Define a comparable view

Choose the search type and comparison dates before interpreting a change. Keep country and device filters consistent. If your business is seasonal, compare a useful seasonal period as well as the immediately preceding period.

Record your filters next to the export. A screenshot of a graph without its filters can lose important context when it reaches another person.

Separate page and query questions

The Performance report provides clicks, impressions, click-through rate, and average position. Use those as observations. Start with a page that matters to the business, then examine the relevant queries rather than chasing every phrase in the property.

Some query data is omitted for privacy. Totals and detailed rows need not add up exactly, so treat exports as a useful view rather than a complete log of every search.

Turn an observation into a hypothesis

ObservationQuestion to investigatePossible work item
Relevant impressions, few clicksDoes the result communicate the page’s value?Align title, visible answer, and snippet expectations.
Clicks down on several related pagesDid demand, access, or search visibility change?Investigate the group before rewriting.
New queries on a product pageDo shoppers need missing details?Add accurate specifications or compatibility information.

These are illustrative questions, not diagnoses. A low click-through rate can have several causes, including query mix and search-result features.

Choose one editing experiment

State what you expect the change to improve. Keep the baseline, the affected page, and the date. If another team changes the template or runs a campaign, include that in your notes.

Close the loop

Review comparable data and customer outcomes, then decide whether to keep, refine, or reverse the change. Use average position cautiously because it aggregates different contexts. The best conclusion may be that the evidence is mixed and the page needs a different investigation.

Sources and further reading

Official documentation checked October 6, 2026. Workflow examples are illustrative editorial recommendations, not measured client results.