SuperAPI AI Directory

decaying

Agents & models

View repository

Updated 10-16 08:48

About this skill

Decaying queries and pages

Use this report to identify retained query/page combinations that lost clicks between adjacent equal-length periods or year-over-year windows. It requires evidence in both windows, excludes URL shifts, and groups returned rows by diagnosis signal and page template. This avoids turning a row missing from GSC's retained query data into zero traffic. The report shows observed movement, not why it happened.

Run the report

Use the compact MCP flow:

  1. Call seo_list_reports with {"category":"opportunities"}.
  2. Call seo_describe_report with {"id":"decaying"}.
  3. Call seo_run_report with {"id":"decaying","params":{"site":"sc-domain:example.com","days":28,"comparison":"previous-period","limit":20,"minDropPct":20,"minPreviousClicks":10,"minClickLoss":5}}.

The CLI uses the same report definition:

seo reports describe decaying --json
seo reports run decaying --params '{"site":"sc-domain:example.com","days":28,"comparison":"previous-period","limit":20,"minDropPct":20,"minPreviousClicks":10,"minClickLoss":5}' --json

site is required. Choose year-over-year only when the requested window fits GSC's available history. Brand terms and includeBrand control navigational-query filtering.

Interpret and act

Read dataStatus, ranges, and source.completeness first. Check current and previous fetched rows, caps, truncation, invalid rows, excluded URL shifts, and output limits in selection. summary.observedRetainedQueryClickLoss covers eligible retained rows; returnedObservedRetainedQueryClickLoss covers only the bounded output. Each item preserves current and previous clicks, impressions, CTR, position, clickLoss, dropPct, signals, and an evidence scope of retained-query-page-row. A diagnosis such as lost_position or lost_ctr labels correlated movement, not a cause.

Safe actions include checking deployments and annotations, live indexability, SERP composition, demand seasonality, and page-template patterns before editing. Use groups to find repeated technical or content hypotheses, then verify individual URLs. Do not sum capped rows into a property-wide loss, treat missing retained rows as zero, or claim that a content change, ranking update, or technical issue caused the decline without separate evidence.