Keyword Clustering From SERP Overlap, Not Volume

Two queries belong on the same page when their top ten results share at least three URLs. That threshold decides every grouping we hand over, and it disagrees with a volume-sorted spreadsheet often enough to matter: French pairs that read as synonyms frequently return two different result sets, and pairs that look unrelated return one. Where the numbers come from: Live SERPs pulled per query from google.fr and google.com through DataForSEO, stored with the date attached so a re-run six months later reads as a comparison rather than a replacement. Search Console across the full sixteen-month window at page and query dimension, joined to the URLs you already hold. Ahrefs and Semrush volumes travel through as a sanity column and nothing more; they are modelled from panels that thin out badly in French below the head, and we have watched a query marked seventy a month carry a six-figure pipeline in developer tools. What comes back: A cluster file in CSV and Google Sheets. Every query carries its cluster id, the overlap score that placed it there, the URL you currently rank with, and one of four verdicts: existing page, new page, merge into, or do not pursue. A written note explaining the do-not-pursue calls, which are usually one of two things, either the result set is held by a page type you cannot honestly produce, or the query resolves inside your product rather than in an article. Then a cannibalisation list: every cluster where two of your own URLs appear in the same top ten. Developer-tools clusters: A library name plus an error string almost always returns a documentation page, a GitHub issue and a Stack Overflow thread. That cluster belongs to your docs, not your blog, and we will put that in writing even when the blog is the thing you were planning to fund. Splitting documentation and marketing into separate cluster sets is the first move on any developer-tools account, because merging them produces a map that recommends articles nobody will ever rank. French-language handling: Accents, elisions and the qu' contraction are normalised before matching, and singular and plural are treated as distinct until the SERP says otherwise, which in home and furniture they often are not. Queries that pull Leroy Merlin, Castorama and a map pack are flagged as retail-local and removed from editorial clusters rather than written about. What this is not: Not a content calendar and not a set of briefs. Those are separate work and you should buy them after you have seen the map, not before. No page templates, no writing, no internal link implementation. We do not deliver difficulty scores, which are a competitor metric dressed as a plan and would amount to re-selling you a tool subscription you already have. Not a fit if: You want the output sorted by volume descending. You need every query in the account touched rather than the ones that resolve into pages. Your site cannot ship a new URL inside a quarter. It is also a poor fit when the answer has already been decided internally and the map is meant to ratify it, because we will produce a map that disagrees, and that disagreement is the thing you are buying from an outside party. Who does the work: One consultant, named in the proposal, from the first SERP pull through to handover. No research team behind them and no step in the middle you cannot see. Handover is ninety minutes, recorded, with your channel lead and whoever owns the CMS in the room, because half the verdicts turn on what your CMS can actually do.