A traffic drop has a small number of possible causes and a defined order for eliminating them. Working that order is what separates a two-hour diagnosis from a two-week one.
The sequence
- Is it search traffic? Split by source before anything else. A fall in direct or referral traffic has nothing to do with Baidu.
- Is the site reachable? Check crawl exceptions and crawl diagnosis in the platform. A server or DNS problem removes pages quickly and quietly.
- Is there a platform message? Algorithm notices arrive there. This is the check people make last and should make second.
- Did index volume fall? If yes, the problem is pages leaving the index — work the index-volume causes. If no, the pages are still there and the problem is ranking or display.
- Did crawl frequency fall first? That points at quota: publishing rate, availability, or trust.
- Did something change on the site? A release, a template change, a robots edit, a URL change, a new advertising layout. Match the date.
- Did demand change? Seasonal or news-driven interest moves independently of the site.
The site-side causes worth checking first
- robots.txt changed — the fastest way to lose everything, and often shipped accidentally with a deployment.
- URL structure changed without a revision declaration or without page-to-page redirects.
- Advertising or permission changes that now breach the landing-page standard — a new interstitial, a login wall, a floating bar.
- Hosting or certificate problems making the site intermittently unreachable to the spider.
The feedback channel
Baidu’s complaint centre is a genuine route when the platform data does not explain the fall — but like any support channel it answers precise questions quickly. Specific URLs, the date the change began, the crawl and index figures around that date, and what you have already ruled out.
The discipline
Do not start with content. Content is the slowest hypothesis to test and the least likely single cause of a sudden fall; availability, directives and algorithm notices account for most of them, and all three can be checked in under an hour.
How we apply this
We work this order strictly, and the two checks that resolve most cases are the platform messages and a robots.txt diff against the deployment log. Clients almost always begin with content, which is the slowest hypothesis to test and rarely the cause of a step change on a single day.
Related services