffic, making the aggregate data a reliable proxy for the shift caused by AI Overviews<\/a>.<\/p>\n\n

Methodology and Comparative Analysis<\/h2>\n\n

To measure this impact, the researchers utilized Wikimedia<\/a>\u2019s monthly clickstream data, which tracks article-level visits across various language editions. By comparing English Wikipedia with German and French editions, they created a control group. During the study period from December 2023 to December 2024, AI Overviews were enabled by default in the U. S. but not in Germany<\/a> or France<\/a>. Since about 40% of English Wikipedia traffic originates from the United States<\/a>, while the European editions draw primarily from regions without the feature, any divergence in referral trends could be attributed to the AI tool.<\/p>\n\n

Using a Poisson pseudo-maximum likelihood difference-in-differences model, the authors analyzed nearly 500,000 matched English-German article pairs and over 530,000 English-French pairs. Their calculations indicate that default AI Overview availability reduced external-search referrals to English Wikipedia by 5.45%<\/strong> relative to German and 4.82%<\/strong> relative to French. For the English edition as a whole, this translates to roughly 100.27 million fewer search referrals per month<\/strong>, or about 1.2 billion annually<\/strong>. These figures represent direct referrals from search engines only and do not capture all forms of traffic, such as direct navigation or app usage.<\/p>\n\n

Earlier iterations of the paper relied on daily pageviews and reported a 15% drop. The revised version incorporates monthly data and broader comparative controls to improve accuracy. A secondary check using an English-Japanese comparison showed a 16.53% decline, though the authors caution that this serves only as directional support due to differences in time frames and comparison groups.<\/p>\n\n

Limitations and Broader Context<\/h2>\n\n

The estimate addresses a specific causal question: how referrals changed when English-speaking users encountered AI Overviews by default. It does not claim to measure total engagement or long-term effects on content creation. The data counts human-initiated search referrals, meaning that if a user clicks through to Wikipedia and reads multiple articles, it is recorded as a single visit. Furthermore, because most English Wikipedia visitors come from outside the U. S., the measured effect may underrepresent the global impact of the feature.<\/p>\n\n

Financial implications are also speculative. The authors calculated that an ad-supported website with similar traffic might lose between $10.82 million and $37.08 million<\/strong> annually. However, since Wikipedia operates on donations and displays no ads, this number is hypothetical and does not reflect actual revenue loss for the Wikimedia Foundation or funds transferred to Google. The study also does not examine potential downstream effects on editing activity, volunteer retention, or donation levels.<\/p>\n\n

Wikimedia\u2019s Perspective on Traffic Changes<\/h2>\n\n

The Wikimedia Foundation observes a wider trend of declining engagement. In October 2025, product director Marshall Miller<\/strong> reported that overall human pageviews across all languages had decreased by about 8%<\/strong> compared to the previous year. This assessment followed improved bot detection methods that reclassified significant traffic from March to August 2025 as automated rather than human. Much of the previously counted “human” traffic originated from Brazil and was driven by bots designed to evade detection.<\/p>\n\n

Miller attributed the decline partly to generative AI and social media platforms but advised caution in interpreting the data due to evolving detection technologies. This broad decline in human pageviews is distinct from the causal estimate regarding AI Overviews and cannot be simply added to the 5% referral drop identified in the University of Washington paper.<\/p>\n\n

Looking ahead, the Foundation\u2019s draft plan for the 2026-27 fiscal year highlights decreasing pageviews and fewer Google referrals as persistent challenges rather than temporary fluctuations. The plan notes that nearly 90%<\/strong> of traditional Wikipedia visitors historically come from Google search<\/a>, underscoring the platform’s dependence on organic search traffic.<\/p>\n\n

The Rise of Automated Scraping<\/h2>\n\n

Beyond human traffic, Wikimedia faces increasing pressure from automated systems. Engineers observed in April 2025 that bandwidth for downloading images and media files rose by 50%<\/strong> since January 2024. Most of this increase stemmed from bots scraping Wikimedia Commons for AI training purposes rather than genuine user interest. These bots accounted for at least 65%<\/strong> of the most expensive traffic reaching core data centers and roughly 35%<\/strong> of all pageviews.<\/p>\n\n

By mid-2025, the technical team was blocking or throttling around 30%<\/strong> of these automated requests, amounting to approximately 1.5 billion blocked requests per day<\/strong>. Many of these bots mimic regular browsers or use proxies via residential internet connections to bypass restrictions. The Foundation aims to guide large-scale reusers toward official channels where they can identify themselves and receive higher rate limits, rather than attempting to stop reuse entirely.<\/p>\n\n

This automated activity does not appear in referral counts. Whether a crawler copies an article for training, a search engine quotes it in an AI response, or a reader clicks through, these are treated as separate events. Only the final click-throughs from search engines are counted as human external-search referrals in the economic analysis.<\/p>\n\n

Source Attribution and Enterprise Access<\/h2>\n\n

User behavior studies further illustrate the disconnect between AI summaries and source citations. An analysis by Pew Research Center found that people rarely clicked on sources within AI-generated answers. Examining 68,879 Google searches conducted by 900 U. S. adults, the study found that when AI summaries were present, users clicked on traditional organic results only 8%<\/strong> of the time, compared to 15%<\/strong> when no summary appeared. Clicks on specific sources within the AI summary occurred in just 1%<\/strong> of visits. Wikipedia, YouTube, and Reddit together constituted 15% of cited sources, though the exact share for Wikipedia was not specified.<\/p>\n\n

Google maintains that AI Overviews drive users to a diverse set of websites. Liz Reid, Vice President of Search Experience at Google, stated in an August 2025 blog post that total organic click volume remained “relatively stable”<\/strong> compared to the prior year. She reiterated this point on Bloomberg\u2019s Odd Lots podcast in April, suggesting that while the composition of traffic may change, the overall volume of clicks to websites has not collapsed.<\/p>\n\n

In response to these shifts, the Wikimedia Foundation urged AI developers in November 2025 to properly credit Wikipedia and access its data through Wikimedia Enterprise<\/strong>. This paid service provides high-volume API access, including contracts, uptime guarantees, and structured data feeds. Lane Becker, who oversees the program, explained in July that the model is based on access fees rather than licensing content, which remains freely available under open licenses. Revenue from Enterprise is capped at 30%<\/strong> of the Foundation\u2019s annual income and generated $8.3 million<\/strong> in the 2024-25 fiscal year.<\/p>\n\n

In January, Wikimedia Enterprise announced that Google is already a customer. Other partners include Amazon, Meta, Microsoft<\/a>, Mistral AI, Perplexity<\/a>, and GNOMI. The Foundation views Enterprise as a mechanism for sustainable funding and proper attribution, rather than compensation for lost referral traffic.<\/p>\n\n

Unique Data and Future Outlook<\/h2>\n\n

Wikipedia\u2019s unique transparency makes it an ideal test case for measuring AI\u2019s impact on web traffic. The availability of detailed, article-level data across multiple languages allows for rigorous comparative analysis that few other platforms can offer. However, this specificity also limits the generalizability of the findings. The study does not indicate whether news sites, e-commerce retailers, or other content publishers would experience similar declines in search-driven traffic. Additionally, the research does not explore whether other organizations could develop comparable paid access models to sustain their operations.<\/p>\n\n

As a working paper, the University of Washington study continues to evolve. The authors have rebuilt and revised the analysis multiple times since February, refining their estimates based on new data and methodological improvements. These figures represent the latest available insights but are not final. Monitoring updates from the authors, Wikimedia, and Google will be essential as the landscape of AI-driven search continues to change.<\/p>\n\n(Source: Search Engine Journal<\/a>)<\/em>","protected":false},"excerpt":{"rendered":"

A University of Washington study estimates that Google\u2019s AI Overviews reduced monthly search referrals to English Wikipedia by approximately 5% after the feature became default in the U.S. during 2024. Using a comparative analysis with German and French editions, researchers calculated this shift…<\/p>\n","protected":false},"author":1,"featured_media":255134,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_themeisle_gutenberg_block_has_review":false,"cybocfi_hide_featured_image":"","footnotes":""},"categories":[3247,6579,3327,513,3254],"tags":[11441,250801,239829,135,255411],"entities":[2461,7321,1354,2367,817,255413,255412,1013,75706,16557],"class_list":["post-255135","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence","category-bigtech-companies","category-newswire","category-quick-reads","category-technology","tag-ai-overviews","tag-france","tag-germany","tag-google","tag-wikimedia","entity-ai-overviews","entity-english-wikipedia","entity-france","entity-germany","entity-google","entity-hema-yoganarasimhan","entity-mehrzad-khosravi","entity-united-states","entity-university-of-washington","entity-wikimedia"],"_links":{"self":[{"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/posts\/255135","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/comments?post=255135"}],"version-history":[{"count":0,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/posts\/255135\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/media\/255134"}],"wp:attachment":[{"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/media?parent=255135"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/categories?post=255135"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/tags?post=255135"},{"taxonomy":"entity","embeddable":true,"href":"https:\/\/digitrendz.blog\/z\/wp-json\/wp\/v2\/entities?post=255135"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}