AI writer: Marcus Reed Systems & Infrastructure Writer

Search used to be the front door to the web. Now it is starting to look like a room with the answers already on the table. Google’s AI features are pushing more queries toward summaries and conversational follow-ups, and the practical result is fewer clicks to the open web.[4][7][10][11] That is convenient for users. It is also a structural problem for the sites that used to rely on search as their main distribution channel.

Google has been expanding AI Overviews across countries and languages since 2025, and it has also described a more seamless AI Search experience that blends AI Overviews with AI Mode.[1][4][10][11] In its own framing, the goal is to make search more conversational and more useful for complex questions. The company also says people are using Search more than ever.[1][4] Usage and referral traffic are not the same thing. A search box can be busy while the open web gets less of the benefit.

A 2025 behavioral study tracking 900 U.S. adults across 68,879 Google searches found that when an AI summary appeared, users clicked a traditional result 8% of the time, versus 15% without one.[8] Links inside the summary were clicked 1% of the time.[8] Other reporting on zero-click search put the broader trend at 56% of searches ending without a click.[5] That is not a rounding error. It is a change in the default path from question to destination.

A Reuters Institute survey said news executives expect search referrals to fall by more than 40% over the next three years as answer engines take over more of the first interaction.[3][6][9] Separate industry analysis projects organic traffic declines of 25% to 50% for some sites and argues that AEO and GEO are becoming the new optimization vocabulary.[3][13] Those terms are less important than the underlying fact: the old bargain was visibility in exchange for clicks. That bargain is getting rewritten.

There is a technical reason this shift is hard to reverse. Classic search ranked pages and sent users out to read them.[4][7] AI search systems compress the page into an answer layer first, then decide whether a click is still useful.[4][7][10] That changes measurement, attribution, and incentives at the same time.[2][3][8] If the answer is good enough, the site becomes a source to be extracted, not a destination to be visited. That is efficient for information retrieval. It is much less friendly to ad-supported publishing, lead generation, and any business model that assumes the page visit is the product.

It also changes how web teams have to think about their own work. If your audience is increasingly reached through snippets, summaries, and AI-generated responses, then page design alone is not enough.[1][2][4][10] Structure matters. Machine-readable content matters. Clear entities, stable URLs, and explicit provenance matter.[1][2][11] This is why some developers are now treating the page less like a brochure and more like a data object that has to survive crawling, summarization, and reuse. That is not a moral judgment. It is just where the traffic is going.

But there are still important limits to the story. The available sources do not prove that every category of search is collapsing at the same rate.[2][5][8][12] Informational queries are usually the most exposed to summaries, while navigational and transactional searches may behave differently.[5][12][14] We also do not yet have enough clean, long-running public data to separate short-term product rollout effects from a permanent structural break.[2][5][12] That distinction matters. A big interface change can look like a new regime before the market has fully adjusted to it.

Google keeps emphasizing that people are finding more useful results and that AI features bring more questions into search.[1][4][10][11] That does not mean the old web referral model survives unchanged. It does mean the right question is not whether Google is disappearing. The better question is which part of the web’s value chain it is absorbing. Search can remain a giant business and still hollow out the traffic mechanics that publishers and developers depended on.

For site owners, the practical answer is not to chase every new acronym. It is to watch three things. First, whether AI answer layers keep expanding across more query types.[1][4][10][11] Second, whether referral traffic keeps weakening even when rankings hold.[2][3][5][8] Third, whether publishers, product teams, and SEO shops start treating attribution loss as a normal operating condition instead of a temporary glitch.[3][6][13] If those signals persist, then the web is not being killed. It is being reorganized around different incentives. That is slower, messier, and more important than a headline about the end of search.