With AI serving as the "golden formula," Baidu Search has quietly transformed, growing in capabilities. Users now perceive the once-familiar search box as significantly "larger" and more versatile.

This "largeness" pertains to the expanded content and tools that the search box can accommodate and invoke. For instance, users used to stack keywords like "family travel," "rafting," and "one day" to plan a trip, but now, they can simply express their needs in natural language.

Baidu refers to this enhanced search box as the Smart Box.

Furthermore, users now have access to a growing array of auxiliary tools during the search process, including the AI-reconstructed "BaiKan" function on the search results page, AI-generated cameras, upgraded AI assistants, and more.

However, this is just the beginning. Reports indicate that this upgrade of Baidu Search is the most significant overhaul in nearly a decade. The "largeness" signifies not just a shift towards AI-native search in form, but also reflects Baidu's new philosophy on AI search logic: the new search must maximize breaking out of the constraints of the "box" and leverage AI tools and ecosystems to transition from information acquisition to task completion.

Coincidentally, at the 2025 I/O Developer Conference, Google also dedicated a significant amount of time demonstrating application cases and envisioning the future of AI search.

Will AI disrupt search? The attitudes of Google and Baidu, the two industry giants, towards AI in 2025 are becoming increasingly clear, and the answer seems to be a resounding yes. Rather than being passive, it is better to proactively use AI to revolutionize oneself. At this crucial juncture, both players have presented their respective strategies.

AI Search: Revolutionizing the Landscape

What does AI search actually entail? Opinions vary on this question. As a derivative of ChatBot, AI search presents itself in two straightforward ways: dialogue boxes replacing search boxes and aggregated results replacing traditional web page presentations.

Following this logic, two types of existing practices have emerged. One type involves creating comprehensive assistants for web pages and applications, incorporating all necessary AI functions. The other type is more vertical and community-oriented, akin to an AI version of "Little Red Book".

The similarity in form has led to product homogenization, stemming from a failure to grasp the underlying logic of AI search. Leveraging Baidu Search's major revamp, we can re-examine AI search from search methods, processes to results.

Whether it's a search box or a dialogue box, the first step is to break free from the constraints of the "box." The so-called "box" is simply the interactive interface where users generate search behavior, which can manifest in various forms. At the current stage, the result of transforming traditional search methods is that the search box has "grown larger," and AI search as a whole is transitioning from a tool-oriented to a content-oriented approach.

On the input side, the forms of input have become more diverse, transcending the limitations of past keyword searches and catering to various search needs such as ultra-long text search, file search, and AI tool invocation. Essentially, this increases the information content of searches from the source, enhancing the accuracy of results.

On the output side, it fills in the previously missing "puzzle pieces" to match search actions with output results. Specifically, it abandons the presentation of single web page links and replaces them with multi-modal forms such as audio, video, and digital humans. For example, Baidu also considers localization needs in its output. In daily scenarios, users can inquire about the distance, ratings, specialties, etc., of nearby restaurants.

Localization relies on ecosystems, similar to the relationship between OpenAI and Apple. No one denies the advancement of GPT, but the world still anticipates the synergy produced by "iOS system + AI." The same applies to AI search. As time passes, the advantages of traditional search companies gradually emerge, and the popularity of MCPs (Multi-functional Content Platforms) allows the search AI ecosystem to expand once again.

At Baidu's "AI Day," a case was demonstrated where, after entering "cafes in the mountains near Beijing" into the search box, AI would recommend suitable locations based on the user's historical preferences, accompanied by images and videos for reference and decision-making. This process involves recalling the Baidu Map component to locate the position and invoking relevant MCPs to complete a series of consumer behaviors such as ordering and group buying, highlighting the features of cross-platform linkage and internal and external ecosystem integration.

According to official sources, currently, Baidu Search's MCP service includes over 18,000 MCPs, covering various scenarios such as life, finance, and e-commerce. In the AI search ecosystem, two value fields have formed. On the consumer end, users utilize search tools to meet their needs, while also providing traffic and potential marketing conversion opportunities for businesses. As contributors to the ecosystem, businesses not only serve users but also create value.

In the new AI search ecosystem, Baidu maintains its role as a "link." For instance, in the home decoration field, Baidu cooperates with a company that generates 3D home decoration designs using AI, providing it with traffic exposure and promotion.

The company's AI application has been well-received by Baidu users, allowing them to directly generate panoramic home decoration schemes simply by expressing their ideas in colloquial language. The data conversion is very intuitive, with the number of users this company acquires from Baidu Search already accounting for 25% of its overall users.

We observe a phenomenon where AI search acts like an integration center, where numerous scattered functions can be broken down and recombined. This signifies that the boundaries of search have extended far beyond mere information acquisition.

In this upgrade of Baidu Search, a series of functions have been added: an upgraded built-in AI assistant with a new video call feature, enabling interaction with the real world to solve search problems; a one-stop workspace to fulfill the needs of one-click text generation, graphic creation, and code; enhanced deep search capabilities, simplifying complex problems through multi-level reasoning of thought chains and multi-stage autonomous reflection abilities of large models. The search results page has bolstered rich media capabilities, supporting mixed layout and output of images, videos, notes, and chart content, and can also invoke AI tools such as MCPs and intelligent agents to identify products, plan itineraries, etc.

What you think is what you get, what you see is the result. The logic of AI search has undergone a fundamental shift, transitioning from information acquisition to task completion. It functions like a "personal butler" that utilizes all tools to handle tasks for you. Instead of a long list of links, users receive directly usable decoration schemes, travel plans, and videos.

Master of Technology Integration

Search has traversed the cycle of eras, evolving from the Internet age to the present. The biggest variable driving AI search onto the path of revolution comes from new technologies. To some extent, search serves as Baidu AI's super entrance and also Baidu's technology super entrance.

Some so-called AI search applications on the market partially utilize shell search APIs to pass themselves off as genuine, causing market confusion. However, the threshold behind AI search is actually very high, involving professional knowledge bases, ultra-large-scale indexes, RAG (Retrieve-and-Generate) capabilities, etc., which require numerous core capabilities to support.

"Baidu Search's past accumulation has integrated a set of solutions, which determines the ceiling of AI search."

This major revamp of the AI search box can be seen as a display window for Baidu's AI technology integration. Today, Baidu has formed a three-tier AI technology ecosystem, with the bottom layer being an open model ecosystem centered on the self-developed Wenxin large model, assisted by DeepSeek. The middle layer comprises the "intelligent agent + MCP" service for executing tasks, and the top layer is the visible Smart Box, BaiKan, and AI assistant.

By accessing the DeepSeek model first, Baidu set the tone for its open source initiatives. At the end of June, the Wenxin 4.5 series opened source with ten new models. In multiple mainstream evaluations, the Wenxin 4.5 model surpassed popular models like DeepSeek-V3 and Qwen3 in comprehensive strength, once becoming a hit in the overseas AI circle.

It's worth noting that Baidu released "MuseSteamer," a video generation model developed by its commercial team, along with the creative platform "HuiXiang," and Baidu Search has also integrated MuseSteamer.

Compared to competitors like Keeling and Hailuo on the market, MuseSteamer's most prominent feature is integrated sound effect generation, meaning users receive a complete video that integrates visuals, voiceovers, and background music. Staff at the event booth told us that their future goal is to continue lowering the user threshold by launching various templates for users to freely replace and modify.

Reportedly, this project is internally known as "Steamer" at Baidu, with the R&D team hailing from Baidu's commercial team. It has been relatively low-key before its official debut, with the project initiated around the Spring Festival this year. It took two months to process data, started training in April, and won first place in the VBench I2V image-to-video ranking in early June.

There was a time when large companies and AI entrepreneurs were obsessed with "starting from scratch," aiming to completely abandon old things with an Internet flavor and establish a new AI-native world. But today, behind the search revolution lies new technology. Traditional search entrances come with inherent traffic attributes. As long as users are willing to use them, they are the best "vessels" for AI-native search.

Perhaps, as Zhao Shiqi, Vice President of Baidu and General Manager of Baidu Search, said, if products must change, they cannot be defined by the original search engine products. "Forgetting search means forgetting the old form of search."

The Change and Unchanging of AI Search

Across the ocean, another search giant, Google, has also demonstrated its emphasis on AI search.

At this year's I/O Developer Conference, Google rolled out a combination of "AI Mode + Task Assistant." During the demonstration, with just one sentence, AI search completed the entire process of seat selection and ticket purchase for developers.

As "veterans" in search, Baidu and Google hold national-level traffic and products, and must face the dilemma of trade-offs and balances in the process of seeking change. On the issue of AI search revolution, Google and Baidu share similar visions, but their self-reform determination creates a gap.

Google once temporarily suspended the advancement of AI search due to a deadlock in communication with advertisers, while Baidu has been more aggressive. Previously, there were upgrades for single-point AI search functions. This major revamp is a full-scale release, skipping the AI switch mode. It's not just an improvement in the appearance of the results page but also a full-link leap for Baidu Search from information search to goal achievement, and it directly places the results on the main search page used by hundreds of millions of people every day.

In response, Zhao Shiqi explained, "Baidu is more urgent and hopes to become faster."

Technology is advancing rapidly, and AI search continues to evolve, perhaps welcoming a new change in three or six months. As of now, there is no final conclusion on AI search within the industry, and change is the norm.

However, standing at the current juncture and periodically reviewing AI search, we can already discern some trends of change and constancy.

AI search understands users' needs better, and search is transitioning from humans adapting to search to search actively serving humans. Subtle changes have already occurred in Baidu's AI search. When the proportion of users asking questions in natural language increased by 18%, the proportion of complex problems solved also increased by 15%, indicating that AI search is truly effective in practice.

Humanization and personalization are the directions for the future development of AI search. A case that reflects the current thinking on AI search products is Baidu Search's "AI Travel" function.

For many users, a prompt of several hundred words is an insurmountable threshold. First, they may not be very clear about their own needs, and second, it is difficult for them to express their outputs. Considering this, Baidu's AI Travel function is equipped with options such as the number of days, location, budget, and travel mode. Selecting these options automatically forms a prompt template, instantly solving the above problems.

The entry value of the search box remains crucial; it serves as both the entry point for functions and a hub for traffic. Its value lies not just in asking questions and inputting but in clarifying users' needs and paving the way for invoking tools and executing tasks later. The result-oriented nature of search is that it no longer merely aims to give users answers but also starts finding ways to help them get things done.

Of course, some characteristics of the search core remain unchanged. Ordinary users need the fastest and most timely searches, which puts pressure on Baidu. When hundreds of millions of people are simultaneously using AI search functions, the consumption of tokens and task processing speed will be rigorously tested.

Currently, for questions that rely on speed to obtain answers, such as weather, stock prices, and dates, Baidu has invested heavily in infrastructure to enable search to complete invocation and output within milliseconds.

The accuracy of information acquisition surpasses speed in importance. While the Internet is abundant with answers, trustworthy ones are genuinely scarce. Posing questions to an AI assistant and subsequently verifying the response on Baidu indirectly highlights a prevalent issue in AI search. Baidu AI Search is actively addressing industry-wide challenges, including traceability of authoritative information, responses from real individuals, intelligent integration of authoritative mini-programs, and expert intelligent agents.

AI represents not just a change but a disruption to the search landscape.

"Baidu must change, and it has no other option but to do so." After a quarter-century, Baidu Search has embarked on a mission to revolutionize itself.