Baidu

By Bill Hartzer · Updated 19 August 2026

What Baidu is, and where its results come from

Baidu is China's largest search engine and one of the very small number of organisations worldwide that crawls, indexes and ranks the open web entirely with its own technology. There is no partner feed. Nobody supplies Baidu's organic results, and nobody ever has.

That is worth stating up front, because the most common English-language assumption about Baidu — that it is a localised copy of somebody else's search engine — is simply wrong. Baidu has crawled the Chinese web itself since 2001 and operates the full apparatus of an independent engine: a crawler, an index (the engine's own stored copy of the web), a ranking system, a public platform for site owners, and its own large language models.

A Baidu results page reads as a portal rather than a list of links. Paid listings sit at the top. Below them Baidu's own properties are weighted heavily — Baike (an encyclopedia), Zhidao (question-and-answer), Tieba (forums), Wenku (documents), Baijiahao (its publishing platform) — with external web results following, increasingly under an AI-generated summary.

Two structural facts define the experience. Results are censored — material the Chinese state has ruled impermissible is not returned, and the user is not shown what is missing. And the line between advertising and organic results has historically been drawn far less sharply than on Western engines, with documented real-world consequences.

How the crawler and index work

Baidu's crawler is Baiduspider. Its primary user-agent string declares Baidu's own documentation URL — the crawler identifying itself. A mobile variant announces an iPhone user agent carrying the same Baiduspider token, a rendering agent fetches JavaScript-dependent resources, and separate vertical agents exist for images, video, news and advertising.

Because Baiduspider is among the most heavily spoofed user agents on the web, Baidu documents reverse-DNS verification against its own domains so site owners can tell the real crawler from impersonators. Baidu states that Baiduspider observes robots.txt; practitioner reports going back years describe aggressive crawling and slow adoption of newly added disallow rules, a complaint from webmasters rather than a stated policy. It has also historically ignored crawl-delay directives, Baidu directing publishers to set crawl frequency through its webmaster platform instead.

The index is overwhelmingly Simplified Chinese and China-hosted. Sites hosted outside mainland China, and sites without an ICP licence, are crawled and ranked far less favourably — partly a consequence of latency at the national network border, partly a ranking choice. Coverage of the English-language web is poor by comparison, and this is the single most consequential difference between Baidu and a global engine.

Filtering is not a final scrub applied on the way out. It runs through what is crawled, what is retained and what is served — which is why some queries return an official notice and others return nothing that looks unusual. It is structural, not occasional.

Who runs Baidu, and how it makes money

Baidu, Inc. is a publicly listed company, not a state-owned one. It has traded on Nasdaq as BIDU since 5 August 2005 and holds a secondary Hong Kong Stock Exchange listing as 9888 since March 2021, taken out as a hedge against US delisting risk. Co-founder Robin Li remains chairman and chief executive as of August 2026. The listed entity is Cayman-incorporated and reaches its Chinese operations through a variable interest entity structure — the standard arrangement for US-listed Chinese internet companies, in which shareholders hold contractual rights to the economics rather than equity in the operating business.

The revenue model is advertising: keyword-auction search ads, plus in-feed advertising across the Baidu App and its content network. Non-advertising revenue comes from Baidu AI Cloud, sold as metered infrastructure and model access, and Baidu consolidates iQIYI, a subscription-and-advertising video service it controls.

One detail about the ad system matters more than the rest. Baidu's auction historically let advertisers occupy the top of the page in a format visually similar to organic results. The correction came from regulators, not Baidu: after the 2016 Wei Zexi case, authorities required clearer labelling of paid listings and capped the share of a page they may occupy.

Privacy, jurisdiction and state access

Baidu publishes a privacy policy and, under China's Personal Information Protection Law of 2021, offers Chinese users consent, access and deletion mechanisms. It stores their personal information inside China, as data-localisation rules require.

Jurisdiction is the whole story here. Baidu operates under the Cybersecurity Law of 2017, the Data Security Law of 2021, PIPL, and content rules administered by the Cyberspace Administration of China. Under the Cybersecurity Law and the National Intelligence Law of 2017, Chinese organisations must support and cooperate with state intelligence work. Baidu must retain user logs and produce data on lawful request. No product setting alters that.

There is a documented technical finding as well as a legal one. In February 2016 the Citizen Lab at the University of Toronto found that Baidu's Android and Windows browser and SDK components transmitted GPS coordinates, search terms and hardware identifiers with weak or absent encryption, exposing them to interception by third parties as well as to Baidu. Baidu subsequently patched the components.

Chinese law also requires real-name registration for many internet services, so a logged-in Baidu account is generally tied to a verified mobile number and thence to a national identity document. Account search history is attributable to an identified individual in a way that is not typically true elsewhere.

AI: ERNIE, and a rebuilt results page

Baidu moved into large language models earlier than most Chinese peers, and builds the whole stack itself.

ERNIE (文心) — Baidu's model family, which began as a pretrained language model in 2019 and has been applied to search relevance since.
ERNIE Bot (文心一言) — demonstrated March 2023 to a poor market reception, opened to the Chinese public in August 2023 after one of the first CAC approvals for a public generative-AI service. ERNIE 4.0 followed in October 2023.
ERNIE 4.5 and ERNIE X1 — a foundation model and a reasoning model, launched 16 March 2025. Baidu open-sourced the ERNIE 4.5 family on 30 June 2025, reversing its closed-model stance in response to DeepSeek.

Baidu has rebuilt search around generated answers, placing AI summaries above results and adding conversational follow-ups in its app; Robin Li has publicly called this the largest change to Baidu Search in a decade. Whether those answers can be permanently switched off is unclear — no documented, persistent setting to disable them has been confirmed as of August 2026.

The models are Baidu's own throughout, as is the surrounding infrastructure: PaddlePaddle, China's most widely used domestic deep-learning framework, and the Kunlun AI accelerator chips. Baidu depends on no Western AI vendor, and under US export controls could not.

Market share, and why the number is contested

The citable figure is 50.83% of the Chinese search market in July 2026, per StatCounter GlobalStats across all platforms. The same table reports Bing at 22.84%, Yandex at 13.2%, Haosou at 8.95%, Sogou at 2.12% and Google at 1.86%.

Quote that figure with its methodology attached. StatCounter measures page views on sites carrying its tracking code. In China that panel is small, skewed toward internationally oriented and expatriate-facing sites, and distorted by VPN traffic. A Yandex share of 13.2% in China is not a credible description of mainstream Chinese search behaviour, and if that row is inflated the Baidu row is depressed by the same mechanism.

Other measurements run materially higher. Statista's China search page notes Baidu "controlled about 70 to 80 percent" of Chinese online search, though its current chart is paywalled and it cites StatCounter as a source. Chinese domestic panels have historically placed Baidu in the 60–75% range. An industry report, Chinese search engines v4.5 of 29 January 2026, gives Baidu 54.64% for 2025 and 55.02% for 2024 — but its Bing figure of around 24.5% suggests it is itself StatCounter-derived, corroborating the trend rather than checking it.

These numbers should not be averaged, and none is settled. What every methodology agrees on is direction: Baidu's share has been falling for years — and not because a rival search engine displaced it. Chinese search intent moved into apps: WeChat's in-app search, Douyin for video discovery, Xiaohongshu for product queries, Zhihu for questions, Taobao and JD for shopping. None appear in a market-share table, and all take queries Baidu used to serve.

History and the events that shaped it

Baidu was founded in Beijing on 18 January 2000 by Robin Li and Eric Xu. Li had patented RankDex, a hyperlink-analysis ranking method, while working in the United States. The name comes from a Song dynasty poem by Xin Qiji — 众里寻他千百度, "hundreds of times, searching for her in the crowd".

5 August 2005 — Nasdaq IPO; the shares more than quadrupled on the first day.
2005–2008 — Baike, Zhidao, Tieba and Wenku launch, building the owned-content ecosystem that still dominates the page.
March 2010 — Google withdraws its censored google.cn service from mainland China. Baidu's share rose sharply afterwards; this, more than any product decision, established its dominance.
April–May 2016 — the Wei Zexi case. A 21-year-old student with synovial sarcoma found a hospital through Baidu's paid listings, underwent an unproven immunotherapy and died. The outcry triggered a Cyberspace Administration of China investigation and the paid-listing rules that followed — the defining event in Baidu's regulatory history.
January 2019 — Fang Kecheng's essay "Search Engine Baidu Is Dead" goes viral in China, arguing Baidu had become a walled garden mostly returning its own content. Baidu disputed it; it landed anyway.

Who it suits, who it does not, and what people get wrong

Baidu is genuinely excellent at Chinese-language web search. Its segmentation, synonym handling and query understanding are strong, and its index of China-hosted content has no serious rival. Structurally it is a real search engine: own crawler, index, ranking, models, framework and chips. Very few organisations anywhere have all six.

It is a poor choice, or an outright wrong one, for several groups. Anyone searching outside China in a non-Chinese language: the index is not there. Anyone needing unfiltered results on Chinese politics, Xinjiang, Taiwan, Tiananmen, Hong Kong or the leadership — and filtering extends beyond politics to health, pharmaceutical, gambling and financial content. Journalists, human-rights researchers and anyone whose threat model includes the Chinese state, given real-name account linkage. Anyone requiring an auditable separation between paid and organic listings, a persistent and regulator-acknowledged weakness. The index also carries a long-standing reputation for scraped and duplicate content.

Common misconceptions

"Baidu is state-owned." It is not. It is a Cayman-incorporated, Nasdaq- and Hong Kong-listed private company. It is state-regulated and legally obliged to censor — a different thing, and the accurate one.
"Baidu is the Chinese Google" / "Baidu copies Google's results." The first is superficially true and structurally misleading: Baidu is far more portal-driven, more commercial on the page, and faces a threat Google does not — search intent migrating into closed apps. The second is simply false. Baidu has crawled and indexed independently since 2001, and runs its own webmaster platform at ziyuan.baidu.com.

Frequently asked questions

Does Baidu have its own search index?

Yes. Baidu crawls the web with its own crawler, Baiduspider, stores the results in its own index and ranks them with its own systems. No external partner supplies its organic web results, and none ever has. It also runs its own webmaster platform, Baidu Search Resource Platform, with submission, indexing and diagnostic tooling — the infrastructure of a genuinely independent engine rather than a reseller.

Is Baidu owned by the Chinese government?

No. Baidu, Inc. is a publicly traded company, listed on Nasdaq as BIDU since August 2005 and on the Hong Kong Stock Exchange as 9888 since March 2021, and incorporated in the Cayman Islands. It is not state-owned. It is, however, state-regulated: it operates under Chinese content-management rules and is legally obliged to filter search results and to produce user data on lawful request.

What is Baidu's market share in China?

StatCounter GlobalStats put Baidu at 50.83% of Chinese search in July 2026, ahead of Bing at 22.84% and Haosou at 8.95%. That figure carries a caveat: StatCounter's Chinese panel is small and skewed, and the same table reports Yandex at 13.2%, which is not a credible description of Chinese search behaviour. Statista and Chinese domestic panels have placed Baidu considerably higher, in the 60–80% range.

Does Baidu censor search results?

Yes, and it is required to by law. Baidu operates under rules administered by the Cyberspace Administration of China and filters content the state has ruled impermissible. Filtering is applied to what is crawled, what is indexed and what is served, so some queries return an official notice and others simply return nothing unusual-looking. Coverage extends beyond politics to health, pharmaceutical, gambling and financial content.

Can you use Baidu outside China?

You can reach baidu.com from most countries, but it is a poor general-purpose engine outside its home market. The index is overwhelmingly Simplified Chinese and China-hosted, and sites hosted outside mainland China or lacking an ICP licence are crawled and ranked much less favourably. For English-language queries, coverage is weak. Baidu is the right tool for Chinese-language, China-hosted content and little else.

What is ERNIE, and does Baidu use AI in search?

ERNIE is Baidu's own family of language models, first released as a pretrained model in 2019 and applied to search relevance since. ERNIE Bot opened to the Chinese public in August 2023; ERNIE 4.5 and the ERNIE X1 reasoning model launched on 16 March 2025, and the ERNIE 4.5 family was open-sourced on 30 June 2025. Baidu now places AI-generated summaries above search results.

Why is Baidu losing search share?

Not because another search engine beat it. Chinese search intent moved into apps: WeChat for in-app search, Douyin for video-first discovery, Xiaohongshu for product and lifestyle queries, Zhihu for questions, Taobao and JD for shopping. None of those appear in a search-engine market-share table, and all of them take queries Baidu once served, so any share figure understates the scale of the shift.

Is Baidu safe to use for private searches?

It depends entirely on your threat model, and if that model includes the Chinese state, no. Baidu retains user logs and must provide data on lawful request under the Cybersecurity Law and the National Intelligence Law. Real-name rules mean a logged-in account is generally tied to a verified phone number and national identity document. Citizen Lab documented insecure transmission of location and search data by Baidu browser components in February 2016; Baidu later patched it.