Check this page with an assistantOpens a chat asking it to summarise this article and name the evidence behind each claim.
Claude opens with the prompt on your clipboard: Anthropic does not support prefilled prompts on the web, and we would rather copy it than ship a button that drops it.
Search Console Is a Diagnostic Instrument, Not a Rank Tracker
The job of a weekly Search Console review is to answer one question: where did observed search behavior change enough to justify an action? If the meeting ends with a dashboard tour, the system has produced attention but not operations.
Google reports four core metrics: clicks, impressions, CTR, and average position. They are related, but they do not have equal diagnostic value. Impressions show that a result was eligible and seen under Google’s counting rules. Clicks show visits from those results. CTR is the ratio between them. Position is an average of the topmost result, not a fixed rank you can reproduce from one manual search.
Start with the outcome metric closest to the concern. If traffic fell, inspect clicks and impressions. If impressions held while clicks fell, investigate appearance and intent. If both fell on one page, inspect the queries, demand, technical state, and competitors. Average position becomes supporting evidence, not the headline.
Set Up a Baseline Once, Then Protect It
- Use the domain property when possible.
It brings protocols and subdomains into one property. Keep URL-prefix properties when a team needs a deliberately narrow operational view.
- Submit the canonical sitemap.
A sitemap is a discovery and canonicalization hint, not an indexing guarantee. List the URLs you actually want selected. The XML sitemap guide defines the clean URL and lastmod contract.
- Record the business events.
Keep launch dates, migrations, redesigns, large publishing batches, pricing changes, and outages beside the search data. Search Console does not know why your site changed.
- Define page groups.
Use URL patterns for service pages, product pages, comparisons, blog posts, tools, and locations. A site total can hide opposite movements in different page types.
Compare complete, like-for-like periods. A Monday-through-Sunday week should not be compared with five and a half preliminary days. For a seasonal business, year-over-year context can be more useful than the immediately previous period. Keep the comparison logic stable long enough to recognize a real change.
The 30-Minute Weekly Operating Routine
| Time | View | Decision |
|---|---|---|
| 0–3 min | Overview, Page indexing, manual actions, security | Is there an incident that overrides normal optimization? |
| 3–8 min | Performance: complete period vs previous period | Did clicks or impressions change beyond normal noise? |
| 8–15 min | Pages sorted by difference | Which URL group accounts for the movement? |
| 15–23 min | Queries filtered to the selected page | Demand, relevance, appearance, or competition? |
| 23–28 min | URL Inspection and live page | What is the smallest defensible intervention? |
| 28–30 min | Operations log | Owner, release date, metric, and check date |
Choose at most one or two actions from a weekly review. Search changes take time to be crawled, processed, and reflected; stacking several rewrites on the same page makes the result impossible to interpret. Google’s starter guidance notes that some changes can appear within hours while others take months, and recommends waiting at least a few weeks to assess many SEO changes.
Five Patterns Worth Acting On
1. Impressions rise before clicks
This is often an early discovery signal. Filter to the page, inspect its queries, and decide whether the page truly satisfies them. If it does, strengthen the exact sections and internal links that support those needs. If it does not, decide whether a separate page is warranted; do not stretch one URL across incompatible intent.
2. Impressions hold while CTR falls
The result remains visible, but fewer impressions become visits. Compare query mix, device mix, title promise, visible search features, and the opening answer. A falling blended CTR is not automatically a bad title: the page may have expanded into broader queries or more AI and rich-result contexts.
3. Clicks and impressions fall on one page
Compare the page’s query set between periods. A broad decline suggests a page or template issue; a decline concentrated in one query cluster suggests demand or relevance. Inspect indexing, canonical selection, last meaningful update, content accuracy, and internal links before rewriting.
4. Two URLs appear for the same query
This is not sufficient evidence of harmful cannibalization. Open the two pages and ask whether they serve different stages or intents. If their jobs are distinct, clarify each and link between them. If they duplicate the same job, consolidate the stronger material and align canonical and internal signals.
5. A top-three query begins to soften
Protect established demand before chasing new volume. Check whether facts, screenshots, prices, or examples aged; whether the destination still completes the user’s task; and whether site changes weakened its internal paths. “Defend” work is often cheaper and more certain than net-new production.
Read the Dimensions Without Lying to Yourself
Search Console aggregation changes with the view. Google explains that charts are aggregated by property, while page groupings count unique URLs. When several URLs from one property appear in the same result, property-level metrics credit the topmost result and count the property once; page-level data can count each URL.
- Queries: exact rows exclude anonymized searches and can be truncated. Chart totals include anonymized demand even when the query table cannot show it.
- Pages: performance is generally credited to the canonical URL, even when a duplicate URL received the click.
- Device and country: use them to test a specific hypothesis, not to create dozens of tiny segments with unstable data.
- Search appearance: compare like with like. Rich result contexts can have different counting and click behavior.
- Branded/non-branded: Google’s built-in filter uses data beginning in March 2025 and may be unavailable for small or sub-properties. Treat the classification as informative, not a ranking signal.
Discover is a separate, interest-driven surface with different volatility and reporting. If the property has enough data to expose that report, use the Google Discover field guide rather than blending feed traffic into Web Search decisions.
Use the broadest view that can answer the question. Add a filter only when it changes the decision you could make.
Map Every Signal to a Bounded Action
| Signal | Evidence to collect | Bounded action |
|---|---|---|
| High impressions, position 4–15 | Matching query, ranking page, competing intent | Improve one relevant section and add 2–3 editorial links |
| Top-ten, low relative CTR | Query mix, current result, title/H1/opening | Test a clearer promise without changing page scope |
| Previously strong page decays | Query losses, factual age, template changes | Refresh outdated evidence and restore missing utility |
| Demand has no suitable page | Repeated query pattern and business relevance | Create a brief; do not publish until the page has a distinct job |
| Excluded priority page | URL Inspection, response, directives, canonical | Fix the exact eligibility conflict before content work |
Avoid verbs that hide scope: “optimize,” “improve authority,” and “fix technical SEO.” The action should be reviewable in a diff. “Add a comparison table using current plan limits and link it from the two adjacent guides” is a unit of work; “improve topical authority” is not.
Measure the Intervention, Not Just the Site
Keep an operations log beside Search Console. For every shipped change, record the URL, release date, exact change, hypothesis, primary metric, guardrail metric, and earliest check date. Compare the selected page and query set with a sensible prior period, while keeping seasonality and larger site changes in view.
A title test might use clicks as the primary metric and impressions as the guardrail. A content expansion might use impressions for the intended query cluster as primary and conversion rate as guardrail. A consolidation might use clicks across both old URLs and the destination rather than judging the surviving URL alone.
Search Console does not isolate traffic from Google AI features into a separate performance type; Google includes those interactions in the Web search type. Pair Search Console with analytics referral and conversion data, and label any inference about AI visibility as an inference.
Questions About Operating Search Console
- What should I check in Google Search Console every week?
In Google Search Console, check indexing or security alerts first, compare complete periods for clicks and impressions, review the pages and queries with the largest meaningful change, and choose one intervention. A weekly review should create a decision, not a screenshot deck.
- Is average position the same as my Google ranking?
No. It is the average position of the topmost result for the selected property or page across the impressions included in the report. It varies by query, device, location, and search-result feature. Google recommends focusing more on click and impression trends than position alone.
- Why do the chart totals and table rows not add up?
The chart is aggregated by property while page views can be aggregated by URL, and anonymized queries are omitted from query tables for privacy. Filtering can also change totals because of data truncation. Treat the report as directional product data, not a complete server log.
- How much Search Console history is available?
The standard interface provides up to 16 months for most Search performance analysis. Bulk data export is the better route when a site needs longer retention, more complete rows, or warehouse-level analysis.
