Last verified: June 11, 2026
TL;DR
AI models are now answering questions that used to send people to websites, and the shift is happening faster than most publishers and marketers have noticed. When a user gets a complete answer from an AI interface, the click to the source often never happens. The brands and content creators whose information feeds those answers are losing visibility they cannot measure with traditional analytics.
The Click Is No Longer the Default Outcome
Search used to have a predictable shape. A person typed a question, scanned a list of links, and clicked through to read. The website visit was the point. That assumption is now structurally broken for a growing share of queries.
AI-powered interfaces, including ChatGPT, Google's AI Overviews, Perplexity, Microsoft Copilot, and Claude, now synthesize answers directly in the response window. The user reads the answer, closes the tab, and moves on. No click. No session. No conversion signal. For the publisher whose content trained or informed that answer, the transaction is invisible.
This is not a marginal edge case. Google's own AI Overviews now appear on a significant portion of informational queries in the United States. Perplexity reported hundreds of millions of queries per month as of early 2025. The behavioral shift is real: when an answer is complete and credible, most users do not click to verify it. Research on zero-click searches predates AI by years, but AI-generated answers accelerate the pattern dramatically because the answers are longer, more conversational, and more directly responsive than a featured snippet ever was.
The practical consequence is that organic traffic to informational content is declining for many publishers even as their content continues to be consumed, just indirectly, as training data or retrieval context. The metric that used to signal reach (the page view) no longer captures the full picture.
Why AI Models Answer Instead of Redirect
Understanding why this happens requires understanding how large language models and retrieval-augmented systems work, at least at a functional level.
Large language models (LLMs) like GPT-4o, Gemini 1.5, and Claude 3 are trained on vast corpora of text. They do not retrieve a webpage and display it. They generate a response based on patterns learned during training, sometimes supplemented by real-time retrieval. The result is a synthesized answer that may draw on dozens of sources without surfacing any of them prominently. Even when citations appear, studies of user behavior consistently show that most readers do not follow those citations.
Retrieval-augmented generation (RAG) systems, which power interfaces like Perplexity and the cited-answer features in Google AI Overviews, do pull from live web content. But the architecture is still answer-first. The model reads the source, extracts what it needs, and presents a summary. The source URL appears as a footnote, not as the destination.
The incentive structure of these products reinforces the pattern. AI interfaces compete on answer quality and speed. A response that sends the user away to read a webpage is, from the product's perspective, a failure. The goal is to keep the user in the interface, satisfied. That goal is largely achieved. The publisher's goal, to earn the visit, is not.
What the Traffic Data Is Actually Showing in 2025
The data picture is uneven, which makes it easy to dismiss. Some categories of content are seeing steep declines in organic traffic. Others are holding steady or growing. The difference largely tracks query type.
Informational queries, the kind that start with "what is," "how does," or "why does," are the most exposed. These are exactly the queries AI models answer best. Publishers who built traffic on definitional content, explainers, and how-to guides are reporting the sharpest declines. Navigational queries (searching for a specific brand or site) and transactional queries (searching to buy something) are less affected, because the user's intent requires a destination, not just an answer.
Similarweb and other web analytics providers have published data showing that referral traffic from Google to many content-heavy sites declined measurably through 2024 and into 2025, coinciding with the broader rollout of AI Overviews. The correlation is not proof of causation, but the timing is hard to ignore. Independent analyses of specific publisher categories, including health information sites, recipe sites, and general reference content, have documented traffic drops ranging from modest to severe depending on the site's reliance on informational queries.
The harder problem is what the traffic data does not show. A brand whose positioning is being described, accurately or inaccurately, by an AI model has no standard analytics signal for that event. The AI answer happens. The user forms an impression. The brand never sees it. Traditional web analytics tools like Google Analytics 4 measure sessions, not AI citations. The gap between what is being said about a brand in AI responses and what the brand can observe is, for most organizations, total.
The Asymmetry Between Visibility and Influence
The deepest problem here is not the traffic loss itself. Traffic loss is measurable, at least in aggregate. The deeper problem is the asymmetry between how much influence AI models now have over buyer perception and how little visibility most brands have into that influence.
When a potential customer asks an AI model to compare options in a category, the model answers. It names some brands, describes their strengths, and may omit others entirely. That answer shapes the buyer's mental shortlist before they visit a single website. The brand that gets described accurately and favorably has an advantage. The brand that gets described with outdated information, or not mentioned at all, has already lost ground in a conversation it never knew was happening.
This is the structural shift that traffic metrics miss. Share of voice in AI-generated answers is becoming a meaningful competitive variable, but almost no organization is measuring it systematically. The brands that recognize this early are building processes to audit what AI models say about them, identify where the descriptions are wrong or incomplete, and publish content structured to correct the record. The brands that do not recognize it are optimizing for a search environment that is already changing beneath them.
The click was never the goal. The goal was always the impression, the consideration, the decision. AI models are now shaping all three, often before a website is ever visited. That is the shift worth understanding.