Keyword Cannibalization Analysis

Find queries where several pages of one property compete against each other, decide which page should win each cluster, and recommend concrete fixes. All data comes from the GSC Wizard MCP tools.

Optional user input: the user's request may contain a property (site URL or domain), a date range, or a focus hint such as a specific query or URL path. Honor it when present.

Workflow

1. Resolve the property.

   • If the user's request or the conversation names a property, use it. Site URLs use GSC formats: sc-domain:example.com or https://www.example.com/.
   • Otherwise call list_sites and pick the obvious match. If several plausible properties exist, ask the user which one before continuing.

2. Set the date range.

   • Default: last 28 days (the tools default to this when you omit dates). Use the user's range if they gave one.
   • GSC data lags about 2 to 3 days; never treat the most recent 2 days as complete. If a dip is visible at the edge of the range, say it is likely reporting lag, not a real drop.

3. Find cannibalized clusters. Call analyze_cannibalization with the siteUrl (plus startDate/endDate if the user specified a range). Defaults are fine to start: minImpressions: 10, limit: 100. Raise minImpressions (for example to 100) on large properties to cut noise; if the user gave a focus query or path, filter the returned clusters to it.

   • Each result is a query with 2 or more competing pages, scored by impression entropy: higher entropy means impressions are split more evenly across pages, which is worse cannibalization. Each contender already carries clicks, impressions, ctr, and position.
   • Present the top clusters (typically 5 to 10) as a compact markdown table: query, number of pages, total clicks, total impressions, entropy. If nothing clears the threshold, say the property shows no meaningful cannibalization in this range and stop.