Author | Lu Yao

The topic of whether AI search will replace traditional search has been discussed too many times.

The reasons supporting the "replacement theory" are: AI provides faster, more accurate, and more integrated results, and traditional search has even become a tool for AI search calls.

However, in reality, this argument is clearly unfounded.

At present, traditional search browsers can be roughly divided into four tiers: universal search engines represented by Google and Baidu; default browsers represented by Edge and Safari, which are closely integrated with the Windows operating system and Apple ecosystem, respectively, and have a large user base; as well as Firefox, 360, UC, Quark, and other distinctive third and fourth tiers.

Search is just an action from the user's perspective, but its development has gone through several stages.

From the earliest directory search based on manual classification, to keyword matching, and then to using crawlers to capture web pages and store their content in an index database, search engines have continued to evolve over more than two decades. With the increasing maturity of semantic search and intelligent recommendation technologies, the ability to understand user needs has significantly improved. Nowadays, search engines can not only provide more accurate results but also give more personalized information by mining user data.

It is important to note that whether it's hyperlink jumps, box searches, keyword searches, semantic searches, Baidu Encyclopedia, Google, or Wikipedia, these entries and content are still manually written. The essence of the results users obtain - content created by humans - has not changed.

Perhaps considering that the technology of twenty years ago is vastly different from today's big data, AI, and mobile internet, new search concepts and products have been proposed in the current search field. There is even a belief that emerging technologies can disrupt decades of accumulation in traditional search.

However, current AI search is more focused on a narrow scope, believing that simply changing the user's interaction mode and improving the relevance of search results can achieve "redefinition." But this should actually be called an "answer engine" rather than a true "search engine."

In fact, search is essentially an ecological concept.

From a broader perspective, search is a complex system encompassing technology, data, content, users, business, and other aspects. Take Baidu as an example; it not only provides search services but also has a search ecosystem accumulated over decades. Baidu entries, Baidu Encyclopedia, Baidu Post Bar, Baidu Library, and other ancillary products are all based on its ecosystem.

Founded in 2000, Baidu achieved its first-year profitability around 2004 through pay-per-click advertising. At that time, the internet industry was in a stage of rapid development, with increasing demand for information search among users, and enterprises also recognized the value of online advertising. As China's largest search engine, Baidu had a huge user base and market share far exceeding other competitors.

Analogous to the mobile phone market, Apple's success lies not only in its hardware but also in its operating system, application ecosystem, developer ecosystem, and the entire industrial chain layout, making it difficult to surpass.

Years ago, Robin Li proposed the concept of "Box Computing," providing users with a one-stop service based on the internet. The core is that users can find what they need instantly. "On the one hand, Baidu keeps as much traffic within its platform as possible; on the other hand, Baidu will have more say in the distribution mechanism."

This is a rather high strategic perspective and almost represents Baidu's ultimate vision for its ecosystem.

It is understood that after the concept of Box Computing, Baidu launched an open platform. Less than a year after its launch, it jointly launched over 600 applications with partners on the PC main site, covering search, daily travel, entertainment, e-commerce, investment, and other fields. At the same time, Baidu also launched the Baidu Mobile Open Platform, centered on mobile devices, intending to transfer the Box Computing strategy.

Looking at a longer timeline, we can see that whether it's Box Computing, the open platform, Baidu Light Apps, or later investments in many "middle page" companies such as iQIYI and Qunar.com, the underlying logic is to further expand the ecosystem by gathering offline scenarios, allowing users to complete a series of activities such as search, information acquisition, consumption, and services within the search box.

In the PC era, products such as Baidu Encyclopedia, Baidu Map, Baidu Zhidao, and Baidu Wenku grew alongside Baidu Search. These products originally parasitized on the main site and were nourished by the huge traffic generated by the search engine, supporting more vertical endogenous products. While enriching the product line, it also further segmented Baidu's original user base for subsequent commercial operations.

However, in the mobile internet era, although the search box has not disappeared, people's reliance on native apps has increased significantly. In-app search and in-app conversion have developed new user habits, and search has been continuously fragmented, indeed weakening the traffic to Baidu's search entrance.

Unlike Baidu, Google's search ecosystem is a different concept.

Google also has a suite of products, such as Gmail, Maps, and Translate, which attract a large number of users. However, Google's smooth transition from PC to mobile devices largely depends on its complete operating system for mobile phones. In addition to the Android open-source project, it also includes GMS (Google Mobile Services), which is crucial to the mobile phone ecosystem.

Features such as the notification system, account system, software authentication system, and payment and revenue sharing system in GMS provide convenient service experiences for mobile phone users while also providing profit models and revenue sharing mechanisms for developers. Mobile phones with GMS are considered part of the Google ecosystem, and the overall user experience of phones without GMS will be affected.

From search engines to terminal operating systems, this is something that Baidu has not achieved. Google's case can be considered the ceiling in the current search field.

No matter what kind of search, there is always substitutability between products.

As China's second-largest search engine, Sogou witnessed fierce market competition and limited market share from its inception in 2004 to its delisting in 2021.

In fact, Sogou has also tried multiple businesses during its development, including Sogou Map, Sogou Reader, Sogou App, and Sogou Hao. However, these businesses did not form effective synergies, instead increasing the company's operating costs and risks.

Why is this the case?

The reasons are complex, but essentially, the search market has vast potential but a small pie. The top player often occupies a significant portion of the market share and profits, so even Sogou, which once covered over 560 million people, faced significant survival pressure.

After all, when you enter keywords or questions into a search box, the result options matched by different browsers are often similar, and there is only one closest to the answer.

Most current AI search software is evaluated based on daily and monthly active users. However, if we extend the timeline, these metrics are not stable. Imagine when a software is first launched; many people use it out of curiosity or for evaluation, resulting in significant overlap in user data across different products. A more critical issue - user retention rate - is often overlooked.

In the AI field, the base model is the foundation for building various applications, and open-source architectures provide more flexibility and scalability. For example, open-source models like Llama are encapsulated in the technology stacks of many AI software both domestically and internationally, leading to similar output effects across different software.

However, to ensure stable operation, traditional search engine vendors still prefer to retain and optimize their technology architectures and algorithms. Similarly, users are more accustomed to using familiar search methods and tend to be conservative about trying out new interaction modes or functions.

Traditional search focuses on the ecological concept and the breadth of universal search, while current AI search is primarily focused on the depth of search. This means that the commercialization path of traditional search poses a significant challenge for AI search.

Furthermore, in the past, we believed that users would come and go on Baidu. However, this is inaccurate. Traditional search often provides numerous result options, requiring users to continuously retrieve and browse until they find the most suitable answer. On the other hand, the search ecosystem also allows users to browse more content.

There is an unwritten rule in the internet industry that user engagement time is one of the critical indicators to measure the value of a product or service.

In the first stage of the internet, people mainly focused on user numbers and scale. In the second stage, companies like Meituan and Douyin paid more attention to the LTC (Lifetime Customer Value) of users, with those having longer user engagement time and higher potential value being more favored by investors. Especially in the mobile internet era, vertical traffic conversion within apps is valued highly by investors in terms of LTC.

In contrast, AI search pursues faster and more accurate results, actively shortening the user's dwell time on the app. It is difficult to guarantee that users will not simply use and leave after getting the answer, making it impossible to measure with traditional business logic.

Let's take a look at Perplexity, a star in AI search. This team, previously involved in databases, has iterated and optimized its functions centered on "search and answer generation capabilities" since its launch in 2022. Perplexity's charm lies in its RAG technology, which enables large language models (LLMs) to connect to external knowledge bases, allowing users to essentially converse with any data repository.

Simply put, in vertical fields like healthcare, manufacturing, and education, Perplexity can transform into a specialized information assistant. Each company has its knowledge assets, and Perplexity can serve as a closed database for these companies. The company recognizes that accumulating users around its products is more important than training its models.

Focusing on specific fields or user groups before gradually expanding into the market provides another possible commercialization path for AI search.

For Kimi, relying solely on the narrative of "redefining search" is insufficient. The most important thing is to maintain imagination and continuous improvement in the future search landscape.

Currently, domestic valuations of AI startups often reference overseas business models and primary markets. This overseas benchmarking-driven market bubble is not uncommon in China. Everyone is groping in the dark, unsure of the future direction, with many enterprises and investors emulating foreign practices while ignoring the significant differences in user habits between China and abroad.

Let's consider another question: If the future form of search will indeed undergo significant changes, have industry giants like Baidu and Google made substantial alterations? The answer is clearly no.

Even for Baidu, which has always emphasized AI, AI only plays an auxiliary role in its search products.

On the one hand, according to Baidu's annual DAU and other user data, it maintains a stable level of around 200 million, and the existing search model can still meet the needs of most users. On the other hand, by analyzing Baidu and Google's revenue over the past four years, we find that there has been little fluctuation in these companies' revenue since 2020, and external changes have not had a significant impact.

It is worth mentioning that although the annual revenue of Baidu Mobile Ecosystem Group (MEG) has not been detailed, relevant reports indicate that its revenue contribution has consistently accounted for about 60% of Baidu's total revenue, making it one of Baidu's most profitable business groups. Recently, Baidu Health Group (HCG) was integrated into MEG to strengthen the synergy between health business and mobile ecosystem, signifying that MEG's market competitiveness remains Baidu's core barrier.

Even the advent of AI has not broken this situation.