What Same-Condition Retesting Can and Cannot Show

Google says AI features can use different models and techniques, while Baidu's standard interface exposes settings that can affect generation. Generated answers vary with conditions and over time.

GrayRhino AI compares results only when the question plan, source plan, and analysis rules qualify as the same conditions, and describes the difference as an observation.

A retest can show changes in completion coverage, brand discovery, owned citations, individual answer evidence, and source classes. Keep the latest result, the previous comparable result, and the earliest comparable baseline visible so a single convenient pair does not dominate the story.

Before comparing, check question text and order, audience facts, model and endpoint settings, and analysis rules. If a condition differs, both raw records remain readable, but the report should not compute a same-condition rise or fall. A partial run also needs its complete expected denominator and a limitation.

Synthetic teaching example: four questions on two sources complete 5/8 tasks in the first run and 6/8 in the second. One answer that previously lacked an owned citation now includes a website link. The supported observation is that this question-source answer gained that citation; show both answers and URLs. It does not prove that a website edit caused the change.

If the second run changed its interface, model, or audience, a same-condition comparison is not eligible at all. A new link in one answer must not be extrapolated to every consumer interface. The numbers and outcomes here are instructional only.

The system checks comparison eligibility and preserves the latest record, previous comparable record, and earliest comparable baseline. Version mismatches are labeled incomparable rather than assigned a rise or fall. A proposed action needs an answer-level evidence anchor, not just a changed percentage.

A brand mention may rise after a website edit while models, indexes, sources, or random variation also changed. Without controlled evidence, the report states the observed difference and does not claim that the edit caused it.