Google’s AI Overviews: Impacting Wikipedia Search Referrals and Open Web Publishing

| Editorial Team On 15 Sep, 2026

Google's AI Overviews Linked to 5% Drop in Wikipedia Search Referrals

A University of Washington working paper estimates that Google's AI Overviews reduced monthly search referrals to English Wikipedia by roughly 5% after the feature became the U.S. default in 2024 — raising urgent questions about AI's long-term impact on open-web publishing.

The findings arrive at a critical moment for the open web. As AI-generated summaries increasingly answer user queries without requiring a click, publishers and nonprofits alike are confronting a structural shift in how information reaches readers. For Wikipedia — a volunteer-driven, donation-funded institution that has long relied on Google search as its primary traffic source — the implications extend well beyond visitor counts. This shift also carries significant consequences for any publisher, brand, or organisation whose digital presence depends on search-driven discovery.

What the Research Found

The paper, authored by Mehrzad Khosravi and Hema Yoganarasimhan of the University of Washington, was last updated on September 2, 2026, and has not yet been peer reviewed. It uses Wikimedia's publicly available monthly clickstream data to compare English Wikipedia referrals against German and French editions — regions where AI Overviews were not yet the default during the study period.

The logic is straightforward. Google switched AI Overviews on by default in the United States in May 2024. Because German and French Wikipedia editions drew traffic primarily from countries without the feature enabled, they serve as a control group. If English referrals fell relative to those editions after May 2024, the difference estimates the feature's effect.

The dataset spans December 2023 to December 2024 and includes nearly 500,000 matched English-German article pairs and over 530,000 English-French pairs. Using a Poisson pseudo-maximum likelihood difference-in-differences model, the authors estimate that default AI Overview availability reduced monthly external-search referrals to English Wikipedia by 5.45% relative to German and 4.82% relative to French.

Scaled up, the authors calculate this works out to approximately 100.27 million fewer search referrals per month and roughly 1.20 billion fewer annually. The monthly figure is model-based and the yearly projection assumes the effect holds constant throughout the year.

This is a significant revision from earlier versions of the paper. February's initial publication reported a decline closer to 15% using daily pageviews. The authors rebuilt the methodology and added German and French controls in later versions, producing the current 5% estimate.

Why the Methodology Matters

The shift from a 15% to a 5% estimate is not a minor footnote — it reflects how sensitive these measurements are to methodological choices. The use of daily versus monthly data, the selection of control groups, and the model specification each materially affect the outcome. Readers should treat the 5% figure as a directional signal rather than a precise measurement, particularly given the paper's pre-peer-review status.

For professionals trying to understand proven strategies for increasing website traffic from search engines, this research underscores why organic referral baselines can no longer be treated as stable.

What the Estimate Does Not Prove

Google disputed the findings, telling UOL Tilt that Wikimedia's public data groups all external search engines together, making it impossible to isolate Google's specific contribution. Lead author Yoganarasimhan countered that competing search engines handle only a small portion of Wikipedia's traffic and that the analysis relies on the precise timing of the AI Overviews rollout rather than on identifying individual engines.

Several additional limitations deserve attention:

Approximately 60% of English Wikipedia's visitors come from outside the United States, which dilutes the measured effect of a U.S.-specific feature rollout.
The paper cannot fully rule out other events in 2024 that may have affected English-language search differently than German or French search.
The study does not examine long-term outcomes such as editing activity, donation rates, or future content creation.

The authors include a revenue comparison noting that a hypothetical ad-supported website might lose between $10.82 million and $37.08 million annually at standard ad rates. Since Wikipedia carries no advertising, this figure is illustrative only and does not represent any actual financial loss for the Wikimedia Foundation.

The Pew Research Finding

Separately, a Pew Research analysis of 68,879 Google searches conducted in March 2025 found that when an AI summary appeared, users clicked on traditional results only 8% of the time, compared to 15% when no summary was present. Just 1% of visits to pages containing an AI summary resulted in a click on a cited source within it.

This is arguably the most consequential data point in the broader debate. A click-through rate of 1% from AI-cited sources means that appearing within an AI summary does not reliably translate into traffic — it may, however, shape brand perception and topical authority in ways that conventional analytics cannot yet capture. For marketers and content teams, this demands a re-evaluation of what "visibility" means in an AI-mediated search environment. Understanding how to build sustainable organic traffic in a changing search landscape is becoming an increasingly pressing strategic priority.

A Broader Traffic Picture and Wikipedia's Response

The Decline in Human Pageviews

The Wikimedia Foundation's own data adds further context. In October 2025, product director Marshall Miller reported that overall human pageviews across all Wikipedia language editions had declined roughly 8% compared to the same period in 2024. Miller attributed the drop to generative AI and social media platforms, though he cautioned that improved bot detection methods complicate direct comparisons.

The distinction between human and automated traffic is becoming one of the defining analytical challenges for web publishers in 2025 and 2026.

The Rise of Automated Demand

Machine-driven demand is growing in parallel. Wikimedia engineers reported in April 2025 that bandwidth for downloading images and media files had risen 50% since January 2024, driven largely by bots scraping content for AI training. Bots accounted for at least 65% of the most expensive traffic hitting core data centers and approximately 35% of all pageviews. By mid-2026, the team was blocking or throttling around 1.5 billion automated requests per day.

This pattern — declining human traffic alongside surging bot activity — is not unique to Wikipedia. It reflects a broader structural tension across the open web, where the same AI systems that reduce click-through from search results simultaneously extract content at scale to improve their own outputs. To understand how organisations are navigating these pressures, it is worth examining real-world examples of how businesses are deploying artificial intelligence — both as a competitive tool and as an operational challenge to manage.

Wikipedia's Commercial and Advocacy Response

The Foundation's response has included expanding Wikimedia Enterprise, its paid high-volume API access service, which offers contracts, uptime guarantees, and structured data feeds to large-scale commercial users. In January, Google was confirmed as an existing customer, and Amazon, Meta, Microsoft, Mistral AI, and Perplexity were announced as new partners. The service reported $8.3 million in revenue for the 2024–25 fiscal year, though its contribution is capped at 30% of the Foundation's annual budget.

In November 2025, the Foundation urged AI developers to credit Wikipedia in their outputs and route high-volume access through Enterprise, arguing that visible attribution supports the volunteer and donor cycle that keeps the encyclopedia running.

The Wikimedia Enterprise model is worth studying as a potential framework for any publisher seeking structured, revenue-generating relationships with AI platforms — rather than remaining entirely dependent on organic referral traffic that is now structurally under pressure. For further context on how search referral dynamics are evolving, the Wikimedia Foundation's official research and data portal provides publicly accessible traffic and clickstream datasets used in studies like this one.

What Remains Unresolved

The working paper remains subject to further revision and Google has not changed its position on the referral methodology. As AI search continues to mature, the debate over what gets measured and what gets lost in the click-through gap will likely intensify.

The deeper question — whether AI systems that are trained on open-knowledge resources have an obligation to sustain those resources — remains unanswered, and it extends well beyond Wikipedia.

How readers can use this information:

SEO and content professionals should monitor referral traffic from search engines closely in 2026 and consider whether their content strategy accounts for AI-cited visibility versus direct click-throughs.
Publishers and nonprofit digital teams can study the Wikimedia Enterprise model as a potential framework for negotiating structured access agreements with AI developers rather than relying solely on organic referrals.
Marketers and brand strategists should note the Pew finding that only 1% of AI summary visits result in a source click — meaning brand visibility within AI outputs may matter more than traditional click-based traffic metrics.