Write a comparison specification
Before reading a percentage change, write the question you want the comparison to answer. For example: did clicks to the same service-page group change between two complete four-week windows? This is more useful than comparing whichever figures happen to be visible in two screenshots.
Record the property, report, search type, date boundaries, page selection, query selection, country, device and grouping dimension. Use an explicit value such as 'all' where no filter applies. This specification becomes the reference for both exports and prevents an unnoticed filter from changing the population being measured.
Freeze settings and change one variable
Google's Performance report lets you choose metrics, grouping dimensions, date ranges and search type. Those controls determine the view you are reading. For a period comparison, preserve the non-date settings while changing the dates; do not switch from one page to the whole property midway through the review.
Capture the active filter labels before exporting each view. Compare them with the written specification. If you discover a country or device filter left over from an earlier investigation, decide which scope is intended and regenerate both sides. Renaming an old export does not correct the data inside it.
Check coverage before calculating the change
For a routine weekly-business comparison, equal-length windows with the same weekday mix are a useful starting point. State when a different comparison is intentional, such as a campaign period versus the same seasonal period in another year. Do not silently compare a partial week with a full week.
Google notes that recent performance data can be preliminary. Check the report's data state before treating the latest interval as settled. Keep the extraction time in your record so a later reviewer can distinguish a revised dataset from an unexplained spreadsheet change. Matching dates alone does not establish that demand or business conditions were identical.
Example: a country filter creates a false decline
Illustrative scenario: an earlier export contains 120 clicks across all countries, while a later export contains 70 clicks with a UAE filter applied. Subtracting these values produces a number, but it does not measure change for the same audience. The reviewer finds the filter mismatch in the saved report settings.
After recreating both exports with the UAE filter, the hypothetical figures are 60 and 70 clicks. The observed difference is 10 clicks, or about 16.7% relative to 60. These invented values explain the method, not a customer result. If the baseline were zero, report the absolute change rather than a percentage that divides by zero.
Keep the finding separate from the explanation
A defensible finding states the metric, comparable scope, dates and difference. It does not automatically credit a title edit, a new article or an agency campaign. Put possible explanations in a separate column with the evidence needed to evaluate each one, such as release dates or changes in the query mix.
Save both original exports, the comparison specification and the calculation. Have another reviewer check that the filters match before the result reaches a client report. If a mismatch cannot be corrected, label the comparison as unsuitable for a trend claim and obtain a consistent pair instead of presenting a precise but misleading percentage.