Skip to main content
Visionnment Estimate value
Menu

Value & Improve

Keyword research for an existing website: turn Search Console queries into a content plan

Use Search Console queries, existing URLs, and search intent to choose what to refresh, what to write, and what to leave alone—with a worked decision sheet.

A search graph connecting query, page, and backlink nodes under a diagnostic lens.

For an existing website, start keyword research with the queries and pages you already have. Connect each promising reader task to its best existing URL, inspect what the page actually covers, then choose a refresh, a distinct new guide, a small link change, or a hold. A larger keyword list is useful only when it improves that decision.

The finished artifact is a short content plan with evidence and acceptance conditions. You need authorized Google Search Console access, an inventory of published and scheduled pages, and time to read the nearest articles in full. A spreadsheet is enough for a small site; a paid research subscription is optional.

This workflow uses Google's public documentation checked on November 4, 2026. It does not promise rankings or treat another site's estimated traffic as your opportunity. If your site has no useful query history yet, begin with the first useful content-cluster exercise and label demand assumptions explicitly.

1. Write down the decision and the reporting scope

Choose a manageable question, such as “Which two existing guides should we improve next?” Avoid combining a redesign, a migration and a full content expansion into one unexplained traffic diagnosis.

Record the Search Console property, export date, start and end dates, search type, country/device filters, and any query or page filters. Use completed days, and compare equal-length periods when interpreting movement. Add an older comparable seasonal period where available; equal duration alone does not remove seasonality or a business change.

As an editorial starting point, inspect both a shorter and a longer window—for example, 28 and 90 completed days. Those are working choices, not Google's thresholds for a valid opportunity. A low-volume site may still have too little evidence after 90 days.

Google documents that ordinary report dates use Pacific Time, while the 24-hour view uses local time. Some queries are omitted for privacy, query tables are truncated, and query filters exclude anonymized queries from totals. Preserve those limitations in your sheet instead of forcing visible query rows to explain every property impression. See Google's dimensions and data-grouping guidance.

2. Read the numbers before sorting by them

Use the report's metrics for their defined purpose:

Metric What it helps you see What it does not establish
Clicks Recorded search clicks in the selected scope Readers who completed your useful task
Impressions Recorded appearances under Google's counting rules Total market search volume
CTR Clicks divided by impressions A universal title-quality score
Average position Average topmost position for the grouping One fixed rank everyone sees

For a hypothetical row with 12 clicks and 600 impressions, CTR is 12 ÷ 600 = 2%. That calculation is sound; calling 2% “bad” without the query, result layout, device and page purpose is not. Google's Performance report overview explains the metrics, report setup and aggregation differences.

Keep unknown values distinct from true zeros. Google notes that unavailable values displayed as a dash or approximation can become zero in downloaded reports. Inspect the original view when a zero would change your decision.

Do not average several row CTR percentages to get a combined CTR. Calculate combined clicks divided by combined impressions within a consistent scope. Likewise, do not add page-grouped observations to property totals as if they were separate audiences.

3. Join the queries to their existing pages

In Search Console's search performance report, open Pages, select one relevant URL, then inspect Queries. Export that filtered view and write the page filter beside the file. Start with a few business-relevant pages so you can verify each relationship manually.

Reverse the investigation for an interesting query: select the query and inspect Pages. You are asking which URLs appeared for that observed query, not assuming the first page you noticed is its permanent owner. Google's performance-report walkthrough describes these query and page filters.

For each candidate, retain the query, observed URL, period, clicks, impressions, CTR and average position. Keep separate rows for different pages. Two unfiltered exports—one of queries and one of pages—do not supply a reliable query-to-page join by themselves.

An authorized analyst can automate this through the Search Analytics API, grouping by query and page. Record the request dimensions and filters. The API returns leading rows within its limits; it does not guarantee an exhaustive query log.

Before concluding that an unexpected URL is wrong, check attribution. Most search performance is credited to Google's selected canonical URL rather than a duplicate, as explained in its canonical data-attribution guidance. If a canonical or discovery problem is unresolved, use the new-site discovery checklist for the technical investigation before commissioning another article.

4. Group reader tasks and read the nearest owners

Add an “intended reader task” column in plain language: compare two providers, reconcile a settlement, restore a site, or diagnose a failed welcome email. Related wording can belong to one task. Similar words can also hide different tasks.

Read each nearest existing article from beginning to end. Include scheduled and held content in the comparison, since an unpublished but complete guide can already own the task. Record the exact useful sections, the missing step, and what a refresh must preserve.

Inspect a relevant live search result page for a few representative queries. Record the check date, location/device context where known, and prevalent page types. Are people being offered product pages, comparisons, tutorials or support answers? Treat this as an intent clue: Google explains that results vary with time, place, device and recent history in its report overview.

Do not copy a competitor's headings as your evidence. Identify what your reader still cannot do after reading your current page. If a broad guide already completes the task, a new keyword variation may need a clearer heading or contextual link rather than another URL.

5. Complete a worked query-to-action sheet

The following is an original hypothetical exercise for an imaginary website-operator publication. The observations and editorial findings are invented for teaching; they are not Visionnment results or search-volume estimates. Each count represents one query/page pair over the same 90-day Web-search window.

Query Observed page Clicks / impressions Full-body finding Decision
newsletter revenue reconciliation /newsletter-revenue/ 12 / 600 Reconciliation is the task, but refund timing is missing Refresh that owner
migrate newsletter suppression list /newsletter-tools/ 3 / 120 Comparison mentions exports; no import method exists elsewhere Research a distinct guide
newsletter revenue worksheet /newsletter-revenue/ 4 / 160 Worksheet exists but its download link is unclear Repair the link and label
unfamiliar vendor versus tool /seo-tools/ 0 / 1 Audience fit and useful evidence are absent Hold; do not create a versus page

The migration row has a 2.5% observed CTR: 3 ÷ 120. That figure does not establish that a new article will win more clicks. The new-guide decision depends on the missing reader task, source-backed method, editorial fit and absence of an existing or scheduled owner.

For the refresh row, preserve the URL, useful worked example and valid source material. Add the missing refund-period bridge and an acceptance check. Renaming the whole page around a nearby query would risk weakening the job it already does.

If two URLs appear for the same query, read both before declaring cannibalization. Different intents, result formats or canonical attribution can explain the observation. Consolidation needs its own evidence and redirect plan; a shared query alone is insufficient.

6. Turn the chosen row into a bounded brief

Choose an action because its expected reader benefit is clear and its evidence is available. Rank your shortlist by task importance, confidence in the identified gap, effort, maintenance burden and business relevance. Treat uncertain demand as a named uncertainty, rather than assigning invented volume or difficulty scores.

An accepted refresh brief should state:

  • The reader and task already owned by the URL.
  • The exact missing explanation, example or acceptance check.
  • The useful sections, identity and evidence to preserve.
  • Primary sources needed for changed factual claims.
  • The smallest edit that completes the task and its review condition.

A new-guide brief also needs a one-sentence difference from every nearest owner. For the hypothetical migration row: “This teaches a same-owner subscriber-data rehearsal; the existing comparison chooses a platform and the sale checklist transfers ownership.” If that difference disappears when you read the bodies, return the task to the existing owner.

Paid tools can help expand terms or inspect competitors after this step. Use the SE Ranking and Mangools comparison to assess that purchase separately. Their estimates should not replace the query evidence or the full-body editorial decision.

7. Check completion and learn from the change

Before approving the plan, have another reader trace one row from the saved filter through the existing body to the proposed action. They should be able to reproduce the arithmetic, identify what is observed versus hypothetical, and explain why a refresh or a new URL was chosen.

After a selected edit is released, record the actual date, changed sections and material concurrent events. Revisit the same query/page scope after a suitable observation period for the site's volume. A changed click count alone cannot separate the edit's effect from seasonality, result changes, demand or technical conditions.

If impressions are sparse, keep the task useful and the measurement claim modest. If the query is unrelated, hold it. If the page already answers it, improve discovery only where needed. If the site is technically unavailable, repair that prerequisite. The output of good existing-site keyword research can be two careful refreshes and no new article.