How to retest and judge change
Retests must lock the same scope, question set, language and window. The conclusion can only be improved, declined, no clear change or insufficient evidence. A change in the same period is not causal proof of one action.
Comparable conditions
- The same brand, site, market and language.
- The same question-set version. A changed set needs a breakpoint or an intersection comparison.
- Equal-length windows, with holidays or platform delay noted.
- The same valid-sample rule. Failed requests stay out of the denominator and are counted separately.
Interference
Other publishes, competitor changes, model updates or collection failures may happen in the same period. Record them. Do not delete them and then claim a clean comparison.
How to read the result
- Improved or declined: the direction is clear under comparable conditions.
- No clear change: the sample is enough but the difference is not.
- Insufficient evidence: sample, source or time cannot support a comparison.
Sources
- GEO 2.0 — Ongoing verification, not a recommendation guarantee
- FAQ — Monitoring and execution deliver different things