Baidu Bets on AI Search to Reverse Market Share Losses in China

A Search Overhaul Under Pressure

Baidu announced a major overhaul of its search engine in July 2025 as its share of China's search market stood at roughly 50% to 54%, down from around 65% in 2020. The Chinese technology company, often called "China's Google," is integrating its Ernie artificial intelligence models into its main search service, seeking to turn a familiar tool for finding links into an assistant that can answer questions, compare products and help users complete tasks.

As of July 5, the upgrade represents Baidu's most substantial search redesign in years. Its new features include support for complex questions exceeding 1,000 words, voice searches in several Chinese dialects, and answers that combine text, video and other information. The company promises greater accuracy, responses that understand the context of a question, and recommendations suited to individual needs.

The stakes extend across China's 1.1 billion internet users. Baidu faces competition from search engines such as Microsoft's Bing and from Douyin, Xiaohongshu and WeChat, where people increasingly find recommendations and information without opening a conventional search service. Younger users in particular are turning to these platforms for content discovery.

Baidu also faces financial pressure: its online advertising revenue has declined for four consecutive quarters. Search advertising has traditionally supported its business, so the company needs to attract users while finding ways to earn money from an experience increasingly built around answers rather than lists of links.

The overhaul brings together several parts of Baidu's AI strategy: more capable Ernie models, free access for individual users, lower prices for AI services and a move toward making models available for developers to use and adapt. Its success will depend on both the usefulness of the new search experience and the revenue it produces.

What Changes When Users Search?

Baidu's redesigned search can handle long requests containing several connected questions. Someone shopping for a phone could ask which models cost less than 2,000 yuan, which take the best photographs and which would make suitable gifts. Instead of presenting only links, the system can generate recommendations, arrange products by price and bring together relevant videos and blog analyses.

Natural language search means users can write as they would speak, explaining their needs rather than trying to guess the right keywords. Understanding context means connecting the parts of that request: the price limit, the photography requirement and the intended recipient all need to shape the answer.

The upgrade includes a larger search box and a more capable voice function supporting several Chinese dialects. Baidu's mobile app is also becoming an AI chatbot that can assist with writing, drawing and planning trips. These changes expand the service from finding information to helping people work with it.

Baidu wants the system to anticipate what a user needs next. An economic query, for example, could produce a graph showing gross domestic product (GDP), alongside an explanation and material from trusted bloggers and experts. GDP measures the value of goods and services produced in an economy; displaying it in a graph can make changes easier to understand.

Zhao Shiqi, Baidu's general manager for search, said the AI features are integrated into the main search page, contrasting that approach with Google's separate AI Mode. He described the company's reasons for changing its core product:

Baidu search has to change, and we are voluntarily looking for changes. We don't intend to beat others, we need to beat our former self.

How Much Ground Has Baidu Lost?

StatCounter and The Egg put Baidu's share of China's search market between 50% and 54% in mid-2025. Against the roughly 65% figure for 2020, that represents a decline of about 11 to 15 percentage points. The figures describe a company that still leads the market but has lost a substantial part of its previous advantage.

Microsoft's Bing has risen to more than 24% of the Chinese search market. On desktop computers, it now rivals Baidu, with figures indicating that it matches or exceeds Baidu's share. Baidu retains a stronger lead on mobile devices, although Bing's mobile presence is growing quickly. Haosou and Yandex are also gaining ground.

Bing's gains have been attributed to more varied, responsive search results and its rapid adoption of AI features. Its familiar interface and access to global content also make it attractive to international brands and users.

Douyin, China's counterpart to TikTok, Xiaohongshu, also known as Little Red Book, and WeChat exert a different kind of competitive pressure. Users can find product recommendations, travel advice and lifestyle content within those apps. A person who begins and finishes a search there has less reason to visit Baidu.

Baidu is consequently competing on two fronts. Rival search engines challenge its position within traditional search, while social and content platforms compete for the underlying habit of seeking information. The AI overhaul is an attempt to make Baidu useful enough for people to bring more of those questions back.

The Ernie Models Behind the Upgrade

Baidu's search transformation relies on Ernie, short for Enhanced Representation through kNowledge Integration. The Ernie family consists of large language models, AI systems that generate responses to written requests. Within search, those models help turn a question and relevant information into a conversational answer.

The latest versions, Ernie 4.5 Turbo and Ernie X1 Turbo, are presented as faster, cheaper and more capable than their predecessors. Baidu says X1 Turbo matches the performance of leading models whose code or model resources are publicly available, including DeepSeek R1, at half the price. The company also claims advanced reasoning and planning abilities, allowing the model to work through requests involving several steps.

AI already generates around 11% of Baidu's core search results, a proportion expected to increase as the technology develops. The July redesign therefore builds on AI use already present in the service while giving the technology a more prominent role in the search experience.

Ernie also supports a wider collection of Baidu businesses and applications. These include Wenku's document creation tools, cloud computing and autonomous driving through the Apollo Project. Search is the most familiar entry point for many users, but Baidu's AI investment reaches beyond answering questions on a results page.

The main developments span several dates and periods:

• 2020: Baidu held around 65% of China's search market.
• 2024: An investment analysis projected a contraction in Baidu's revenue.
• June 30, 2025: Baidu set this date for beginning to make its next generation of Ernie models open source.
• July 2025: Baidu announced its major search overhaul.
• By July 5, 2025: The company was pursuing the redesign with its search market share at roughly 50% to 54% and online advertising revenue down for four consecutive quarters.

Why Open Ernie to Developers?

Baidu announced that it would make its next generation of Ernie models open source from June 30, 2025. Opening models gives developers access to resources they can try, build on and adapt, helping the technology reach beyond the company's own products.

The strategy is intended to accelerate AI adoption, attract developers and establish Baidu's technology as an industry standard. More people using a model can mean more applications built around it, extending its reach into products that Baidu does not create itself.

Robin Li, Baidu's chief executive, explained why wider access could encourage adoption:

If you open things up, a lot of people will be curious enough to try it. This will help spread the technology much faster.

Baidu has also made its latest Ernie models free for individual users, abandoning earlier subscription fees. That decision addresses a Chinese market where users are sensitive to prices, while price cuts for AI services seek to make the technology more attractive to businesses and developers.

Free consumer access and open source models serve related purposes: bringing more people into contact with Ernie and encouraging them to keep using it. Baidu must then earn revenue through other activities, including cloud services, partnerships and marketing tools powered by AI.

A Global Race Meets China's Divided Internet

Google, Microsoft and Yandex are also changing search through generative AI, technology that produces new text or other content in response to a request. Google's Gemini powers AI Mode, Microsoft Bing offers AI question and answer features, and Yandex uses its YATI and Neuro models. Across these services, users increasingly receive an assembled response instead of having to open several pages and combine the information themselves.

Baidu's approach places its new capabilities directly in the main search page. That could make AI part of a routine search rather than a separate destination users must choose. The company is trying to change the experience where its existing audience already arrives.

China's internet presents a particular difficulty. Much valuable material sits inside platforms that restrict access by outside search engines. These platforms are often called walled gardens: people can explore the information within them, while external services have limited ability to index it, meaning to catalogue it for search.

Douyin, Xiaohongshu and WeChat contain the product reviews, travel tips and lifestyle information that many users want. Restrictions on indexing make it harder for Baidu to gather a comprehensive selection of material for its answers. More capable AI still depends on the information the search service can reach.

Google operates from a position of global search dominance, while Baidu faces a Chinese market fragmented among powerful platforms. Baidu's challenge therefore includes both improving how it answers and widening the useful information available to those answers.

Content Choices and User Trust

Fang Kecheng, an assistant professor at the Chinese University of Hong Kong, argues that Baidu's search results have not always improved. He points partly to the company's preference for its own content platforms, including Baijiahao and Baidu Baike, instead of directing users to the most relevant external sites.

Giving priority to Baidu properties keeps users within the company's services. It can also limit the range and quality of information surfaced in search. Fang's criticism concerns the tension between Baidu acting as a gateway to the wider web and becoming a destination that supplies more of the content itself.

AI answers make those content choices especially consequential. When a system combines material into a single response, its selection of information shapes what the user sees first. A broader range of useful material matters for comparisons, recommendations and questions requiring several perspectives.

Baidu also faces concerns about accuracy, bias and transparency as AI generated content becomes more common. Its answers need reliable, current information, and users need to understand where advertising enters the experience. The company's effort to increase engagement depends on preserving confidence in the results.

Can Answers Produce Advertising Revenue?

Baidu's traditional search business earns much of its revenue from online advertising associated with search queries. AI changes that arrangement by answering more questions directly, reducing the central role of the familiar list of links.

The company is experimenting with native advertising within AI generated content. Native advertising is designed to fit the surrounding experience rather than appear as a separate display. For Baidu, the commercial task is to make such advertising useful and natural while protecting trust in its recommendations.

The original comparison is with American startup Perplexity's subscription based AI search service. Baidu, however, has removed earlier subscription charges for individual access to its latest Ernie models. Its strategy must account for the price sensitivity of Chinese users while developing other sources of income.

Partnerships, cloud services and AI marketing tools are among the routes Baidu is exploring. Investment analysis on Seeking Alpha identifies AI Cloud, Intelligent Driving and Ernie Bot as businesses positioned for monetization and possible growth. Despite a projected revenue contraction in 2024, that analysis views the focus on AI as a potential source of lasting value if engagement becomes profit.

The four consecutive quarters of falling online advertising revenue give this work urgency. The unresolved commercial question is how much the new search experience can earn, and whether revenue from other AI activities can support growth as the traditional advertising business weakens.

Investors Welcome the Push as Rivals Advance

Baidu's shares rose more than 10% following the release of new AI models that users could access for free. Investors responded positively to the announcements, while the opening of Ernie models and aggressive service price cuts were viewed as efforts to regain market share and influence among AI developers.

Competition extends beyond search results. ByteDance's Doubao chatbot, DeepSeek's open source models and Alibaba's Qwen are all gaining traction. Some competing services have surpassed Ernie Bot in user numbers, adding pressure on Baidu to turn its established reach into regular use of its AI products.

An unnamed analyst described the company's prospects in terms of whether its technology meets its promises:

Baidu's competitiveness hinges on whether its new models truly deliver on the promised performance and cost advantages.

Baidu's redesign brings conversational answers, multimedia content and personalized recommendations into a service already used at scale. Success could establish a stronger standard for AI search in China and influence approaches elsewhere. Failure would leave more room for domestic and international competitors to attract the next generation of Chinese internet users.

For users, the immediate change is practical: they can ask longer questions, speak in more dialects and seek help with tasks as well as information. For Baidu, the test is whether those capabilities restore usage, deliver dependable answers and generate revenue in an internet increasingly divided among competing platforms.

Key Points

• Baidu announced a major AI search overhaul in July 2025, integrating Ernie models into its main search experience.
• Its Chinese search market share fell from around 65% in 2020 to roughly 50% to 54% in mid-2025, while Bing rose above 24%.
• The upgrade supports complex requests exceeding 1,000 words, voice searches in several Chinese dialects and answers combining different forms of content.
• AI generates around 11% of Baidu's core search results, with that share expected to grow.
• Baidu set June 30, 2025, for opening its next generation of Ernie models and made its latest models free for individual users.
• Online advertising revenue has declined for four consecutive quarters, increasing pressure to earn money from AI search and related businesses.
• Restricted access to rival platforms, concerns about content choices and competition from other AI services remain central challenges.