Start with the unit of observation
Two columns can both be called impressions while describing different groupings. Before comparing them, write down what each row represents: a page, a query or the property as a whole. Also preserve the active filters and date range. The heading above a spreadsheet column is not enough to establish that the values can be added together.
Google explains that impression counting can be aggregated by property or by page, with the report configuration affecting the result. Several links belonging to one property can contribute differently when grouped by URL. That distinction is a counting rule, not evidence that a report is broken or that visitors disappeared.
Use a small observation ledger to understand the difference
Create a teaching ledger with an observation ID, query label and the distinct site pages involved. Keep it separate from real reporting data. The purpose is to make overlap visible: one search observation can involve more than one page from the same website. Counting those page appearances does not turn them into separate people.
For an actual review, identify the grouping used by each saved report and ask what decision it supports. Page-level evidence helps examine individual destinations. A property-level view answers a broader visibility question. Neither is automatically the more correct report; the error is using one as if it measured the other.
Example: one observation involving two pages
Illustrative scenario: an observation contains two eligible links from the same website, one to a service page and one to a guide. In a simplified ledger, the property appears in one observation, while each of the two pages appears once. Summing the page rows gives two page appearances, not two independent observations of the property.
Do not use that sum to claim that the website reached twice as many people. This example isolates overlap for explanation; it does not attempt to reproduce every search-result format or Google's full reporting rules. When analyzing a real discrepancy, consult the relevant report configuration and the official metric documentation.
Keep separate reporting questions separate
If the question is which landing pages are receiving visibility, use the page view and label its totals accordingly. If the question concerns search terms for the property, retain the query grouping and its scope. Do not add a page total to a query total: the two are descriptions of potentially overlapping activity, not independent channels.
Likewise, avoid interpreting an impression as a unique visitor or an on-site session. In a client report, write the unit beside the figure rather than relying on an explanatory footnote several pages later. A clear label is often more useful than trying to force every export into one combined total.
Investigate a mismatch in a reproducible order
Preserve both original reports. First compare the property, dates and filters, then identify their grouping and aggregation. If these differ, recreate a comparable pair before checking arithmetic. If the mismatch remains, record the exact report settings and example rows for a deeper export or reporting investigation; do not invent missing rows to make totals balance.
Conclude with what the evidence supports: for example, 'These totals use different units and are not additive.' Keep the distinction visible in dashboard labels and handover notes. This prevents the same apparent discrepancy from returning every month and keeps decisions focused on useful page and query patterns rather than an artificial reconciliation.