How Baidu Search works

Baidu’s own framing of site building is refreshingly plain: the search engine is just an ordinary visitor to the site. It crawls, and it judges value, from the user’s point of view — so any improvement to user experience is an improvement for search.

The three stages

Baidu divides the work into collection, ranking and display — getting content into the index, ranking well within it, and earning the click from the results page. Every technical recommendation belongs to one of the three, and confusing them is the usual reason effort produces nothing.

Machine-readable content

Baiduspider reads text. Flash, images and JavaScript content are not processed, and text inside them does not exist. The consequences Baidu spells out:

• Use text, not Flash, images or JavaScript, for important content and links.
• Where Flash is unavoidable, publish a text version and link to it from the home page.
• Use Ajax only for genuine interaction — never for navigation or body content you want indexed.
• Avoid frame and iframe: content shown through an iframe may be discarded.

Site structure

Baidu recommends a flat tree that is also a network: home page → channel → article, with as few levels as possible, and cross-links between related pages at every level. Its stated rules — every page reachable by at least one text link; important content reachable from the home page or a shallow level; categories sensible rather than over-divided; breadcrumb navigation on content-heavy sites; text links rather than JavaScript or Flash in navigation.

Subdomain or directory — a genuinely Baidu-specific answer

Use a directory until a section is rich enough to stand as an independent site, then convert it to a subdomain. Baidu explains why: the main site normally carries the highest weight, and a subdomain inherits only part of it, depending on its own quality. Content unrelated to the site’s main theme belongs on a separate subdomain, because unrelated content confuses topic identification. And abusing subdomains — many of them without substantial content — is treated as cheating.

URLs

One page, one URL. Where several URLs serve the same content: use only the canonical form internally, keep session IDs and tracking codes out of URLs, 301 the variants, and block the rest with robots.txt. Keep URLs short and descriptive; dynamic URLs are fine, but every extra parameter increases the risk of the crawler falling into a black hole.