Search Console keyword cannibalization (Growth+)
Changed onQueries with at least min_impressions impressions for which two or more pages each take at least min_share (default 0.1) of the query's impressions, most impressions first, each with its competing pages (share, impressions, clicks, ctr, position). Use it to find the keywords several of the site's pages fight over, and decide which pages to merge, redirect or differentiate. Google counts page-by-query impressions per page, so a query's impressions here can exceed its property-wide figure. Reads Google's top 25,000 page-query rows per property (truncated: true when reached). With output=flat or csv, each competing page is its own row, with its URL in page_url. Requires a connected Search Console property (Growth+). Live reads only: under stored reads it returns ERR_INVALID_PARAM. Live reads return ERR_SEARCH_CONSOLE_ACCESS_REVOKED (403) for revoked Google access; reconnect the property in Preferences > Project Settings > Data Connections. They return ERR_SEARCH_CONSOLE_UPSTREAM (503) when Google Search Console is unavailable or over quota; wait for the number of seconds in Retry-After before retrying.
Request
- Base URLhttps://api.llmpulse.ai/api/v1
- URLhttps://api.llmpulse.ai/api/v1/search_console/cannibalization
- AuthAPI key in header Authorization
Query parameters
Project ID
Which Search Console properties to read; /search_console/properties lists them. Omit it for the primary property, or pass one property id (e.g. sc-domain:example.com) to read that one; both answer with property. A comma-separated list of ids, or all, combines several and answers with properties, excluded_properties and property_errors instead: excluded_properties are left out because another selected property already counts them (a URL-prefix property under a domain property), so no click is counted twice; property_errors are properties Google refused, left out of the figures and also named in the X-Search-Console-Property-Errors header. When Google refuses every property, the request fails as a single-property read would.
Number of days to look back, today included (alternative to from/to). At most 487, Google's retention.
Start date (YYYY-MM-DD). Defaults to 30 days ago.
End date (YYYY-MM-DD), inclusive. Defaults to today. Google publishes data with a 2-3 day lag, so data_through in the response says how far it reaches.
Smallest impressions a query needs.
Smallest fraction (0.01 to 0.5) of the query's impressions a page needs to count as competing.
Rectangular output for BI tools (Tableau, Excel, Sheets, ELT). Omit for the default nested JSON. 'flat' returns the same metadata plus 'columns' and 'rows'; 'csv' returns those rows as text/csv. Errors are always returned as JSON.
Which search surface to measure. Defaults to web. discover and googleNews carry no query dimension, so Google rejects /search_console/queries for them.
Narrow the query; every entry must match (AND). Send the whole list as one JSON value: filters=[{"dimension":"page","operator":"contains","expression":"/blog/"}] (URL-encoded). An array of objects has no query-parameter form a generated client can produce, so the string is what the official SDKs send; see the SearchConsoleFilters schema for the shape it encodes. includingRegex and excludingRegex take RE2 syntax. At most 10 entries, expression at most 500 characters. The bracket form filters[][dimension]=page&filters[][operator]=contains&filters[][expression]=/blog/ is also accepted.
final (default) counts only rows Google has finalized. all also counts the most recent days, which are still being filled in and will change.
How Google counts impressions. auto (default) is what Search Console shows (by page when page is grouped or filtered, else by property). Google rejects byProperty with a page dimension or filter.
Response
Competing pages per query
Changes
- info
endpoint added
- info