By Xiao Luyu, Edited by Yang Xuran
August 20 marked Baidu's AIDay Open Day. During the event, the Baidu APP announced multiple product upgrades, with the latest progress in the AI reconstruction of Baidu Search being particularly noteworthy.
On the surface, this is a product feature iteration; however, the deeper change lies in Baidu redefining the next step of search: Baidu's search function is evolving from "providing answers" to "helping users truly understand and directly delivering results." Currently, these advanced AI capabilities are all available to users for free.
Three years ago, and even today, there are still voices in the industry discussing whether AI will disrupt search engines. Today, the answer has become increasingly clear—AI has not disrupted search engines but has instead accelerated their evolution. Leveraging its existing advantages in user scale, data accumulation, and scenarios, the search engine continues to occupy a crucial entry point in the AI era.
Currently, the two search giants, Baidu and Google, have coincidentally chosen to deeply inject AI capabilities into their existing search entry points, allowing AI businesses to expand the search market and bring about incremental growth spaces that were hard to predict in the internet era.
After all, AI did not appear out of thin air. The foundational businesses that existed before the advent of the AI era, on the one hand, gave birth to the emergence and prosperity of AI; on the other hand, these foundational businesses will also undergo a thorough reconstruction by AI, gaining new room for growth.
This article is an in-depth value piece from the content team of "Juchao WAVE." Welcome to follow us across multiple platforms.
Delivering Results
Baidu has spent three years on AI-reconstructed search. Throughout this process, a product logic has gradually become clear: when using search, users are not just asking "what they want to know," but rather "what they want to achieve." Therefore, what search engines need to do in the AI era is actually to help users understand problems, complete tasks, and deliver results.
With this as the original intention, the two most noteworthy upgrades at this AIDAY both point in the same direction—providing users with more specific and usable "results."
The first upgrade is the ERNIE Task Engine 2.0. It is no longer just a Q&A tool but has become an end-to-end task delivery engine. It can autonomously plan, think, and execute based on user goals, completing the entire process from information collection and processing to creative expression and final delivery. The whole process is automatically closed-loop without manual intervention.
According to public evaluation results, this task engine has demonstrated strong capabilities in handling complex tasks. In July this year, the task engine topped the global engineering-oriented AI agent evaluation list PinchBench v2 with an overall success rate of 94.6%. In August, it ranked first in the SuperCLUE Lobster-type product evaluation with a score of 97.62.
However, what is more noteworthy than the evaluation scores is that, in task mode, the endpoint of the search engine has been redefined.
Today, AI search no longer just provides intermediate forms such as pages and URLs, but rather proactive task planning, content generation, and result delivery.
The second upgrade is the world's first "full-page interactive" search result format—the GUI General Knowledge Card. When users face abstract and complex questions, the search results they see are no longer just text and images. Instead, structured knowledge cards, step-by-step illustrations, and interactive animations are integrated into a complete page that can be directly operated.
For example, when searching for "the law of image formation by a concave lens," users no longer need to struggle with boring text descriptions; instead, they can move objects with their own hands and observe in real-time how the position and size of the image change. When searching for "how to calculate volume using the water displacement method," users can directly adjust parameters on the result page to intuitively see a dynamic demonstration of the water level changes.
The change in the format of search results transforms users from "seeing the answer" to "understanding the answer," because the information has been pre-deconstructed, digested, and reorganized by AI. What users get is a ready-to-use final product.
The combination of interactive pages and task execution brings longer user dwell time, deeper interaction levels, and clearer user intent recognition.
A noteworthy data point is that in June this year, the daily active users of the ERNIE Assistant increased by 83% year-on-year, and the average daily conversation rounds grew by more than two times. This at least indicates that the relationship between users and search is changing. People are no longer just typing keywords, browsing a few web pages, and leaving; instead, they are willing to continuously ask follow-up questions, interact repeatedly around a single issue, and complete more tasks within the search entry point.
As the person in charge of the Baidu APP stated, over the past two decades, Baidu Search has been shortening the distance between people and information. Now, it continues to shorten the distance between people and understanding, as well as between people and delivery.
The deeper the AI reconstruction goes, the shorter the distance between people and understanding, and between people and delivery, and the stronger the willingness of users to stay at the search entry point to complete tasks.
If we still use the framework of the old search to view AI search, it is inevitable to underestimate the significance of this change.
The logic of the old search is "keywords → links → users filter by themselves," while the logic of AI search is "intent → understanding → interaction → execution → delivery." The search business under AI reconstruction will no longer be confined to the logic of a traffic distribution entry point, but will move towards an entry point for understanding, interaction, and tasks.
After all, how could a search box capable of delivering complex task results be just a search box?
Validating Value
Expanding our vision to the entire industry, we might as well observe the actual value of AI-reconstructed search through Google's practices.
At the Google I/O conference in May this year, Google CEO Sundar Pichai announced the biggest overhaul of Google Search in 25 years, with the core being the comprehensive reshaping of the search entry point and interaction methods using AI. The new version of the search box will dynamically expand, supporting searches initiated by text, images, videos, files, and even currently open Chrome tabs.
Users can also directly ask continuous follow-up questions from AI Overviews, with the system retaining the context to form a back-and-forth Q&A close to a chat.
At the same time, Google plans to have search agents running in the background 24/7, continuously scanning blogs, news, and financial data, and pushing updates according to user settings.
Google and Baidu are taking the same path: not creating another AI entry point parallel to search, but using AI to reconstruct search itself.
This path was once met with market skepticism, with some worrying that AI chatbots would replace search engines—if users ask questions directly to ChatGPT, will Google Search's entry position and business model be weakened?
Google's current data has falsified this concern. Looking at market share, data released by Bank of America in April this year shows that Google's global search market share has not only not been eroded but has instead risen against the trend to 90%.
Looking at query volume, Google's Q1 2026 earnings report shows that AI-related queries contributed 22% of incremental traffic, while the overall query volume hit a record high. Pichai reiterated during the earnings call that AI features are strongly driving the growth of search query volume, with billions of clicks sent to external websites via AI features every week.
Data from native competitors such as OpenAI also shows a disadvantage, illustrating the unique value of search engines from another perspective.
A research report by Semrush shows that in ChatGPT user interactions, queries triggering real-time web searches account for only 34.5%, which is still declining compared to 46% at the end of 2024. This means that the vast majority of ChatGPT usage scenarios are unrelated to information retrieval; chatting, creation, and entertainment are the main battlegrounds for AI-native platforms. Among the search-related queries derived from AI-native applications like ChatGPT, the proportion that can truly be monetized will only be lower.
Search engines, on the other hand, face more explicit, higher-frequency information needs that are closer to decision-making and transactions. In contrast, Google's advantages are obvious, as it has over twenty years of accumulated search data, user habits, and ecosystem. Coupled with AI capabilities, the user value released is far greater than that of a pure AI chat assistant—which is precisely Baidu's competitive advantage as well.
However, compared to Google, Baidu faces fiercer competition for entry points. Platforms such as short videos, social media, and e-commerce have long held a large amount of vertical content and user time, and a new round of AI applications is also competing for the position of the first and last stop for user inquiries in the AI era.
Therefore, Baidu's AI search reconstruction started earlier than Google's and has been promoted more aggressively.
As early as last year, Baidu had completed the "largest overhaul in a decade" of its search, upgrading the search box to a "smart box." By this year, this reconstruction has entered a critical time node where its effects can be verified. Judging from the data on the daily active users of the ERNIE Assistant in June, the growth trend of Baidu's AI search is basically consistent with Google's: AI has not weakened search demand but is expanding the boundary of demand that search can accommodate.
Creating Incremental Growth
Google has already proven to the market that AI is an amplifier for search, and the commercial value of traditional search giants has instead gained greater room for imagination because of AI.
Baidu is following the same path, exploring further and deeper in the Chinese market, which has richer application scenarios but a more complex competitive environment.
In the Chinese internet company market, the core of platform competition has long revolved around fighting for user attention and usage habits, ultimately pointing towards traffic monetization. Therefore, the business model of traditional search is also mainly built on advertising and clicks.
But when the endpoint of AI search shifts to delivering service results and value, the old business model of the internet era changes accordingly. What users consume may no longer be just a single click, but a complete content production or task execution. The value that search can create will also extend further from "distributing traffic" to "completing tasks."
This is precisely where the future incremental space for Baidu's AI search lies. AI search can continue to extend to more scenarios such as education, office work, knowledge services, consumption decision-making, and enterprise services. The more search resembles a task entry point, the broader the commercialization boundaries; the richer the scenarios it accommodates, the more the value of search goes beyond just traffic.
The confidence for AI to open up search growth space mainly comes from three levels. Baidu's "search gene" provides an irreplaceable starting point and real, rich application scenarios; continuous "AI feature iteration" drives the continuous evolution of search products and experiences; the full-stack advantage of "Chip-Cloud-Model-Agent" provides Baidu with systematic competitiveness from chips to applications, running through the entire technology stack.
This combined capability is already playing a role in high-value landing scenarios.
Taking the financial industry as an example, the financial sector needs to process massive, high-value unstructured data (financial reports, research reports, regulatory documents, contracts, etc.). Baidu's knowledge enhancement and search technologies accumulated over more than twenty years give its large models an innate ability to efficiently process such professional documents.
Based on the understanding of financial scenarios, Baidu can deeply customize and rapidly iterate its own AI capabilities. Baidu AI Cloud provides full-stack infrastructure from computing power scheduling to cluster management, coupled with its self-developed Kunlun Chip, jointly ensuring the high concurrency and stability required by the financial industry.
According to a report released by IDC, Baidu AI Cloud, with a market share of 16.6%, has ranked first in China's financial industry generative AI platform and application solution market for the second consecutive year, covering 100% of systemically important banks.
For individual users, utilizing the ERNIE Task Engine to assist in quantitative trading and investment research allows tasks that previously required professional tools and high time costs to be completed within a single entry point.
These experiences, technologies, and solutions accumulated in the financial industry, through the platform of Baidu AI Cloud, form valuable experience feedback and are further replicated to other high-value industries such as government affairs, industry, and energy.
This constitutes a complete set of "Chip-Cloud-Model-Agent" Looping capabilities: the four layers of chip computing power, intelligent cloud, model capabilities, and agent applications mutually reinforce each other, forming cross-layer loop optimization, allowing the entire technology stack to continuously evolve and become stronger with use.
Looping thus opens up the incremental space for Baidu's AI businesses, including search.
AI-driven businesses have already accounted for half of Baidu's core revenue, creating a new logic for the capital market's valuation of Baidu.
In Q2 2026, legendary investor Stanley Druckenmiller established a position in Baidu ADR for the first time, and top investment banks such as Morgan Stanley, UBS, and Bank of America also increased their holdings in Baidu ADR. These capital movements all suggest that the commercial value of Baidu's AI upgrades has entered a stage of strong realization.
Navigating through cycles has never relied on betting on a single trend, but on foundations, patience, and continuous evolution. Whether it is Google or Baidu, both possess the foundation of search, the first-mover advantage in AI, and the confidence of a full-stack approach. Therefore, in this long-distance race of the agent era, Baidu and Google still have a great opportunity to successfully navigate through the cycle once again.