Trace free text through every outgoing route
Map the search field, URL, search provider, page measurement and other recipients. A separate search service may need the query to return results while analytics does not. Review the recipient and purpose of each route separately.
Use a clearly fictional marker, never actual personal information, and inspect requests in a safe environment. Look at URLs, referrers and parameters as well as custom events. Hiding a value in the reporting interface does not prove it was never transmitted.
Test the boundary of the collection contract
Decide which groups may be sent, such as none or some results and approved destination types. Do not quietly add a free-text debugging field. Rare combinations can require restricted reporting even when raw query text is absent.
Run the same fictional search, change its wording and reload the result view. Recorded events should follow your definitions of an attempt and a view. Another event does not necessarily represent another visitor.
| Observation | Supported conclusion | Unresolved question |
|---|---|---|
| Zero results recorded | An observed attempt returned no match | Was relevant content available or was the request outside scope? |
| Query reformulated | Another attempt may have occurred | Correction, refinement or a new need? |
| Result opened | A destination type was selected | Did the destination solve the task? |
| Signal recorded twice | Possible duplication needs review | Two triggers or two actions? |
| No signal in the period | No event was observed in this sample | Was collection active and available? |
State the denominator and the gaps
Record the period, language, content scope and technical states that could be observed. A zero-result rate needs an explicit denominator. Do not mix searches, people and sessions in the same calculation.
Explain changes in measurement definitions before comparing periods. Fewer recorded searches can follow a change in collection. A selected result indicates a click rather than automatic understanding, a qualified lead or revenue.
Use task tests before commissioning content
Choose a bounded issue from the report and reproduce the task manually with prepared terms. Check whether content, a title, the index or the user journey caused the problem. The existing search task guide provides a structure for this review.
Report the observed pattern, sample, alternative explanations and next owner. Google's contractual PII definition is not a definition of all personal data. Your requirements for collection and access need their own agreed basis.