Baidu Wen Xiaoyan enters a new battlefield

Author | Zhou Zhiyu

Editor | Zhang Xiaoling

Catalyzed by AI, the search engine battle is reigniting.

In the past year and a half, within the Alibaba ecosystem, DingTalk and Quark, Tencent's Yuanbao, 360, as well as Douyin's Bean Bag and others have launched deep search functions or tools centered around AI search. Beyond traditional search engines, users' choices have become more diversified. Targeting the search giant Baidu, both new and old players are making moves one after another.

Baidu is also making significant moves. Along with this year's adjustment of Baidu's mobile ecosystem strategy, Baidu Search is continuously advancing the iteration of search in the AI era internally. On September 4th, Baidu announced that the Wenxin Yiyuan App will be rebranded as Wenxiaoyan, simultaneously launching a series of new AI functions. With this move, Baidu has officially entered the new search engine battle.

From the PC era to the mobile internet era, facing over 20 years of technological waves, Baidu has continuously accumulated the capabilities of search engines, building a moat for its business; now, as AI reshapes various industries, Baidu also aims to use AI to reconstruct the foundation of search and provide its solution for the new search.

Xue Su, Vice President of Baidu and Head of AI Innovation Business, stated that future AI products, especially generative AI products, will possess capabilities that many past utility products do not have.

Of course, as a new application, the new search also needs to acquire enough users to generate strong stickiness among users. Just like how people used to think of Baidu when thinking of search, after achieving this goal, it can become a new traffic entry point in the AI era and lead to more applications and commercial changes based on it.

This new search battle has just begun.

Strategies

What is AI-era search? Currently, players in the market are exploring, with the main changes focused on the "search" process.

Fu Sheng, Chairman of Cheetah Mobile, which has invested in Mita AI Search, once stated that the current form of AI search saves users the process of clicking on web pages, selecting web pages, and viewing web pages, using large models and AI to do this.

In other words, the most significant difference between AI search and traditional search engines lies in the greatly improved efficiency.

This efficiency improvement is also due to a significant reduction in barriers, as well as the diversification of search forms and presentation forms.

When using traditional search engines, besides locking the "keywords" to two or three words, users often need to search for time, use "site:" to limit the website, and other operations to more accurately obtain the search results they need. That is, when using search, users manually filter information once, and then filter again from the results provided by the search engine.

The logic of the new search changes with the support of large models. AI capabilities enable Wenxiaoyan to excel in fuzzy, complex logical reasoning. With the support of the large model of Wenxin, it has generative AI, multi-round interaction, personalized interaction, intelligent entities, and other functions. Users can perform voice search, image search, and ask vague questions.

A typical scenario is when a user wants to find the name of an actor in a scene from a movie or TV show. Without the need to input precise information, simply by using natural language to input vague information through voice to Wenxiaoyan, even with incoherent expressions, Wenxiaoyan can find the actor's name and even output the names of other actors in the scene. Users can further interact with Wenxiaoyan on this basis This is also a concrete manifestation of Wen Xiaoyan's thoughts on the new search. Beyond "search," there should be more scenarios such as "create" and "chat."

Xue Su explained to Wall Street News that "search" implies a passive behavior, a responsive product. Tool functionality is a very obvious feature, it comes when called and goes when dismissed, not proactive.

"New Search" goes beyond tool functionality to make search intelligent assistants more human. This is also one of the considerations for upgrading the Wenxin Yiyuan app to Wen Xiaoyan, hoping it will enter the user's life as not just a tool, but also a user's experienced friend, companion, and confidant.

Based on this consideration, Wen Xiaoyan also has memory and subscription functions, somewhat similar to a personal AI assistant, capable of creating highly realistic chatbots. Through multi-round interactive conversations with users and based on understanding users, it can become an assistant with different personas, even providing emotional value to users in conversations.

For "New Search," under AI empowerment, search should not just be a traditional answer machine, but should be able to form reasoning and generative capabilities on the basis of AI capabilities; users' long-tail needs should make their search results more diversified, with support for capabilities such as multimodal large models, and also be able to create music and videos in search results.

All of this will make new search applications fundamentally different from traditional search.

Of course, it is still in an exploratory period. Xue Su also admitted that Wen Xiaoyan is still just a phased answer, perhaps in half a year or a year, Wen Xiaoyan can grow into a product that better fits its name, goals, and vision.

Subversion

Baidu's launch of Wen Xiaoyan is due to the fact that although the landscape of the search market has not undergone major changes, facing the changes brought by AI, it needs some bolder attempts.

Before Wen Xiaoyan, the new search that C-end users encountered from Baidu was mainly on the first page of Baidu search results, incorporating generative AI. Now, the launch of Wen Xiaoyan is an important step for Baidu to attract more users to the new search.

If the new search further develops, it can subvert traditional search and the ecosystem formed by content apps in the past. This will also be a key battle for Baidu to win the search market defense in the AI era.

Since 2000, Baidu has been the big brother in the Chinese search industry for 24 years. In the PC era, it replaced portal websites, controlling the main traffic entry point of the Chinese Internet; in the mobile Internet era, different apps formed their own information and content barriers, forming relatively independent search capabilities based on their own databases.

Apps like Douyin and Xiaohongshu, while rising and dispersing user attention and traffic, have also formed information silos and data islands. Many users have become sticky to these apps, even using them as content search tools.

Baidu has entered a defensive phase from its previous high-growth stage, with a market share in the search engine market stabilizing at over 60%.

Large models bring Baidu new opportunities. The capabilities accumulated by Baidu in traditional search in the past, as well as the content ecosystem built by products and applications such as Baidu Tieba and Baidu Wenku, allow it to have an advantage in underlying data over other tech giants in the AI era The so-called new search, the most basic ability, is still to provide users with more accurate search results while responding more promptly. This is the fundamental reason why users are willing and dare to trust the new search.

However, search is still just a temporary form, the purpose is only to allow users to smoothly transition to the new form of search. After product and technological iterations, the new search should also have new interactive forms and product models based on search.

In addition, with more users using it, the new search can also enrich the content ecosystem behind the search. The reason is simple, only with a strong content ecosystem support, search results can be more precise, and users can be retained to a greater extent.

Seeing the disruption of the search market by AI, as well as the opportunity to create a completely new traffic entry point, many players have accelerated towards this track over the past year. There are startups like Mitata AI, Monica, as well as established companies like 360, iFlytek, and Douyin; the capital market also has high expectations for this track. Just in August this year, Perplexity, founded only two years ago, received investment from SoftBank Vision Fund II under Masayoshi Son, and its valuation increased from $1 billion in April to $3 billion in August.

From Baidu's perspective, now is a good time to launch Wenxiaoyan.

Xue Su also mentioned that when Wenxinyiyan was first launched, it was mainly based on large models, targeting mainly the B2B market and developer community, of course, there were also some trendsetters who understand AI. However, after one year of operation, it has grown from adolescence to adulthood, aiming at the consumer market and consumer products, targeting a larger user base.

Wenxiaoyan is a vanguard of Baidu's entry into the new search channel from the traditional search. In the process of exploration, it can move faster, bolder, and more aggressively. It will advance side by side with traditional Baidu search, with complementary capabilities.

Of course, it is premature to say that new search will overturn traditional search in the short term. The inference cost of AI search on data is much higher than that of traditional search, and commercialization is still under exploration.

Compared to changes in the landscape, it is more likely to be changes in scenarios. The search battlefield is shifting from mobile apps to terminal-side inference on devices such as mobile phones, computers, and smart speakers.

Xue Su also believes that future search products will be a combination of AI and product results. There are increasingly new possibilities outside of search. "This is still an area we are observing, the changes are happening too fast."