Over the past decade, the most certain thing in China's internet was this: if you wanted to find information, you opened a search box. Today, that is no longer certain. Users' behavior has not actually changed—they are still looking for answers, just through a different door.
This article does not judge who is right or wrong. We only review two things: how a gateway platform slowly dismantled itself through every seemingly correct decision; and when gateways begin to drift, what businesses should keep in their own hands.
1. Users did not leave; the door changed
The change happened in two steps.
In the first step, search shifted from "standalone search engines" to "the platforms where content lives." If you want to see real user reviews, you search in content communities; if you want to browse news and check trending topics, you search on short-video platforms; if you want to compare prices and place orders, you search directly within e-commerce sites. These searches all take place inside someone else's walled garden, and redirects no longer happen.
In the second step, search shifted from "lists" to "answers." AI applications turned keyword matching into directly giving conclusions: users ask a complete sentence, receive a well-organized answer, and leave after reading it. The industry calls this phenomenon "zero-click"—the answer is seen, but the link is not clicked. It is worth noting that AI search has a significantly higher traffic conversion rate than traditional search, for a simple reason: questions people ask AI tend to be more specific and closer to the point of spending money.
According to third-party monitoring, the share of traditional search engines in China has continued to narrow in recent years. The figures from different sources vary widely and should not be cited directly, but the direction is consistent: gateways are fragmenting, and after fragmentation, there is little flow back.
To be clear, this is not a problem for one company; it is a common challenge for all "gateway-type" businesses. The essence of a gateway is a transfer—users temporarily hand over to you the power to decide "what I should believe." This power is not taken away by force; it exists only while it is being used.
2. It once held the best hand
In reviewing this, we must acknowledge the starting point: the holder of this position once had everything—the largest traffic, the most cash, the most complete index, and the earliest technology investments.
The turning point was never about capability. It did technology earlier than anyone else; models, computing power, talent, and data—none were lacking. What actually happened was the ordering of priorities: between "the user's experience this time" and "this quarter's revenue," it chose the latter again and again.
The sequence was like this:
First, ad slots were used to replace content providers' revenue; content providers could not make money in this ecosystem and shifted to their own platforms; users followed the content, because where the answers are, that is where people go; the value of the gateway declined accordingly; so it monetized more aggressively to make up for what had fallen away.
The platform did not collapse; it was dismantled layer by layer. The person dismantling it did not feel he was dismantling it—every year, he was making a correct decision that made the current-period financial statements look better.
3. How the balance tipped: there is no "user" column in the scorecard
The mechanical principle behind the imbalance is actually very simple.
A company's behavior is determined by what it can measure. Whether an ad is labeled is checkable: regulators can inspect it, legal can sign off on it, quarterly reports can state it clearly. But "after users finish searching, how many useful pieces of information remain" is not checkable: no metric, no number, no one signing off.
So what is checkable was optimized to the extreme, and what is not checkable was completely ignored. Nine ads are not a loss of control; they are the accurate execution of a real business objective.
What is more troublesome is that the metrics themselves become distorted. Numbers such as installs and monthly active users often measure "installed" rather than "in use." When the numbers on the dashboard begin to diverge from users' real experience, negative feedback breaks down—the reports show prosperity, but the experience is an empty shell, and when management adjusts the "balance" based on that, the direction is wrong from the first step.
So this does not need to be elevated to a question of character. It is not that it does not want to consider users; it is that its scorecard has no such column.
4. "Only one ad in ten results" is a genuinely viable path
Someone once proposed a counter-plan worth taking seriously: put only one ad among ten results, make that single ad the best it can be, and raise the unit price—it can still make money.
This is not a naive idea. The pricing logic of scarce media has always been this way: airport light boxes, high-end magazine covers, and prime-time television are all few, expensive, and gated. The gate itself is part of the effect: advertisers who pay a high price will care about their image and will not ruin it with garbage creative. Users get clean results, advertisers get genuine attention, and the platform receives a higher unit price—these three things can hold true at the same time.
Then why is it unworkable? There are three real obstacles, none of them moral:
First, scarcity can be expanded. If there is not enough inventory, just add one more slot. As long as someone is willing to buy the tenth slot at a low price, the temptation of that marginal revenue will always hang at the door. So "only one ad" does not rely on business rationality—rationality will inevitably lead to expansion—it relies on institutional self-discipline.
Second, the measurable always beats the more valuable. Brand effects are hard to quantify, while performance ads can be written into reports by click. When two groups compete for the same budget, the side with better numbers will always win. To sign off on the decision of "revenue declines, users improve," you need an owner who does not evaluate him quarterly.
Third, a high threshold means fewer customers. If an ad slot is sold only to big brands that can afford a high price, small and medium-sized advertisers are all shut out—and long-tail advertising is exactly what sells best.
So the conclusion is clear: "Few but expensive" is not a mistake of the business model; it is a mistake of the assessment cycle. Whoever can withstand the current-period decline can take this path.
5. Change ships, or paint the old one
Among companies that started around the same period, some changed ships: they replaced their product structure, replaced their revenue sources, and traded short-term pain for long-term survival. Others kept painting the old ship: reinforcing the gateway, consolidating traffic, and improving monetization efficiency.
Taken individually, every act of painting is correct. But the direction is to contract inward, while the world outside is scattering outward.
The harder you walk on the old map, the farther you get from the new continent.
6. What it means for businesses
Gateways will drift; this is the norm, not an accident. What businesses truly need to think through is another matter: where customer relationships are stored.
Buying traffic buys a lease term, not an asset. It is real and effective, but it stops the moment you stop spending; and the longer the lease term, the more you will find two things happening at once—unit prices are rising, and the quality of the traffic you buy is falling. The reason is not complicated: those who leave first are exactly the people who care most about answer quality; they want to trace sources, verify, and choose suppliers; those who remain are more often the people driven by marketing. A platform drives away its most valuable users yet records it under total traffic.
What businesses can decide for themselves is the other part: content that can be searched, cited, and retained.
Specifically, it means organizing real accumulated knowledge into fact blocks that machines can understand and people are willing to trust—parameters, processes, comparisons, standards, and the real details of cases. Where these things are placed determines two things: when customers come to verify you, can they find you; and when AI answers that question, are you among the cited sources.
The gateway determines who sees you; assets determine whether you can stay. The former is increasingly not up to you; the latter is entirely up to you.
This is our current judgment, and it is also the basis for how we work.