Damian Frost

Clustering by word similarity produces clusters that are linguistically tidy and commercially wrong, and the cost shows up months later as pages competing with each other. A comparison query wants a comparison page. A what-is query wants an explanation. A query with a brand name in it wants the brand's page.

Similarity metrics see one topic; intent separates three. That is why I start by classifying intent rather than by clustering, and cluster within a class. The classification is a judgement call and I write down the basis for it, because a classification you cannot audit is a classification you cannot correct. Volume is a separate question and worth being careful about.

Figures come from a source with its own methodology, they vary substantially between providers, and a low-volume long phrase can convert far better than a high-volume short one. I present a range and the source rather than a single number. Search volume for a phrase nobody types says nothing about demand for the topic. A niche can have demand expressed in phrasings that no volume tool reports, and that gap is exactly where a smaller site finds openings.

Once clusters exist, the decision is one page per cluster, and that is where most duplication starts. Two clusters with the same intent are one page, and the second cluster becomes a section on the first rather than a new address. I keep the classification note in the page so the reasoning survives being disagreed with. The classification is worth revisiting as the site grows, since a page created to serve one cluster often ends up serving three.