Baidu Now Measures AI Citations: Here’s What SEOs Should (And Shouldn’t) Read Into It

Baidu's New AI-Citation Dashboard Measures What Baidu Wants You To See, Not the Whole Picture

Baidu has added an AI Traffic Analysis section to its webmaster tools, tracking four metrics: AI Citations, AI Impressions, AI Clicks, and an AI CTR calculated as clicks divided by citations. Chinese webmasters spotted the addition and Baidu confirmed it in an official notice, according to Search Engine Journal, though the dashboards remain empty as the feature rolls out in phases through September 2026. A sample chart Baidu showed ran its citation axis from 0 up past 5,000 to 20,000, and the keyword and page tables underneath are sorted by clicks and capped at the top 50,000 entries.

What actually counts as an "AI citation" here?

A citation is logged whenever a page is referenced inside an AI-generated answer that appears in Baidu's regular search results, nothing broader.

Baidu tracks citations and impressions as two separate counts rather than one funnel, which means a page can rack up citations without its link ever appearing where a user can actually see and click it — buried in a collapsed source list, for example. That split, tucked into the metric definitions rather than the headline, is arguably more useful than any single number the dashboard will eventually populate.

Why should buyers question Baidu's CTR formula?

Baidu divides clicks by citations rather than clicks by impressions, a formula that can make a link's real performance look worse than it is.

Standard search-console logic compares clicks to impressions — how often something was actually seen. Comparing clicks to citations instead folds in every case where a page was referenced but never visibly surfaced, dragging the ratio down. Anyone relying on this data should also calculate clicks divided by impressions and watch the gap between the two figures as a signal in its own right, not as noise.

Does this cover China's whole AI search market?

No — the report is limited to AI answers inside Baidu's own traditional search scenario, excluding its standalone chat product and every rival platform.

Baidu's Wenxin chat product isn't included, despite some Chinese commentators assuming otherwise, and the report has no visibility into DeepSeek, Doubao, Qwen/Quark, or Kimi, each running a closed ecosystem of its own. For categories where those platforms drive more traffic than Baidu's in-results AI answers, this dashboard offers a partial, and potentially misleading, view of AI visibility.

What should buyers of GEO and AI-visibility tools ask their own vendor?

Ask exactly how a vendor defines citation, impression, and click, and whether being referenced is tracked separately from being seen.

Baidu's launch is a useful reminder that AI-visibility metrics are set by whichever platform is doing the measuring, not by any shared standard. A tool reporting citation counts for Google's AI Overviews, ChatGPT search, or Perplexity may be counting something entirely different from what Baidu counts, and none of those platforms are obligated to distinguish "referenced" from "seen" the way Baidu's own split between citations and impressions does. Before trusting a vendor's citation number, find out what triggers it and whether it was ever actually exposed to a user.

Frequently asked questions

Is Baidu's AI citation data live yet?

No. As of the rollout in September 2026, the dashboard fields are still empty while Baidu phases the feature in, and it has warned that updates may lag once live.

Can a page be cited without generating a click?

Yes. Because Baidu counts citations and impressions separately, a page can be referenced in an AI answer without its link ever becoming visible, let alone clicked.

Should GEO practitioners outside China copy Baidu's citation definitions?

Not directly. Each AI answer engine sets its own citation logic, so what holds for Baidu's traditional search scenario won't necessarily hold for Google, ChatGPT search, or Perplexity.

Source: Search Engine Journal.