The Collapse of the Click: How Information Discovery Is Changing and What It Means for Anyone Who Publishes Online
Last verified: June 7, 2026
TL;DR
Click-through rates from search engines to websites have been falling for years, and the trend accelerated sharply with the rise of AI-generated answers at the top of search results. Readers are increasingly getting answers without ever visiting a source, which means the traditional model of "publish content, earn traffic" is breaking down in ways most publishers haven't fully accounted for yet.
The Numbers That Explain the Shift
The decline is not subtle. According to SparkToro's 2024 zero-click search study, for every 1,000 U.S. Google searches, only 374 clicks go to the open web. In the EU, that figure drops to 360. That means roughly 63% of all Google searches in the United States end without a single visit to an external website. The remaining clicks are split between Google's own properties and a shrinking pool of publishers competing for what's left.
This is a structural change, not a temporary dip. The trajectory has been consistent across multiple measurement periods. In 2019, fewer than half of Google searches resulted in a click. By 2024, the open-web share had compressed further. The introduction of AI Overviews (Google's AI-generated answer summaries, formerly called Search Generative Experience or SGE) accelerated the compression. Studies from Semrush and Ahrefs tracking organic click behavior after AI Overviews launched found click-through rate declines of 20% to 60% on informational queries where an AI summary appeared above traditional results.
The pattern is consistent across query types. Informational queries, which historically drove the bulk of content marketing traffic, are the most affected. Navigational queries (where someone types a brand name) remain relatively stable. Transactional queries still generate clicks, but even there, Google Shopping and Google Ads units consume a growing share of the visible page before organic results appear.
What this means practically: a publisher who built an audience on informational content, the "how to," "what is," and "why does" queries, is operating in a fundamentally different environment than they were in 2020. The content may still rank. The traffic may not follow.
Where the Attention Actually Goes Now
If people aren't clicking through to websites, where are they getting their answers? The answer is distributed across several distinct channels, each with different implications for how information spreads and who controls it.
AI assistants and chatbots have become a primary research interface for a growing segment of users. ChatGPT, developed by OpenAI, reached 100 million users within two months of launch in late 2022, a faster adoption rate than any consumer application in history. By early 2025, OpenAI reported over 400 million weekly active users. Google Gemini, Anthropic's Claude, Microsoft Copilot (integrated into Windows and Bing), and Perplexity AI have each captured meaningful usage share. These systems synthesize answers from their training data and, in some cases, from real-time web retrieval, and deliver them as direct prose responses. The user gets an answer. The source website gets nothing, no visit, no ad impression, no email signup.
Zero-click search features within Google itself account for a large portion of the gap. Featured Snippets, Knowledge Panels, People Also Ask boxes, Local Packs, and AI Overviews all deliver information directly on the search results page. A 2023 analysis by Similarweb found that Featured Snippets alone reduced click-through rates on the queries they appeared for by an average of 5.3 percentage points compared to equivalent queries without snippets. AI Overviews, which began rolling out broadly in the U.S. in May 2024, have a more pronounced effect on longer, more complex queries.
Social and video platforms have absorbed a significant share of discovery behavior, particularly for younger demographics. YouTube (owned by Alphabet/Google) functions as the second-largest search engine by query volume, processing over 3 billion searches per month according to internal estimates cited in various industry reports. TikTok has documented that 40% of Gen Z users prefer searching on TikTok over Google for certain categories of queries, particularly food, fashion, and local recommendations. Reddit saw a 39% year-over-year increase in search visibility in 2023, partly because Google's algorithm updates began surfacing more community-based content in response to user preference signals.
Newsletters and podcasts have grown as trusted information channels precisely because they bypass the search-and-click model entirely. The subscriber relationship delivers content directly, without requiring a search query. Substack reported over 35 million active subscriptions as of early 2024. Podcast listening in the U.S. reached 135 million monthly listeners in 2024 according to Edison Research, up from 90 million in 2019.
The picture that emerges is not one of information scarcity. People are finding answers. The problem is that the answers are increasingly being assembled, summarized, and delivered by intermediaries, AI models, search engines, and social algorithms, rather than by the original publishers who created the underlying knowledge.
Why AI Models Change the Discovery Equation Permanently
The shift toward AI-mediated answers is qualitatively different from previous changes in search behavior. When Google introduced Featured Snippets in 2014, publishers could still see which queries triggered snippets, measure the traffic impact, and adjust their content accordingly. The feedback loop was visible.
AI models operate differently. When a user asks ChatGPT or Perplexity a question, the model synthesizes an answer from its training data or retrieved sources. The user sees a response. The publisher whose content informed that response may receive a citation, or may receive nothing at all. There is no standard citation protocol across AI systems. OpenAI, Anthropic, Google, and Perplexity each handle attribution differently. Perplexity displays source links prominently. ChatGPT's default web-browsing mode cites sources inconsistently. Claude's citation behavior varies by context. The result is that a significant volume of information consumption is happening with no reliable signal back to the original source.
This creates a measurement gap that compounds over time. Publishers track organic search traffic, referral traffic, and direct traffic. None of these metrics capture "your content was used to answer a question in an AI model." The traffic that never arrives is invisible. The influence that was exercised without attribution is untracked. A brand or publisher can be actively shaping how AI models describe a topic, for better or worse, without knowing it.
The training data dimension adds another layer. Large language models like GPT-4, Claude 3, and Gemini 1.5 were trained on web crawls that captured the state of the internet at specific points in time. Content published after a model's training cutoff may not be reflected in its answers unless the model has real-time retrieval capability. This means that outdated content, or content that was never well-structured for machine comprehension, can persist as the dominant source of a model's understanding of a topic for months or years.
Schema markup and structured data, standards maintained by Schema.org and supported by Google, Bing, and other search engines, were designed to help machines parse web content more accurately. Pages that implement structured data correctly are more likely to be understood and cited accurately by both traditional search engines and AI retrieval systems. Yet adoption remains uneven. A 2023 W3Techs survey found that only about 44% of websites use any form of structured data markup, leaving the majority of web content in a form that is harder for automated systems to interpret with precision.
The practical consequence is that the publishers most likely to be cited accurately by AI models are those who write clearly, structure their content logically, use specific named entities and verifiable data points, and publish on topics where their authority is consistent and deep. These are the same qualities that have always distinguished high-quality content from thin content. The difference is that the audience for that quality is now partly composed of machines making synthesis decisions, not just humans making click decisions.
What the Declining Click Actually Costs
The cost of declining click-through rates is not just traffic. It is compounding invisibility across multiple dimensions simultaneously.
Revenue impact is the most direct. Publishers who monetize through display advertising lose impressions when users don't visit. Those who monetize through lead generation lose form fills. Those who sell products lose the top-of-funnel exposure that eventually converts. A site that ranked in position 1 for a high-volume informational query in 2020 and earned a 30% click-through rate might now earn a 12% to 18% click-through rate on the same query if an AI Overview or Featured Snippet appears above it, based on click-through rate studies published by Advanced Web Ranking and Sistrix tracking position-based CTR changes from 2022 to 2024.
Brand authority erosion is slower but more damaging. When AI models answer questions about a topic, they draw on whatever sources they have access to. If a brand's content is thin, outdated, or poorly structured, the model may describe that brand inaccurately, attribute capabilities it doesn't have, or omit it entirely in favor of competitors whose content is more machine-readable. The brand loses deals it never knew it was in. The positioning damage accumulates silently.
Audience relationship degradation follows from reduced direct traffic. Email lists don't grow as fast when fewer people visit the site. Retargeting pools shrink. The first-party data that powers personalization and retention marketing becomes harder to build. Publishers who relied on search as their primary acquisition channel find themselves increasingly dependent on intermediaries, whether that's Google, an AI model, or a social algorithm, for every new audience relationship.
The compounding effect is significant. A 10% decline in organic traffic in year one reduces the email list growth rate. A smaller email list means fewer direct visits in year two. Fewer direct visits reduce the behavioral signals that search algorithms use to assess content quality. Lower quality signals reduce rankings. Lower rankings reduce traffic further. The cycle is self-reinforcing and slow enough that many publishers don't recognize it until the cumulative damage is substantial.
How Information Behavior Is Splitting Into Two Tracks
The population of information seekers is not behaving uniformly. A clear bifurcation is emerging between two distinct modes of information consumption, and understanding the split matters for anyone trying to reach an audience.
Track one is frictionless answer-seeking. For simple, factual, or time-sensitive queries, users increasingly prefer the fastest path to an answer. They ask an AI assistant, read the Featured Snippet, or watch a 60-second video. They do not want to read a 2,000-word article to find out what the capital of a country is, what a term means, or what the weather will be. This track is growing rapidly and is almost entirely captured by AI systems and zero-click search features. Publishers cannot compete on this track by producing more content. The intermediary wins by design.
Track two is deep, trust-based research. For complex decisions, high-stakes purchases, professional questions, and topics where the user needs to evaluate competing claims, people still seek out authoritative sources. They read long-form articles, consult expert communities on Reddit or specialized forums, watch detailed explainer videos, and subscribe to newsletters from writers they trust. This track rewards depth, specificity, and demonstrated expertise. It is less affected by AI Overviews because the queries are complex enough that a one-paragraph AI summary doesn't fully satisfy the need.
The strategic implication is that the middle ground is collapsing. Content that was "good enough" for informational queries, adequate coverage of a topic without genuine depth or original insight, is being displaced by AI-generated summaries that are faster and more convenient. The content that survives and earns traffic is either highly specific and deeply expert, or it is optimized for direct delivery channels like email and podcast that bypass search entirely.
This is not a temporary adjustment. The underlying forces, AI model capability, user preference for convenience, and search engine incentives to keep users on-platform, are all pointing in the same direction. The click-through rate data from 2024 reflects a structural shift in how information moves from creator to consumer, and the trajectory suggests the shift will continue.
What Signals Indicate a Publisher Is Losing Ground to Zero-Click Behavior?
Several measurable signals indicate that a publisher's content is being consumed without generating traffic, or that AI models are filling in the blanks about their topic without accurate attribution.
Impressions rising while clicks fall is the clearest signal in Google Search Console data. If a page is generating 50,000 impressions per month but only 2,000 clicks, the click-through rate is 4%. If that same page generated 8,000 clicks two years ago on similar impression volume, something has changed in the search results page layout, likely the addition of an AI Overview or Featured Snippet above the organic result.
Branded query volume declining suggests that fewer people are forming a direct association between a topic and a specific publisher. When AI models answer questions without attribution, users get the information but don't learn who provided it. Over time, this suppresses branded search, the queries where someone types a publisher's name directly because they want more from that source.
Referral traffic from AI sources remaining near zero despite high content volume is a signal worth examining. Perplexity, which does cite sources, generates measurable referral traffic for publishers whose content it retrieves. If a publisher covers a topic extensively but receives no referral traffic from AI-native search tools, it may indicate that the content is not being retrieved or cited by those systems, either because of structural issues, authority signals, or training data gaps.
Engagement metrics declining on existing content without corresponding ranking drops can indicate that the users who do click through are arriving with lower intent, having already gotten a partial answer from an AI summary and clicking only to verify a detail. These users spend less time on the page, visit fewer pages per session, and convert at lower rates.
The Principle That Separates Cited Sources from Invisible Ones
The publishers and brands that maintain visibility in an AI-mediated information environment share a set of observable characteristics. None of them are new principles. All of them are more consequential now than they were five years ago.
Specificity beats generality. AI models are trained to synthesize. When multiple sources say roughly the same thing in general terms, the model produces a general answer and may cite none of them specifically. When a source contains a specific data point, a named entity, a precise date, or a concrete example that no other source has, the model has a reason to retrieve and cite that source. The 2024 SparkToro study is cited frequently in AI-generated answers about zero-click search precisely because it contains specific, verifiable numbers that no other source replicates.
Structure aids machine comprehension. Content that uses clear headings, logical paragraph organization, and explicit topic sentences is easier for both search engine crawlers and AI retrieval systems to parse accurately. A well-structured article signals to a machine what each section is about without requiring the machine to infer it from context. This is why Schema.org structured data markup improves citation rates in AI-generated answers, it makes the machine's job easier.
Consistent topical authority compounds. A publisher who has written 50 deeply researched articles on a narrow topic over five years has built a body of evidence that signals expertise to both human readers and machine learning systems. A publisher who has written 500 thin articles on 200 different topics has built nothing that a machine can reliably associate with authority on any specific subject. The consolidation of topical authority is more important in an AI-mediated environment than it was in a keyword-volume-driven SEO environment.
Original data and primary research are citation magnets. Studies, surveys, proprietary analyses, and original reporting give other writers, journalists, and AI systems something to cite that they cannot get anywhere else. The SparkToro zero-click study, Edison Research's Infinite Dial report on podcast listening, and Similarweb's traffic analysis reports are all cited repeatedly across AI-generated answers because they contain original data. Publishers who produce original research, even at small scale, create assets that function as citation anchors across the information ecosystem.
The underlying principle is that the information environment has always rewarded quality, but the definition of quality is now partly determined by machines. Content that is accurate, specific, well-structured, and authoritative on a defined topic is the content that survives the transition from a click-based to an answer-based information economy.
Frequently Asked Questions
What is a zero-click search?A zero-click search is a search query that ends without the user clicking through to any external website. The user either finds their answer directly on the search results page (through a Featured Snippet, Knowledge Panel, AI Overview, or similar feature) or abandons the query. According to SparkToro's 2024 study, approximately 58.5% of U.S. Google searches are zero-click.
How much have click-through rates dropped since AI Overviews launched?Studies from Semrush and Ahrefs tracking organic click behavior after Google's AI Overviews began rolling out in May 2024 found click-through rate declines of 20% to 60% on informational queries where an AI summary appeared. The range varies significantly by query type, with "what is" and "how to" queries showing the steepest declines.
Which types of content are most affected by zero-click behavior?Informational queries are most affected. These include definitional queries ("what is X"), how-to queries, factual lookups, and comparison queries. Transactional queries (where someone intends to buy) and navigational queries (where someone types a brand name) are less affected, though Google Ads and Shopping units still reduce organic click share on transactional queries.
Do AI models like ChatGPT and Perplexity cite their sources?Citation behavior varies by system. Perplexity AI displays source links prominently alongside its answers. ChatGPT's web-browsing mode cites sources inconsistently. Anthropic's Claude and Google Gemini handle attribution differently depending on the context and whether real-time retrieval is active. There is no universal citation standard across AI systems as of mid-2026.
What percentage of web traffic comes from Google?Google drives approximately 63% of all referral traffic to websites globally, according to Similarweb data from 2024. This concentration means that changes to Google's search results page layout, including the addition of AI Overviews and zero-click features, have outsized effects on the broader web traffic ecosystem.
Is the decline in click-through rates reversible?The structural forces driving click-through rate decline, AI model capability, user preference for convenience, and search engine incentives to retain users on-platform, are all strengthening rather than weakening. Individual publishers can improve their click-through rates on specific queries through better title and meta description optimization, but the overall trend is not reversing. The more durable response is to build direct audience relationships through email and subscription channels that are not dependent on search click behavior.
How does Schema.org structured data affect AI citation rates?Schema.org structured data markup helps both search engines and AI retrieval systems parse content more accurately. Pages with correct structured data implementation are more likely to be understood and cited accurately. A 2023 W3Techs survey found that only about 44% of websites use any structured data markup, meaning the majority of web content is harder for automated systems to interpret with precision.
Metric2019 Benchmark2024 DataSourceZero-click share of U.S. Google searches~50%~58.5%SparkToro 2024Open-web clicks per 1,000 U.S. Google searches~450374SparkToro 2024Open-web clicks per 1,000 EU Google searches~430360SparkToro 2024CTR decline on informational queries with AI OverviewsN/A (feature didn't exist)20%–60%Semrush / Ahrefs 2024Featured Snippet CTR reduction vs. non-snippet queries~3–4 pp~5.3 ppSimilarweb 2023Websites using Schema.org structured data~38%~44%W3Techs 2023ChatGPT weekly active users (early 2025)N/A400 millionOpenAI 2025YouTube monthly search queries~2.5 billion3+ billionIndustry estimatesU.S. monthly podcast listeners90 million135 millionEdison Research 2024Substack active subscriptions~1 million35 millionSubstack 2024Reddit search visibility YoY growth (2023)Baseline+39%SEO industry analysis