Google Search Console CSV Analyzer

How to use it

1. Export the current Pages table for the date range and filters you want to analyze.
2. Export a current combined page-by-query table for the same scope.
3. Paste current Pages first, add a line containing only ---, then paste current page-by-query rows.
4. For period comparisons, add another separator and a prior-period Pages export with comparable filters. For the update overlay, add one more separator and a GSC Dates export. You may supply Dates without a prior period.
5. Keep or adjust the experimental automated-query thresholds. The defaults flag normalized queries with at least 10 words, at least two impressions, and zero clicks.
6. Select Analyze. Start with the period and content-group reports, opportunities, and cannibalization, then review the query heuristic, topics, CTR curve, and time-chart context.

Example result Illustrative data

This compact static report uses fictional rows. It demonstrates the output fields without claiming typical CTR or traffic.

What the results mean

• Pivot pages is the count of normalized URLs represented in the page-by-query table.
• Winners and losers compare current-minus-prior clicks and impressions by normalized page URL. A click decline with less than one position of movement is labeled stable-ranking decay.
• Template-like URL segments aggregate pages with the shared deterministic URL-pattern registry, such as product detail, article/post, documentation, or a transparent fallback pattern. They are inferred cohorts, not proof of an implementation template.
• Topics groups queries with shared word stems so the grouping can be inspected rather than treated as an opaque semantic label.
• Curve is a position-level CTR model derived from eligible supplied rows; a sparse export may return little or no curve.
• Possible automated-query pattern aggregates the same normalized query across pages, then flags long queries with enough impressions and no clicks. It is an experimental anomaly heuristic, not evidence that a bot made the searches.
• Cannibalization lists queries associated with multiple URLs. Alternation can be intentional, so inspect intent and dates.
• Opportunities ranks pages using modeled click upside plus CTR-gap signals; reasons show which query triggered each score.

How it works

The workbench normalizes page URLs, compares current and prior page metrics, aggregates pages by shared template-like URL patterns, builds a page-to-query pivot, clusters query wording by stem overlap, and derives a CTR curve from queries with at least 10 impressions and average positions 1–20. It separately groups identical normalized queries across URLs for cannibalization and the experimental anomaly heuristic, preventing a multi-page query from being counted as several candidates. Finally, it feeds page totals and query rows into the shared opportunity scorer, which looks for positions 4–20 and CTR below its expected curve.

Features

• Eight related GSC reports from the same two required exports.
• Optional prior-period winners/losers, stable-ranking decay, and inferred URL-cohort deltas with CSV export.
• Experimental long-query, impression, and zero-click anomaly detection with configurable thresholds and full candidate CSV export.
• Optional clicks timeline with confirmed Google update rollout ranges shaded onto the chart.
• Visible query evidence behind page opportunities and cannibalization candidates.
• Local parsing, explicit header checks, and normalized URL joins.
• Optional read-only browser-session connection that loads standard Pages, page×query, prior Pages, and Dates tables without saving a Google credential.

Limitations

• Brand terms are excluded from CTR-curve construction only when you supply them; review the resulting curve before using it as a benchmark.
• Search Console exports and API responses can omit low-volume or anonymized queries and aggregate data differently across dimensions.
• The automated-query heuristic cannot identify bots. Legitimate conversational searches, zero-click behavior, and incomplete top-row exports can all affect the result.
• The session shortcut uses fixed Pacific-time periods and up to 10,000 top rows per table; use uploads for custom filters or preserved exports.
• Algorithm-update overlap is timing context, not evidence that an update caused a traffic change.
• Average position, click-upside estimates, and stem clusters are directional models, not forecasts or causal findings.
• All uploaded tables must use comparable date, country, device, and search-type filters.

Frequently asked questions