On September 2, 2026, a University of Washington working paper dropped a number that should make every website owner uncomfortable. Google's AI Overviews — those AI-generated answer boxes that now sit at the top of search results — reduced monthly search referrals to English Wikipedia by roughly 5 percent. That translates to about 100 million fewer clicks every month, or 1.2 billion per year. From Wikipedia. The fifth most visited website on the planet.

If Google's AI can siphon a billion annual clicks from the site that literally defines the internet's knowledge base, what is it doing to yours?

The study in plain English

The paper, by Mehrzad Khosravi and Hema Yoganarasimhan, used a clever natural experiment. Google switched AI Overviews on by default in the United States in May 2024, but it had not done so in Germany or France during the study period. Wikipedia publishes monthly clickstream data by language edition, so the researchers compared what happened to English Wikipedia referrals (mostly from the U.S.) against German and French referrals (mostly from countries without the feature) for the same articles.

The difference-in-differences estimate: a 5.45 percent decline relative to German Wikipedia and 4.82 percent relative to French. A shorter English-Japanese comparison showed a 16.53 percent drop, though the authors flag that one as directional rather than definitive.

Google disputes the methodology, arguing that Wikipedia's clickstream data lumps all search engines together, making it impossible to isolate Google's specific contribution. The researchers counter that Google dominates search traffic so heavily that the timing break from the rollout is a clean enough signal. Both sides have a point, but the direction is not in dispute. The traffic went down.

Why this matters more than you think

Wikipedia is the canary in the coal mine, but it is also the best-case scenario for a content site. Wikipedia does not run ads. It does not sell anything. It has no conversion funnel to lose. The only thing at stake is whether people read the article on wikipedia.org or read a reconstituted version of it inside a Google AI box.

For a small business website, the stakes are dramatically higher. Your blog posts, service pages, and product descriptions exist to move visitors through a funnel — from awareness to consideration to purchase. An AI Overview can answer the informational query right there in the search results, and the user never needs to click through. You still provided the knowledge. Google still indexed it. But the visitor, the cookie, the session, the chance to show them your call to action — all of that stayed on Google's side of the fence.

This is not a future problem. It has been happening since mid-2024, and the effect is accelerating as Google expands AI Overviews to more query types and more countries.

What the data tells us about who gets hit hardest

The study focused on Wikipedia, but other research paints a broader picture. AI Overviews tend to appear most often on informational queries — the "how to," "what is," and "best way to" searches that content marketing depends on. These are the exact queries that small businesses target with blog posts and guides.

Transactional queries, where someone types "buy running shoes size 10" or "emergency plumber near me," are less likely to trigger an AI Overview. Google still wants to send those users to a page where they can take action. But the top-of-funnel educational content that builds trust and authority? That is squarely in the crosshairs.

The pattern is clear. If your site's traffic strategy relies heavily on ranking for informational keywords, you are already losing clicks to AI Overviews, whether your analytics dashboard makes it obvious or not.

What to actually do about it

Panicking about Google is a WordPress site owner tradition as old as the platform itself, and it is about as productive as it has always been. The better move is to adapt your strategy to the new reality.

Build for clicks, not just answers. If your blog post can be fully summarized in three sentences by an AI, there is no reason for anyone to visit. Write content that requires the reader to be on your site — interactive tools, calculators, comparison tables, embedded demos, personalized recommendations. Give people a reason to click through that an AI box cannot replicate.

Own your distribution. Email lists, push notifications, direct communities, and social followings are channels you control. Google can rewrite how search works, but it cannot touch your subscriber list. The businesses that will weather this shift are the ones that do not depend on a single traffic source.

Optimize for branded and transactional search. People searching for your business name or your specific product are not going to be satisfied by an AI Overview. Double down on the queries where you have a built-in advantage.

Make your content the destination, not the summary. The posts that survive AI Overviews are the ones that offer something an AI cannot package into a box — original data, unique perspective, community discussion, or a tool that does something useful.

The uncomfortable truth

Google is not going to reverse this. AI Overviews are the product now. The company's ad revenue depends on keeping users on Google properties, and AI-generated answers do exactly that. The deal that content creators had with Google for two decades — we make the content, you send us traffic — is being renegotiated unilaterally, and the new terms are worse.

That does not mean SEO is dead or that content marketing is pointless. It means the bar is higher. You cannot just answer a question anymore. You have to be worth the click.

The University of Washington paper is a working paper and has not been peer reviewed, which is worth noting. But the directional signal lines up with what every site owner with access to Google Search Console has been seeing for months. The clicks are moving. The question is whether your site is built to survive the shift.

Sources: University of Washington working paper (Khosravi & Yoganarasimhan, Sept 2026), Search Engine Journal coverage, Wikimedia traffic trends