A buyer types “installation checklist” into a site's search box. The results page is empty. The marketing team adds an installation article to its calendar.
The checklist might already exist.
It could sit outside the search index, carry a different title, or disappear behind a product filter. Writing another page before checking those conditions adds material without identifying the failure.
Internal site search analysis starts with a narrower question: what happened between the submitted query and the available answer? Build an evidence row for that sequence. Then choose the next check.
The workflow below is a proposed editorial method. Every query, count, destination, and sequence in its worksheets is fictional. Use the synthetic rows to agree on classifications before examining any permitted query data.
Keep onsite queries separate from external demand
An onsite query describes a search within your website. It tells you what someone asks the local search system to retrieve. It does not establish how many prospective buyers search the wider market for that topic.
Google Search Console measures your site's appearance and activity in Google Search results. Its queries belong to that external discovery context. Google also documents omitted anonymized queries and limits on displayed rows. Neither table is a complete census of buyer questions. See Google's Performance overview and query limitations.
Keep a third column for external search-volume evidence, with its source, period, and method. Leave it unknown when no research supports it. An onsite count cannot fill that column.
Use the B2B keyword research methodology for the separate demand investigation. A phrase can deserve a better onsite answer even when its external demand remains unmeasured.
Treat the populations separately too. An employee looking for a manual, a returning customer checking instructions, and a prospective buyer can submit identical words. The words alone do not identify the person or buying stage.
Establish what the measurement records
Google documents view_search_results as an enhanced measurement event triggered by a search-results URL query parameter. Site search must be enabled for the web data stream. If the site uses the Google tag, automatic detection must also be enabled for that event. The default parameter names are q, s, search, query, and keyword. Other names can be configured. The search_term parameter populates the Search term dimension. Google's note says unique_search_term is sent only when its value is one, indicating uniqueness within the session. See the enhanced measurement reference.
That mechanism needs validation against the site's search interface. A search box existing on a page does not prove its activity enters the report. Check a synthetic query against the actual URL and event payload in an authorized test environment.
The documented site-search parameters do not supply the whole worksheet below. Do not infer result counts, filter states, refinements, or destinations from search_term. Those fields need separately verified evidence. Mark them unknown when unavailable.
For the wider event plan, use the GA4 setup guide. Here, keep the measurement question bounded: does one intended search produce the expected record, and what does that record prove?
Set privacy controls before collecting text
Google requires enhanced measurement users to prevent personally identifiable information from being collected. Its PII explanation includes email addresses, phone numbers, and full names. A query field needs review before its text becomes analytics data.
Proposed controls begin with the field's purpose. Decide whether raw text is necessary. A restricted category such as “documentation search” may answer the reporting question without retaining the submitted phrase.
Review every destination that could receive the text: event parameters, page URLs, referrers, logs, and exports. Removing a term from one report does not demonstrate that collection was prevented elsewhere.
Test with synthetic markers. Verify suppression before transmission, access restrictions, and the retention decision. Do not paste private queries into a shared spreadsheet to demonstrate the risk. Do not assume a pattern that catches email addresses catches every sensitive phrase.
If the collection boundary remains uncertain, hold raw-query analytics and use a synthetic usability exercise. Record the missing evidence. This article does not enable tracking, establish consent requirements, or certify a privacy program.
Build a worksheet with explicit unknowns
Choose a fixed review period and define the count unit before sorting queries. This worksheet uses recorded search events. It does not count people, companies, opportunities, or successful tasks.
Algolia's site search analysis guide discusses queries, filters, results, and clicks as separate evidence. The following triage worksheet is an editorial adaptation. Its classifications are hypotheses until the next check supports them.
All entries below are synthetic. A refinement is a later search recorded within the fictional example sequence. “Destination” describes an observed click, not proof that the reader completed the task.
| Query | Event count | Result count | Refinements | Destination | Working classification | Next check |
|---|---|---|---|---|---|---|
| installation checklist | 18 | 0 | setup guide | Setup overview | Terminology or indexing | Search the content inventory for the checklist |
| export formats | 11 | 7 | csv export | Export reference | Result ordering or labels | Inspect the first result titles and excerpts |
| setup guide | 8 | 0 with version filter | Same query, filter removed | Version guide | Filter restriction | Repeat with the original version selected |
| renewal process | 5 | 0 | None recorded | Unknown | Possible missing content | Ask the content owner for the current public answer |
| api limits | 9 | Unknown | Unknown | Unknown | Measurement gap | Verify result-count and click evidence |
| warranty | 4 | 3 | None recorded | Warranty page | Possible navigation issue | Inspect the page immediately before search |
Keep the submitted phrase separate from any normalized analysis label. Trimming whitespace can help group records, but do not silently merge version numbers or product identifiers. Record the grouping rule so another reviewer can reproduce the count.
Do not calculate a no-results rate from rows with unknown result counts. State the eligible denominator and exclusions. A smaller, explicit sample carries more information than a precise percentage built from assumed zeros.
Separate repeated activity from repeated need
Eighteen events do not establish eighteen independent requests. One person can retry a phrase, reload a results page, or return to the same task. A tester can also generate repeated searches while checking a release.
Keep the raw event total visible. Add a separate, permitted sequence view when the instrumentation supports it. If identity or session evidence is unavailable, state that independence is unknown. Do not manufacture a deduplicated buyer count.
Record known internal testing periods and the method used to distinguish them. Keep the excluded count beside the retained count. When internal traffic cannot be separated reliably, describe it as a limitation of the sample.
For a small sample, inspect the sequence before ranking topics. Five searches on one afternoon can justify checking a broken destination. They do not establish a recurring market need or support a forecast.
Absence also needs care. No recorded click could mean the reader found an answer in the excerpt. It could mean the click was unmeasured. It could mean the reader stopped. Keep those possibilities open until another observation distinguishes them.
Check the answer before commissioning a page
Start the installation-checklist row with the content inventory. Search titles, headings, document names, and maintained aliases. Read the candidate answer before judging whether it meets the task.
If a usable checklist exists, inspect its inclusion in the onsite search index. This means the site's search system, not Google's index. Record whether the current version is present and which fields the search service receives.
Next, compare terminology. Does the page call the task “initial configuration” while the searcher asks for installation? A verified alias may connect equivalent language. Confirm equivalence with the content owner before treating two terms as interchangeable.
If the answer is present and searchable, inspect its result title and excerpt. A technically matching result can still hide its purpose behind a generic document label. Propose a clearer label only after reading the destination.
If no adequate public answer exists, write a content brief around the missing task. State what the reader needs to do, what evidence will support the answer, and who can verify it. Use buyer-journey content mapping to place that answer without guessing the searcher's identity.
Assign the problem to the right owner
Give each finding a proposed owner and a completion condition. “Improve search” is too broad to verify. “Confirm the approved checklist appears for the permitted alias” names a bounded outcome.
A content owner checks the answer and terminology. A search owner checks indexing and retrieval. A product owner verifies versions and filters. An analytics owner explains missing fields. A navigation owner checks whether the preceding page exposes the expected route.
One person may hold several responsibilities. Keep the decisions separate so a wording change does not conceal an indexing failure.
For the fictional setup-guide row, retain the original version constraint during diagnosis. Removing it changes the question. If the only available guide describes another version, record that mismatch rather than declaring the search repaired.
For the warranty row, inspect the path before search. If the relevant product page omits an obvious warranty link, propose a navigation check. Four searches alone do not prove that navigation is the cause.
Close the row with a checkable decision
Write the conclusion as evidence, interpretation, action, and verification. Keep each part short enough to review without opening the entire analytics export.
A fictional closure reads: “The approved installation checklist exists. It is absent from the current search index. The search owner will review index inclusion. Recheck the original query after the approved change.”
That statement leaves terminology as an unresolved possibility until the recheck. It also avoids claiming that a content brief is necessary merely because the results count was zero.
Keep a no-change decision available. A phrase outside the site's scope may need an honest explanation rather than a new page. A measurement gap may require more evidence before any editorial change.
After an approved correction, compare the same task, constraints, and evidence fields. Record other releases that could affect the result. A lower empty-results count does not prove that buyers found the right answer.
Close the investigation when the evidence supports the decision. Leave the unknowns visible.
Sources
- Google Analytics: enhanced measurement events, including site-search collection and its parameters.
- Google Analytics: understanding PII, for the collection boundary.
- Google Search Console: Performance overview, for external search reporting.
- Google Search Console: query dimensions and limitations, for omitted and truncated query data.
- Algolia: analyzing site search data, for separating queries, filters, results, and clicks.