Keyword Clustering by SERP Overlap, Mapped to Templates

We cluster on SERP overlap rather than on how similar the words look. Two queries belong on one URL when their top ten results share enough of the same URLs, and we work to a threshold of three to four shared results at the same device and locale, because that is Google stating its own opinion about whether the intents are the same. Text similarity puts sofa cum bed and sofa bed on a single page. The Indian mobile SERP separates them, and one page built for both loses to whoever built two. How the clustering runs: Live SERPs per query for the target locale and device, mobile first for India, then overlap computed pairwise and grouped. Seed queries come from your Search Console query export, from your internal site search, which is the most undervalued keyword source most catalogs already own, and from competitor coverage. They do not come from running an expansion tool until it stops producing rows. Volume enters only as a tiebreak between two clusters we could serve equally well. It is not the ranking criterion, because demand you cannot satisfy is a bounce with extra steps. The Hinglish layer: Roman-script Hindi is not a rounding error in Indian commercial search, and volume tools merge or split script variants inconsistently, so we pull the SERP for each variant separately rather than trusting one aggregated number. Transliterated brand and material names, city and pin-code modifiers, and English-Hindi code switching inside a single query all get checked against actual results, because the SERP for a Roman-script query frequently shows a different page type at the top than its English equivalent. Page type is the thing you then have to build. The deliverable is a URL plan: Every cluster is assigned to a template you already ship, category, filtered category, product, comparison or editorial, with the specific facet or attribute combination that serves it and the primary query it should win. Clusters mapping to no existing template get one of two verdicts: a template worth building, stated with the count of clusters that would justify the engineering, or no page at all. That second list is the half that pays for the engagement. Most catalogs we look at already carry more pages than their demand supports, and being told which URL patterns should never be generated is worth more than another two thousand targets. Supply checked against demand: For marketplaces we cross every cluster against live inventory counts per category before recommending anything. Ranking a page backed by four listings converts worse than not ranking at all, and it teaches Google something about the page that takes a year to undo. Where supply is thin we say the page should wait, which is an unpopular sentence in a deliverable and the reason we write it down. What we will not sell: Monthly keyword reports. Volume lists priced by the thousand. Rank tracking as a retainer, since you can run your own for a fraction of what we cost. Content writing: the plan goes to your writers or your existing agency with the intent, the page type and the queries attached, and we will review the first few pieces against the SERP evidence, but we are not a content shop and pretending otherwise would make us a worse one. Who should not buy this: Anyone who wants the list without the argument. The output here is a set of decisions about which pages should exist, and it is only useful to a team with authority to delete pages as well as create them. If the plan has to survive a committee that measures the search function by pages published, this will read as an attack on the department, and you should buy something else.