Internet traffic has collapsed.

Every time you get an answer from an AI assistant, one website dies.

Three quarters, 70 million monthly active users.

This is the report card Baidu App delivered from mid-2025 to the first quarter of 2026 — except that the trend is downward.

In the first quarter of 2026, the number of monthly active users of Baidu App dropped to 655 million, nearly 70 million less than the same period last year.

Baidu is no ordinary app. In China's internet over the past 20 years, "Baidu it" has almost been synonymous with "search for something online". It is the starting point of traffic, the entrance of information, and the "tap" that countless websites rely on for survival.

For the exaggerated decline in traffic, the explanation from Baidu's management is very straightforward — users are increasingly inclined to get answers directly from AI assistants instead of clicking links through search engines.

The same trend is happening across the globe.

On the other side of the Pacific, the same scenario is unfolding in a more drastic way. A "great traffic collapse" that is sweeping the entire internet is taking place.

And people don't care.

01 No One Survives

In March 2026, analytics firm Chartbeat exclusively provided Axios with a set of data that detonated the entire publishing industry.

This set of data tracked traffic changes on thousands of Chartbeat's client websites worldwide over the past two years, and the conclusion is shocking — search referral traffic for small publishers (websites with 1,000 to 10,000 page views per day) plummeted by 60%. Medium-sized publishers saw a 47% drop. The decline for large publishers is relatively "moderate", but still reached 22%.

The smaller the scale, the worse the outcome. The reason is not complicated — large media outlets have brand recognition, dedicated apps, email subscription lists, and loyal users who visit directly. Small publishers almost bet all their assets on Google Search, and when search traffic drops off a cliff, they have no buffer at all.

Page views from Google Search dropped by 34% between December 2024 and December 2025. Google Discover also fell by 15% in the same period. The entire "traffic tap" of Google is being tightened.

But for individual websites, the figures are far more tragic than these averages.

In a report released by Loopex Digital in May 2026, the list of the most affected websites is listed. Business Insider's traffic plummeted by 48.4%. WebMD's monthly visitors dropped from 122 million to 69.5 million. Stack Overflow, the world's largest programming Q&A community, saw its traffic drop by 35.5%, with Google search volume down 32.4%. Quora fell by 28.1%. Even image material websites like Unsplash did not escape, with monthly users dropping from 26.8 million to 17.7 million, a decline of 34%.

If you look across this list, you will find a disturbing rule — the parts that are cut the hardest are precisely the core components of the "internet information infrastructure" in the past ten years.

News media (Business Insider, CNN), knowledge Q&A platforms (Quora, Stack Overflow), medical information websites (WebMD), image material platforms (Unsplash) — these websites have one thing in common: their fundamental value is to answer users' questions.

And now, someone else is answering for them.

02 The "Zero-Click" Era

To understand what is happening, you first need to clarify a concept — "Zero-click Search".

This term refers to the situation where a user enters a question on a search engine and gets the answer directly on the search results page, without clicking any link or visiting any website. Approximately 60% of searches worldwide now end with zero clicks. On mobile, this figure is as high as 77%.

Zero-click search did not emerge in the AI era. Google has long been doing the same thing through features such as "Featured Snippets", "Knowledge Panels" and "People Also Ask" — displaying answers directly on the search results page to reduce the reasons for users to click away from Google. But AI Overviews has pushed this to a completely different magnitude.

According to data from multiple studies, Google AI Overviews currently covers about 48% of search queries, a year-on-year increase of 58%. When an AI Overview appears on a search results page, the click-through rate of the top organic search result drops by 34% to 58% — depending on which study you refer to, but the direction is completely consistent.

Seer Interactive's analysis is more radical, arguing that for queries with AI Overviews, the overall click-through rate of organic search has plummeted by 61%.

To put it bluntly: your website is still in the search results, and the ranking may not have changed at all, but the click volume may have been cut in half. This is the most confusing part for many website operators — "I didn't do anything wrong, but the traffic just dropped off a cliff."

But Google AI Overviews is only half the story.

The other half is a whole emerging "AI search" ecosystem growing wildly.

ChatGPT now processes more than 1 billion search queries per week. Perplexity also has more than 1 billion monthly queries. Claude, Gemini, and Copilot are all growing rapidly — Goodie's 2026 AI Search Traffic Report shows that Claude's AI referral share jumped from 11.8% to 18.5% in early 2026, the fastest growth among all AI search engines.

There is an essential difference between the way these AI products operate and traditional search engines. Google's past business logic was to "send users to websites" — you search for a question, Google gives you a bunch of links, you click in, and Google's job is done. This is a "traffic diversion" model.

The logic of AI chat products is to "give you the answer directly". It reads your content, understands your knowledge, synthesizes a response, and then delivers it directly to the user. The user is satisfied and leaves, without ever needing to know where the answer came from, or visiting any website at all.

This is why just focusing on the traffic changes of Google is not enough. What is really happening is far more profound than "Google algorithm updates".

The entire internet is transforming from a "network of links" to a "network of answers".

Over the past 20 years, the basic chain of internet information flow has been: creators produce content → search engines index and rank → users click links → visit websites → content creators get traffic and advertising revenue. This is a closed loop. Every participant has their own role and interests.

Now, this chain is being cut off. The new chain becomes: creators' content is crawled and learned by AI → AI directly answers users' questions → users leave satisfied → content creators get nothing.

The middle "click" link, the core infrastructure that has supported the entire digital publishing industry, digital advertising industry and content marketing industry for 20 years, is being systematically eliminated.

03 China's "Parallel Universe"

Some people may think that this is mainly a story about Google and the English internet, and has little to do with China.

On the contrary. China is going through the same thing, except that the companies taking away the traffic are different.

In the United States, the main force intercepting traffic is Google's own AI Overviews — Google places an AI-generated summary at the top of its search results, keeping users on the search results page and no longer directing them to external websites. This is equivalent to Google itself "intercepting" its own traffic.

The situation in China is more intense. It is not Baidu itself that intercepts the traffic, but a large number of independent AI native apps, such as Doubao, Wenxin, and DeepSeek, which directly take users away from Baidu.

The figures speak for themselves.

Data from QuestMobile shows that in April 2026, the average number of monthly visits per user to search engine apps decreased by 18.8% year-on-year, and the usage duration decreased by 11.8%. In the same period, the monthly active users of AI native apps have approached 500 million, with the average number of monthly visits per user reaching 91 times, and the average usage duration per user reaching 180 minutes. Among them, Doubao's monthly active users have exceeded 528 million by June.

528 million — this figure is already close to the 655 million monthly active users of Baidu App. Moreover, the former is still rising, while the latter is still falling.

In a survey conducted by QuestMobile in May 2026, among the 25 industry apps it observed, 28% of the industries faced a year-on-year decline in both the number of user visits and usage duration, and another 40% had at least one indicator declining. Search, automotive information and tourism are the first three industries to feel the impact of AI information distribution. 69.4% of users in the online travel industry are using AI native apps at the same time. This proportion is 51.1% among automotive information users.

There is a figure that is particularly worth pondering. After the content of PCauto's website was cited by Doubao, Wenxin and DeepSeek, it reached a total of 12.37 million users. This sounds like good news — until you realize that this figure is 113.8 times the own traffic of the PCauto App.

113.8 times. That is to say, more than a hundred times as many users "consumed" PCauto's content through AI platforms, but these users never stepped into PCauto's website. They asked a question about buying a car in Doubao's chat window, Doubao gave the answer by citing PCauto's content, and the user closed the conversation satisfied.

PCauto contributed the knowledge, the AI platform gained the users, and PCauto got nothing.

The companies that take away the traffic are different, but the logic of erosion is exactly the same — whether it is Google AI Overviews or Doubao, the essence is doing the same thing: read the knowledge of content creators, synthesize answers, and deliver them directly to users, skipping the original source.

04 Traffic That Never Returns

At this point, some people may think: It is true that AI has taken away traffic from search engines, but doesn't AI also cite sources and attach links? Aren't these AI products also "diverting traffic" to websites?

Theoretically yes. In practice, it is almost negligible.

Chartbeat's data shows that from December 2024 to December 2025, referral traffic from AI chat tools did increase by more than 200%. But after a 200% increase, it still accounts for less than 1% of publishers' total page views.

Contentsquare gave more precise figures in its 2026 Digital Experience Benchmark Report — AI referral traffic surged 632% year-on-year, but still only accounts for 0.2% of all traffic. Moreover, the quality of this 0.2% is not optimistic, and 53.6% of AI-referred users bounce away directly.

A 600% increase, 0.2% share, and more than half leave directly.

This set of figures outlines a very cruel picture — AI is eating a lot of traffic from search engines, but it is not "forwarding" this traffic to content creators. It just keeps the traffic on its own platform.

What is more thought-provoking is the behavior of users after clicking from AI to the website.

Multiple analyses point out that many users who click over from AI chat tools actually just want to verify whether the answers given by AI are accurate — they open the web page, glance at it, confirm that the AI is not talking nonsense, and then close the page and leave. They are not here to "consume content", but to "verify facts". This kind of access has almost no commercial value for content creators.

This creates a huge "ecological inequality".

AI companies use the content of publishers and content creators to train their models. When users ask AI questions, AI uses this learned knowledge to generate answers. Users are satisfied and stay on the AI platform. Advertising revenue (if any) stays on the AI platform. And those who created the original knowledge and content get almost nothing.

This is not just a problem of traffic distribution. It is shaking the foundation of the entire digital content economy.

Over the past 20 years, the basic logic of internet digital advertising has been built on the chain of "visit → display → click". Users visit the website, see the ad, and the advertiser pays. When the starting point of "visit" itself begins to disappear, the foundation of this entire chain ceases to exist. For media and content platforms that rely on advertising revenue to survive, this is an existential threat.

Gartner predicts that by 2028, organic traffic from traditional search engines will drop by 50%. Not 5%, not 10% — cut in half. If this prediction comes true, we will live in a completely different internet from today.

05 The Deepest Paradox

This may be the deepest paradox of this "traffic interception".

Back to Baidu at the beginning. Baidu is using AI to offset the decline of search advertising — in the first quarter of 2026, AI-driven business revenue reached 13.6 billion yuan, a year-on-year increase of 49%, accounting for more than half of the core business revenue for the first time. While using AI to kill the old Baidu, Baidu is trying to recreate a new Baidu with AI.

But the bigger problem is not with Baidu. Baidu at least has the ability to transform. What about the countless small and medium websites, content platforms, and vertical communities that were fed by Baidu's traffic diversion? They do not have the ability to build AI, nor do they have a brand that allows users to visit directly. When search engines no longer divert traffic and AI does not divert traffic either, what should they do?

For 20 years, there has been an unwritten "contract" between search engines and websites: you produce content, and I give you traffic. This contract has no legal binding force, and is only built on a simple business logic — search engines need good content to attract users, and websites need traffic from search engines to survive. Both sides benefit.

Now, AI has unbalanced this equation. Search engines (and AI platforms) find: I don't need to send users to websites, I can give answers myself. Users even prefer this way — no need to click, no need to wait for pages to load, no need to find real content among a bunch of ads.

A product that makes users more satisfied is precisely a product that makes content creators more miserable.

When in an ecosystem, the improvement of user experience and the loss of creators' interests become two sides of the same thing, this ecosystem has come to a crossroads where it must be redesigned.

AI is killing the ecosystem that feeds itself.

And no one has yet given a proper answer to what the internet will look like next.

This article is from the WeChat Official Account "GeekPark" (ID: geekpark), author: Wildcard, editor: Jing Yu, published with authorization from 36Kr.