Keyword Clustering: How to Group Keywords by Search Intent and Map Each Group to One Page

Keyword clustering is the practice of grouping search queries that a single page can satisfy, so each group gets one URL. "Email deliverability", "email deliverability issues" and "why do my emails go to spam" differ in wording but may be answered by the same page, while "email deliverability software" asks for a comparison and probably needs another.

The reason to do it is page-level, not keyword-level. Google's SEO starter guide describes content that is organised around what people are looking for, and its guidance on helpful content asks whether a page gives a complete answer. Writing a thin page for every phrase variant works against both. Clustering is how you decide what the pages should be before you write any of them.

SERP-overlap clustering compares the results pages for two queries. If the top results share enough URLs, Google is treating the queries as one intent, and one page can target both. The threshold is a convention among practitioners, not a rule Google publishes; many use a count of shared top-ten results, and you should treat any specific cut-off as a starting point to test on your own topic. Its strength is that it uses the search engine's own behaviour as evidence. Its weakness is cost and noise: it needs a results-page lookup per keyword, and the results change.

Semantic clustering represents each keyword as a numeric vector with a language model and groups the ones that sit close together. It is fast and cheap and works well on thousands of phrases, but closeness in meaning is not the same as shared intent. "Best CRM for small business" and "CRM definition" are related topics with different intents. We recommend semantic grouping to split a long list into rough topics, then SERP checks on the head term of each group to confirm intent. For a list under a few hundred, you may skip the semantic step and do it by hand.

Build one sheet with a row per keyword and these columns: keyword, monthly search volume from your tool, current ranking URL if any, intent label, cluster name, target URL, and notes. Start by lowercasing and deduplicating, and merge plurals and word-order variants, since they almost always share intent. Then sort alphabetically so near-duplicates sit together, and assign a provisional cluster name by reading down the list.

Next, check intent against the live results. For the likely head keyword of each cluster, open the results and note what kind of page ranks: guides, product pages, comparisons, tools, forums. If a cluster's members pull different page types, split it. Illustrative example: a cluster containing "marketing automation workflows", "marketing automation workflow examples" and "marketing automation software" splits into two, because the first two ask for patterns and the third asks for a vendor comparison. Finish by naming the head term (the clearest, most natural phrase for the intent) and putting it in the cluster name column.

Expect to disagree with your first pass. The usual surprise is a cluster that looked tidy by wording but is really two audiences: "keyword clustering tool" wants software, while "how to cluster keywords" wants a method, and a single page cannot serve both without doing neither well. Resolve disputes by what the results page rewards, and write down the reason in the notes column so the decision survives staff changes. Where the live results are themselves mixed, with guides and tools side by side, treat the cluster as a candidate for one page with a clearly separate section, and check again after a few months.

Give every cluster exactly one target URL, and give no URL more than one primary cluster. For an existing site, start with what ranks today: if a page already ranks for part of a cluster, that is the natural home, and the job is to extend it. If nothing on the site matches, the map produces a brief for a new page: head term, supporting phrases to cover in headings and body, the intent, and the page type.

Weigh the page type to the intent rather than the volume. A cluster with comparison intent belongs on a comparison page; a how-to cluster belongs on a guide. Then connect the pages: related clusters link to each other with descriptive anchor text, and a parent topic page gathers its children. The map becomes your editorial calendar, and its gaps (clusters with no page) are the content you owe.

Cannibalisation is when two or more of your pages target the same intent, so they compete and neither ranks as well as one consolidated page would. It is not a penalty. You find it by searching your Search Console performance data for a query and looking at which pages receive impressions for it: if several URLs swap places over time, the pages overlap.

There are three fixes, and the choice depends on whether the weaker page has any value of its own. Merge the two into one stronger page and redirect the old URL; Google's documentation on redirects describes permanent redirects for pages that have moved. Re-aim one page toward a distinct intent and rewrite its title and headings accordingly. Or, where both pages must exist for users, designate one as canonical; Google's guide to consolidating duplicate URLs explains the options, though canonicals suit near-duplicates better than pages that merely compete. Never delete a page that holds links without redirecting it.

Review the map when you add a service or topic, and at least twice a year otherwise. Search results change as new pages publish and as Google reinterprets queries, so a cluster that was one intent last year may have split. Recheck the live results for each head term and look at the Search Console queries driving each page: queries you did not plan for are a prompt to extend the page or give them their own.

A refresh does not mean re-clustering from scratch. Keep the existing map, add new keywords as rows, and ask of each one whether it fits an existing cluster, extends one, or deserves its own. Illustrative example: a new batch of forty keywords usually resolves into a dozen that join existing clusters, a handful that start new ones, and the rest that you decide not to pursue. Date every change in the sheet so the reasoning survives a change of owner.

Who this is not for: a site with ten pages and a narrow offer does not need a clustering project. A short list, read by a person who knows the business, will do the job. Clustering pays back when the keyword list is larger than anyone can hold in their head, or when a site has already published overlapping pages and nobody can say which one is meant to rank.