Evidence points to a real but limited referral hit, not a universal traffic collapse. An MSI listing for the working paper Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia frames the question around whether AI Overviews shift attention away from publisher pages.
Table of Contents
That matters because the best-known number comes from English Wikipedia under a defined rollout period. The key task is separating a measurable Wikipedia effect from broader claims about what any AI overviews web traffic study can prove.
Defining “AI overviews” and study context
In this debate, “AI overviews” means Google’s answer boxes that generate direct responses inside search results. They are not the same as featured snippets. Featured snippets pull a short passage from a page.
AI overviews, by contrast, combine information retrieval with generative large language models. That distinction matters before any traffic claim is judged. If searchers get a fuller answer on the results page, fewer may need to click through.
An arXiv paper titled Auditing Google’s AI Overviews and Featured Snippets: A Case Study on Baby Care and Pregnancy uses that framework in a narrow audit of pregnancy and baby care queries. Its examples also show why context matters.
In one case, the paper describes an overview about helping baby acne that mostly defined the condition instead of answering the “how to help” part. That makes AI overviews more than a design change. They reshape how answers are framed, how much intent is satisfied on the page, and when a visit still feels necessary, which is the key lens for reading any AI overviews web traffic study.
Research findings: referral decline quantified
More concrete evidence comes from the size of the drop, not just the theory behind it. Relevant Audience summarizes a University of Washington working paper by Mehrzad Khosravi and Hema Yoganarasimhan that estimates Google’s default AI Overviews reduced monthly external-search referrals to English Wikipedia by about 5% after the US rollout in May 2024.
The reported estimates were 5.45% against German Wikipedia and 4.82% against French Wikipedia. That does not sound catastrophic on its own. Still, it is large enough to matter at Wikipedia scale. The same summary says the model implies about 100.27 million fewer English referrals per month.
That turns a single-digit percentage into a meaningful audience loss. It also helps explain why this AI overviews web traffic study drew attention beyond search marketing circles. At the same time, the number is narrower than early headlines suggested.
Earlier paper versions reported about a 15% decline, but later revisions switched to monthly search referrals and added German and French controls, producing the lower estimate. For readers, the key takeaway is not collapse.
It is measurable erosion.
Methods, scope, and limitations
Scope matters as much as the headline result. This AI overviews web traffic study is not a census of the whole web. It tracks Wikipedia article traffic in a defined panel, not every publisher or query type.
On arxiv.org, Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia says the sample takes the union of articles that reached the top 1,000 at least once during a pre-treatment window running from July 1, 2023, to March 21, 2024.
That design helps build a larger sample. It also narrows what the estimate can speak to. The timing choice matters too. The paper marks March 22, 2024, as the start of U. S. testing on a subset of queries and traffic.
It then checks a stricter version that starts on May 15, 2024, the full U. S. launch date. The reported effect stays statistically significant and similar in size. That strengthens confidence in the pattern.
Still, it remains evidence about English Wikipedia under one rollout period, not proof that every site or later market will see the same referral loss.
Possible causes and mechanisms
Search behavior may shift for a simple reason: the answer now appears before the visit. On the University of Washington Center for Statistics and the Social Sciences seminar page for Impact of AI Search Summaries on Website Traffic: Evidence from Google AI Overviews and Wikipedia, the abstract says search engines increasingly place AI-generated answers above organic links, which can displace traffic to upstream publishers.
That points to an attention mechanism first, not a penalty mechanism. When the results page satisfies a basic informational need, some clicks never happen. The loss may be strongest when the query is straightforward and the summary feels complete enough.
It may be weaker when the user still needs depth, verification, or a specific page feature that a summary cannot deliver. That distinction matters when reading any AI overviews web traffic study. The likely effect is not that every searcher stops clicking.
It is that more of the answer journey ends on the search page itself. For publishers, that makes referral pressure easier to explain and harder to dismiss as random noise.
Implications for content creators’ strategy
Ranking alone stops looking like a complete strategy under AI Overviews. Even a page in the number 1 position may be hard to see when other results-page features take up more visual space, as Wikipedia notes in its overview of the feature.
That shifts the practical goal. Content has to compete for visibility and for trust, not only for placement. It also raises the value of material that is easy to verify. Wikipedia says some overviews may include hallucinated content for certain searches, which means users still have a reason to seek a source that shows clear facts, context, and ownership.
There is a tradeoff, though. Wikipedia also notes that Google has adjusted link placement inside AI Overviews, so referral opportunities may not disappear in every case. Still, smaller publishers can face weaker visibility when aggregation happens above them.
For creators reading this AI overviews web traffic study, the useful response is not panic or rank chasing. It is building pages that remain worth the click when summaries appear first.
Taken together, the claim is credible, but only in a narrow sense. Google AI Overviews appear to reduce referrals, with English Wikipedia down about 5% after the U. S. rollout. The strongest estimate comes from a defined Wikipedia panel, not the whole web.
It also covers one rollout period, so it cannot predict every site or query. The practical message is clear. Search visibility alone is less dependable when answers are satisfied on the results page. Pages that add depth, verification, and clear ownership still have a stronger reason to earn the click.
