Baidu Qianfan AI Visibility Measurement: Method and Limitations

Baidu documents POST https://qianfan.baidubce.com/v2/ai_search/chat/completions for Intelligent Search Generation. The same document lists baidu_search_v1 and baidu_search_v2 as search_source configurations and describes the answer and references structures.

GrayRhino AI's current Baidu source uses this chat/completions endpoint with baidu_search_v2. Immutable evidence keeps the provider ID; the interface may display the localized label 'Baidu AI Search'.

Baidu's high-performance documentation describes POST https://qianfan.baidubce.com/v2/ai_search/web_summary. It is not the same concept as the baidu_search_v2 configuration inside the standard endpoint.

GrayRhino AI does not list web_summary as an enabled measurement source. This article explains the distinction without claiming that both surfaces are currently measured. The high-performance label belongs to that separate interface; it is not another name for baidu_search_v2 on the standard endpoint.

Adding this interface in the future would require a distinct source ID, request method, cost and comparability review, and product verification. Historical standard-endpoint evidence must not be relabeled as a new-interface result.

Synthetic teaching example: a project asks an unbranded discovery question through Baidu's standard Intelligent Search Generation endpoint with baidu_search_v2 and receives one answer with two references. Its report should preserve the endpoint, configuration, question, and each reference. It must not label this response as web_summary or infer that both references belong to the brand's website.

A reference on an unconfirmed subdomain leaves website ownership unresolved. A failed call remains in the expected task population. This example does not establish a consumer-side Baidu ranking or claim that both Baidu interfaces are measured.

Each run preserves the question, model, API surface, search configuration, time, raw answer, and citations. Brands and competitors use the same task rules, while failures remain in the expected denominator. Inspect the actual interface identity before reading a brand mention or owned-citation conclusion; a broad 'Baidu' label is insufficient evidence.

The result does not represent the full behavior of Baidu Search, Baidu App, Baidu agents, or other consumer products, and it does not guarantee mentions, citations, or recommendations.