Baidu quietly added an “AI Traffic Analysis” section to its webmaster tools in mid-September, creating the first broadly available, first-party view of how often pages are cited inside Baidu’s AI answers. Webmasters noticed the menu around Sept. 15 and Baidu confirmed the addition on Sept. 16 — but the dashboards are still empty as the report rolls out in phases.

Why this matters now

First-party visibility into AI citations matters because it gives site owners direct evidence of how Baidu’s AI is using their content — data you can’t get from third-party scrapers. Even without live data, the report’s metric definitions reveal how Baidu thinks about AI-driven discovery. For international brands and agencies working in China, that matters: this is one of the few times a major Chinese search provider separates AI-sourced traffic from general search performance.

What the dashboard reports

Baidu groups activity into four explicit metrics: AI Citations (how often a page is cited in an AI answer), AI Impressions (how often users see a link to a cited page), AI Clicks (how often users click a cited page) and AI CTR. The interface also shows a citation trend chart, a running count of cited pages, and tables of top keywords and top pages (both sorted by clicks and capped at 50,000 entries). Data is updated daily for the previous day, though Baidu warns updates may be delayed.

Three details that change how you should use the data

1) CTR is measured differently. Baidu defines AI CTR as clicks divided by citations, not clicks divided by impressions. Because citations and impressions are tracked separately, a page can be recorded as cited even if its link isn’t visible to the user (for example, if the source list is collapsed). That means Baidu’s CTR will tend to understate the performance of links that are visible. Calculate clicks divided by impressions yourself alongside Baidu’s figure to see the visibility gap.

2) The scope is limited. Baidu says the data covers AI answers embedded within its traditional search results — not necessarily traffic or citations from standalone Baidu AI products such as Wenxin on chat.baidu.com. It also covers only Baidu’s ecosystem; China’s AI search market is fragmented, with platforms like Doubao, DeepSeek, Qwen/Quark and Kimi operating independently and not sharing data. Treat Baidu’s report as one useful lens, not the whole picture.

3) Keywords are the most actionable element. The keyword table will show which queries trigger AI answers that cite your pages — a demand signal that’s otherwise difficult to obtain in China. Use those queries as testable hypotheses for content and on-page optimization, but don’t assume identical citation behavior across different AI platforms.

Practical next steps for SEOs and brands

  • Verify your site in Baidu Search Resource Platform if you haven’t already so you begin collecting data from day one.
  • Record an early baseline when your dashboard populates. Initial values are most useful for trend detection rather than absolute benchmarking.
  • Report both Baidu’s clicks/citations CTR and your own clicks/impressions CTR to diagnose whether citations are being surfaced visibly.
  • Use the keyword table to generate optimization hypotheses, then test those hypotheses across other Chinese AI platforms and in ranking performance.
  • Monitor server logs for crawler activity from other Chinese AI providers, since Baidu’s report won’t cover them.

What to watch for next

Once data flows at scale, the most valuable signal will likely be the gap between citations and impressions and how that gap varies by query, page type and vertical. That gap will tell you whether Baidu’s AI is crediting sources that users actually see, or whether sources are being recorded behind collapsed UI elements. For international teams, Baidu’s decision to report AI citations separately is a practical reminder: in China, AI visibility is becoming a measurable channel, but it requires platform-specific measurement and testing.

Watch the dashboard as it fills. The metric definitions and the keyword table give you a head start on what to measure; the next step is to turn those signals into experiments across content, metadata and cross-platform testing.