Measurement Sources and Limits
Why API Measurements Are Not Fixed Consumer Rankings
APIs, models, search configurations, and consumer interfaces are different surfaces. One API response is evidence for one disclosed request, not a permanent ranking.
Publisher: GrayRhino AI · Owner: Methodology and content owner · Review: Fact and methodology reviewer · Reviewed 2026-09-22
An API result describes one explicit request
External fact
Baidu documents Intelligent Search Generation and the separate web_summary interface, while Google says its AI features can use different models and techniques. Different interfaces, configurations, and times cannot be collapsed into one fixed consumer ranking.
GrayRhino AI method
GrayRhino AI records each API surface as a specific measurement source together with the model, question, configuration, and time.
Consumer experiences include additional product context
Limitation
Login state, personalization, region, interface behavior, experiments, and undisclosed settings may affect consumer experiences. API evidence does not cover those conditions and is not platform endorsement. Reporting an API answer as 'first place in Baidu Search' or 'visible to every user' exceeds the recorded evidence.
Before comparing two requests, inspect the interface, search configuration, model, market, language, question text, and time. Even interfaces from the same vendor should not be collapsed into a single measurement source.
Two observations from one vendor need not agree
Synthetic example
Synthetic teaching example: a standard Intelligent Search Generation API response mentions a brand for one discovery question. A colleague does not see it in a consumer interface. Neither observation invalidates the other: their surfaces, context, and timing differ. The defensible claim identifies the exact API, configuration, question, and stored answer, not a fixed rank.
If a separate web_summary interface is later added, it needs its own source identity and comparability review. The word 'high-performance' in its documentation is not a configuration of the currently measured standard endpoint. Both observations here are hypothetical.
How GrayRhino AI measures this issue
GrayRhino AI method
Reports disclose the API name and explain that its result is not a consumer-product experience or fixed ranking. Unknown providers retain their stored name to prevent mislabeling. Immutable answer evidence keeps a stable source ID even if a display label is later localized.
A defensible conclusion
Limitation
A report may say that a brand appeared for one question, source, and time. It cannot claim a permanent platform position or promise that the next answer will match.
Sources
- [1] AI features and your website
Google Search Central · Accessed:
- [2] Intelligent Search Generation
百度千帆 AI 应用开发者中心 · Accessed:
- [3] Intelligent Search Generation High Performance
百度千帆 AI 应用开发者中心 · Accessed:
Related reading
Baidu Qianfan AI Visibility Measurement: Method and Limitations
Use Baidu's standard configuration and separate endpoint to inspect a concrete API boundary.
Why Questions, Order, and Measurement Source Versions Must Stay Fixed
Check how a changed interface or configuration affects comparability.
Public measurement methodology
Review the eight public measurement rules, scope, and limitations.
Methodology whitepaper (Chinese)
Review the metric, task-matrix, evidence-chain, and version rules used by these articles.
GEO visibility monitoring
See how GrayRhino AI preserves evidence, reports, and same-condition retests.