AI and the End of the Webpage

For most of the modern internet, webpages were destinations. A user searched, clicked, arrived, and made sense of information inside the context of a source that carried its own voice, institutional identity, archive, and logic. A webpage was not just where information was stored. It was where information gained framing, evidence, authorship, and meaning. Even after Google became the front door to the web, the structure remained familiar: search pointed, and the user left to enter someone else’s domain.

That structure is changing. AI systems now gather material from multiple pages, sort them, compress them, and return a synthesized response as the primary experience. The webpage still exists, but it no longer occupies the same place in the user journey. It is increasingly treated as input rather than endpoint. The web is moving from a network people visit to a resource machines use, and the economic effects are already visible. Search traffic is expected to fall sharply over the next three years, while Google Search traffic to more than 2,500 sites already fell 33 percent worldwide and 38 percent in the United States between November 2024 and November 2025.

The Page Stops Being the Destination

In the old version of the web, search engines largely cataloged webpages and information. The user was given a list, then had to do the research themselves. Search could rank, sort, and surface, but the real labor still belonged to the person at the keyboard: clicking through, comparing sources, judging credibility, and assembling meaning from scattered destinations across the open web. The ritual was simple: search, click, compare. That process was slower and often inefficient, but it preserved a defining feature of the open internet. The user still had to encounter the source in its own environment.

Now the ritual is different: ask, receive, continue. AI silos information, sorts it, and returns it in a structured research format that already feels closer to a finished answer than a search result. The user is no longer being sent outward in the same way. In many respects, this is what people wanted search to become for years: not a ranked directory of places to visit, but an engine that could gather what mattered, process it, and deliver a usable conclusion. The convenience is real, which is exactly why the underlying shift is so easy to underestimate.

That creates a fundamental shift for the internet. The old web was organized around destinations. The emerging AI web is organized around extraction, synthesis, and controlled presentation. Once an AI system can collect data from many places, store it internally, and use that material to answer future prompts, it no longer needs to return to the source in the same way with every request. The logic of the system changes. The webpage is no longer the primary place where meaning is made for the user. It becomes part of a larger informational reserve the machine can draw from, reorganize, and redistribute on its own terms.

The practical result is not just technical efficiency. It is a transfer of value. The system captures the value of the initial request, holds relevant material inside its own answer environment, and reduces the need to send users back to the original source. That dynamic is already contributing to fewer webpage visits, lower traffic, and growing pressure on ad-based monetization as publishers absorb sharp declines in search referrals. It is more efficient, more streamlined, and often more convenient for the user. But it also changes how research happens, how authority is encountered, and how specialized webpages function in the broader information economy. Webpages still matter, but they now behave less like destinations and more like machine-readable reserves that can be reused inside someone else’s interface.

Why the Click Matters Less

What disappears first is not the webpage itself, but the necessity of visiting it. In the older search model, the click linked discovery to publishing. A user who clicked through still encountered the publisher’s framing, authority signals, subscription prompt, ad model, and editorial environment. That encounter mattered because it gave the source a chance to do more than contribute a fact. It could establish credibility, supply context, present evidence in full, and deepen the reader’s understanding. In the AI model, that transfer increasingly fails to occur because the answer can arrive before the visit becomes necessary. The source may still be indexed, visible, and technically available, but it no longer sits at the center of resolution.

Click-through rates for top-ranking pages fell 34.5 percent when AI Overviews appeared. That is more than a traffic statistic. It marks the point at which webpages can remain present in the system while losing their practical role in the user journey. The answer layer starts taking the economic and cognitive value that once belonged to the page itself. The publisher loses the visit, the platform keeps the interaction, and the user loses some of the context that once came with arriving at the source.

The user-facing version of this shift is already becoming ordinary. Someone asks a question about a new regulation, a medical concern, a tax issue, or a software decision. The system returns a clean answer with citations, summaries, and follow-up prompts. The user gets what feels like completed research in seconds and moves on. No tabs open. No comparison between institutions. No real encounter with the source as a destination. What disappears with that speed is not only friction, but exposure to disagreement, institutional difference, uncertainty, and missing context.

The click was never just a monetization event. It was also an exposure event. It brought the user into contact with the source’s architecture of trust: the publication’s editorial standards, the agency’s branding, the specialist site’s method, the university’s evidence base, the regulator’s exact wording. Once those encounters happen less often, source visibility narrows in a deeper sense than traffic analytics alone can capture. The user still receives information, but increasingly without the full context that once distinguished one source from another.

Google’s recent AI Mode updates push that behavior further by letting users open sources side by side within the AI environment and continue asking questions without leaving the answer frame. On the surface, that looks like a usability improvement. In practice, it keeps discovery, inspection, and interpretation inside the same controlled environment. The visit remains nested inside the system that summarized the source first. As clicks decline, direct source exposure narrows, and once users stop encountering sources in their full context, siloed research environments become much easier to build.

From Open Navigation to Research Silos

This is where the economic shift becomes a knowledge shift. Different AI engines retrieve from different corpora, weight different source types, inherit different safety constraints, and reflect different product incentives. Two users asking similar questions in different AI systems may not simply receive different wording. They may encounter different source exposure, different confidence patterns, and different implicit editorial judgments. The result is not only personalization. It is divergence.

Across 11,000 real queries, AI systems showed meaningful source-selection bias, uneven source diversity, and reductions in hedging language of up to 60 percent. AI-mediated discovery does not merely summarize the web. It reorganizes the user’s exposure to the web. Over time, that creates conditions for research silos in which reality is filtered through model-specific patterns of citation, omission, and confidence.

The same pattern appears in user behavior. Citation presence and citation volume shape preference even when the cited material does not fully support the claim being made. In practice, citation structure itself becomes part of the persuasive interface. Authority is no longer only a property of institutions and sources. It is increasingly a property of presentation, fluency, and answer design. The old web was often messy, slow, and inefficient, but it preserved one important form of agency: users still had to move laterally across sources and observe disagreement directly. AI reduces that friction, but it also pre-processes the world before the user sees it.

For the user, this can feel like progress. The answer arrives faster, the interface feels calmer, and the system appears to remove wasted effort. But the cost of that calm is often invisible. The user sees less of the contest between sources, less of the institutional character behind claims, and less of the uncertainty that once signaled where judgment was still required. Research becomes smoother at the exact moment it may also become narrower.

Data Sovereignty, Governance, and the Next Internet Economy

The next fight is not over links. It is over control of the data layer beneath discovery, interpretation, and digital visibility. Search traffic is already expected to fall sharply over the next three years, and Google Search traffic to more than 2,500 sites has already fallen 33 percent worldwide and 38 percent in the United States over a one-year period. Click-through rates for top-ranking pages also fell 34.5 percent when AI Overviews appeared. Those numbers point to something larger than a temporary platform adjustment. They show value moving away from the visit itself and toward the system that captures the query, structures the answer, and retains the user.

That is where data sovereignty starts to matter. Once AI systems become the primary interface for research, the most valuable asset is no longer just content. It is the accumulated record of questions asked, sources selected, patterns reinforced, and answers delivered. Whoever controls query flow and retrieval logic controls not only attention, but the conditions under which information remains visible, monetizable, and institutionally legible. This is what brings the issue back to internet economics. In the earlier web, value was distributed across publishers, search intermediaries, advertisers, and users. In the emerging AI web, more of that value can concentrate inside the answer system itself.

The concentration is not only economic. It is epistemic and political. Across 11,000 real queries, AI systems showed meaningful source-selection bias and reductions in hedging language of up to 60 percent. Across more than 24,000 multi-turn interactions, citation presence itself shaped user preference even when the cited material did not fully support the claim being made. In multimodal generative search, between 3.7 percent and 18.7 percent of video-grounded claims were not supported by their cited sources. In health-related AI Overviews, YouTube accounted for 4.43 percent of citations across more than 50,000 German health queries, outranking hospitals, government health portals, and medical associations. Those figures make the governance issue harder to dismiss. The answer layer is not simply organizing information more efficiently. It is beginning to shape source hierarchy, confidence, and trust at scale.

Governance follows naturally from that concentration. If answer systems become the default gateway for commercial discovery, public information, and research behavior, then data access, source transparency, and competitive interoperability move closer to the center of digital policy. Europe is already moving in that direction, with regulators pushing to open ranking, query, click, and view data to third-party search competitors. That matters because it treats search data as a strategic input in the next internet economy, not just a legacy search-market asset. The internet is not ending because webpages are disappearing. It is changing because webpages are losing their role as the places where value, visibility, and meaning converge. They still matter as archives, evidence bases, and repositories of original work. But they are increasingly encountered through systems that interpret them first, retain the user, and decide how much of the source remains visible. The future of the web will be shaped less by who publishes the most information than by who governs the systems that collect it, structure it, and turn it into the answers people increasingly accept without leaving the machine.

Key Takeaways

• AI is shifting the web from a network of destinations to a reservoir of machine-usable information.
• The deeper disruption is not the disappearance of webpages, but their demotion from destination to source material.
• Search traffic is already under pressure as AI systems reduce the need for users to click through.
• The click matters not only economically, but because it historically exposed users to source context, institutional identity, and trust signals.
• Different AI systems can create different informational realities through source-selection bias and differing citation patterns.
• The emerging battle is not only over traffic, but over control of the data layer beneath discovery, interpretation, and online visibility.
• Data sovereignty and governance are becoming central because answer systems are turning into economic and epistemic infrastructure.
• The future of the web will depend less on who publishes the most information than on who controls the systems that structure and serve it.

Sources

• Reuters Institute; Journalism, Media, and Technology Trends and Predictions 2026; – Link
• Digital Content Next; The publisher’s playbook for the Google Zero era; – Link
• SSRN; The Disruption of Search Engine Optimization by Large Language Models: A Mixed-Methods Analysis of the Evolving Search Landscape; – Link
• arXiv; Answer Bubbles: Information Exposure in AI-Mediated Search; – Link
• arXiv; Search Arena: Analyzing Search-Augmented LLMs; – Link
• arXiv; Auditing the Reliability of Multimodal Generative Search; – Link
• Google; Expanding AI Overviews and introducing AI Mode; – Link
• Google; A new way to explore the web with AI Mode in Chrome; – Link
• European Commission; Commission proposes measures to Google on sharing search engine data with third parties under the Digital Markets Act; – Link
• Journal of Competition Law & Economics; How Future-Proof is the DMA? A Case Study of AI Agents; – Link
• Journal of European Competition Law & Practice; Opt-out remedies will not fix AI overviews; – Link
• SE Ranking; Health AI Overviews Trust YouTube Over Medical Platforms; – Link