# Query-mix review rules

Use Web: text-based only for text-query rows. Web: multimodal has no text-query dimension; review separately. Keep separate AI performance reports distinct. Record completed comparable periods, reporting time basis, property reference and all filters. State enquiry definition/deduplication/completeness separately.

Mutually exclusive group order: brand first, then offered-service, information, wrong-service-or-area, uncertain. A service-fit note may be additional, but do not count a query twice. Confirm service/coverage with the owner. Queries do not identify individuals or prove their location or readiness.

Record chart clicks, exported-row clicks, classified clicks and unclassified clicks separately in the review notes. Anonymised rows and limits mean exported totals may differ from charts. Query filtering can change included totals. Never call the classified-export share a sitewide share without the appropriate denominator. Inspect query-filtered Pages separately and do not add page rows again into property totals.

Worked example: earlier 20+10+10=40; later 18+12+30=60. Change=(60-40)/40=50%. Offered-service extract shares:20/40=50%,18/60=30%; difference=-20 percentage points; service clicks -2 (-10%). No real observation or statistical confidence claimed. Six accepted enquiries in each period is a separate hypothetical business-log observation, with unknown source attribution. Do not divide it by Search Console clicks to label a conversion rate.

No exports/customer details/account identifiers belong in public examples. Preserve original exports privately and use restricted evidence references. Missing information stays blank with a reason. Decision may be hold for missing data; sheet has no account connection or formula execution.