Keyword clustering and the content map: one page for each search intent

What keyword clustering is

Keyword research gives you queries. Writing needs pages. Clustering is the step in between: it decides how many pages you need and which page each query belongs to.

A cluster is a group of queries that one page can fully satisfy. espresso grinder, best grinder for espresso and grinder for espresso machine are three queries, but the searcher wants the same thing each time: to compare and buy a grinder that can handle espresso. One category page can serve all three. espresso grind size is a different job: the searcher wants to know how fine to grind, and a guide serves that better than a shop page.

So the output of clustering is not a list of keywords. It is a list of pages, each with its set of queries. That list, with a few extra columns, is your content map. The map then feeds everything that follows: briefs, internal links, the publishing plan and later the reports.

This lesson builds on the previous two in this module, "Keyword research: sources and method" and "Search intent and SERP analysis". Intent is what you cluster by, so if you are unsure how to read a search results page for intent, start there.

Why one page per intent

The rule is simple: one intent per page, many keywords per page.

Many keywords per page works because Google does not need an exact match. People describe the same need in different words, and a good page about the need can rank for most of them. You do not have to build a page for every wording, and you should not try.

One intent per page matters for two reasons. First, users: a page that tries to be a buying guide, a how-to and a product listing at once does none of these jobs well, and the visitor who wanted one of them has to dig. Second, your own signals: if two of your pages serve the same intent, internal links, external links and your editing effort are split between them. Google usually shows one of them for a given query, and not always the same one or the one you would choose. That problem has its own lesson, "Keyword cannibalisation". Clustering before you write is the cheapest way to prevent it.

One intent per page also stops the opposite mistake: a page for every keyword. Ten near-identical pages for ten wordings of the same question are thin, compete with each other and look like content made for search engines rather than for people.

What Google officially says

Google does not publish a clustering method. Its documentation does say how it treats different wordings, how it views site organisation and what it thinks of content produced only to catch queries.

Read together: write for the words people use, but do not build a page per wording, because Google matches meaning; keep each piece of content on one URL; and organise the site so the relations between pages are clear. Clustering and the content map are how you put that into practice.

Clustering methods

There are two families of methods. In practice most teams combine them.

SERP-overlap clustering

SERP-overlap clustering (also called SERP similarity clustering) uses Google's own results as the judge. If Google shows largely the same pages for two queries, it treats them as the same need, so one page of yours can compete for both.

1. Collect the top 10 organic URLs for every keyword, for the same country, language and device. Ignore ads and SERP features.
2. Compare the keywords in pairs and count how many URLs their top 10 lists share.
3. Set a threshold. Two keywords go into one cluster when they share at least that many URLs. A threshold of 3 or 4 shared URLs out of 10 is a common practical starting point. This is practitioner convention, not a Google rule.
4. Tune it on a sample. A lower threshold merges more and gives fewer, broader clusters. A higher one gives more, narrower clusters. Check a few dozen clusters by hand and adjust.

Two ways of applying the threshold give different clusters:

• Hard clustering: every keyword in a cluster must meet the threshold with every other keyword in it. Clusters are tight and reliable, but you get more of them, and some real variants end up alone.
• Soft clustering: each keyword only has to meet the threshold with the cluster's main keyword. Clusters are larger and more forgiving, but now and then a keyword drifts in that deserves its own page.

Soft clustering is the usual starting point for content planning; hard clustering is useful when the SERPs in your niche are close and you need to separate fine differences.

Semantic and manual grouping

Here you group by meaning instead of by results: shared modifiers (best, how to, vs, near me), shared word stems, embeddings from a language model, or simply a person sorting the list in a spreadsheet. It is the natural first pass on any keyword list, and the only option when you have no SERP data or the topic is so new that results are still unsettled.

Its weakness is that words are not intent. Grouping by words puts coffee grinder and how to clean a coffee grinder together because they share a phrase, and can keep espresso machine and espresso maker apart because they do not.

Which method to use

A practical hybrid: group semantically first, then check the SERP for the main keyword of each group and for every borderline keyword. Split a group when its SERPs disagree; merge two groups when their main keywords share enough results. Re-check the big clusters every few months, because results move.

Edge cases

Same words, different intent

coffee grinder shows shops and category pages; how to use a coffee grinder shows guides and videos. The words overlap, the intent does not, so they belong on different pages. The classic case is a word with two meanings, such as apple the fruit and the brand. The SERP settles it; the keyword text cannot.

Different words, same intent

espresso machine and espresso maker usually return the same pages, so they share one page. Pairs such as burr grinder and conical burr grinder may or may not merge: if the top results are the same, merge them; if the narrower query shows its own set of pages, give it a page or a clearly marked section. Plurals, word order and spelling variants almost always merge.

Local modifiers

coffee machine repair london is a local query. It deserves its own page only if you actually serve that place and the SERP is local (map pack, local businesses). A set of near-identical pages that differ only in the city name gives users nothing new and risks being treated as doorway pages. Make a location page when you have something specific to say about that location.

Brand modifiers

On a shop, delonghi espresso machine often deserves a brand category page, because the searcher wants that brand's range. Your own brand terms map to your home page or the relevant brand page. Queries about a competitor's brand rarely map to a page of yours; if you do not sell that brand, a page for it does not help the searcher.

Mixed-intent SERPs

Sometimes the top 10 mixes shops and guides. Look at which type dominates the top positions and build that page type. If the SERP is genuinely split, a page that serves both can work, for example a category page with a short buying guide above the products. Mixed SERPs are also where SERP-overlap scores are least stable, so check them by hand.

Year and freshness modifiers

best espresso grinder 2026 normally belongs to the evergreen best espresso grinder page, which you keep updated, rather than to a new page each year.

Small and zero-volume queries

SERP data for queries with almost no searches is noisy. Attach them to the closest cluster by meaning and answer them inside that page. Create a separate page only when the intent is clearly different and worth serving.

The content map

The content map is one row per cluster, which means one row per page. A plain spreadsheet is enough. These are the core columns:

• Cluster: a short name for the need, such as "Espresso grinders".
• Main keyword: the query that best describes the cluster, usually the one with the most demand. It guides the title and H1.
• Secondary keywords: the other queries in the cluster. They guide headings, sections and FAQs.
• Intent: informational, commercial, transactional, navigational or local.
• Page type: category, product, guide, comparison, service, location or tool. It follows from the intent and from what the SERP shows.
• URL: the existing URL, or the planned one for a new page.
• Status: keep, improve, merge, create or redirect.
• Priority: P1, P2 or P3, from the scoring described below.

Useful extra columns: total monthly demand of the cluster, current best position, the hub the page belongs to, owner and due date.

Two checks keep the map honest. No URL may appear in two rows: if it does, two clusters are fighting for one page, or one cluster is really two. And no row may list two URLs: if it does, you have found a merge.

Mapping clusters to existing URLs

On an existing site, most clusters already have a page that Google associates with them, even if nobody planned it. Keyword mapping starts from that evidence, not from a blank sheet.

1. Export Search Console data with page and query together (Performance report, or the API for larger sites) for the last few months.
2. Match the queries to your clusters and note which URL gets the impressions and clicks for each cluster.
3. One URL per cluster: it becomes the cluster's URL. Mark it keep if it matches the intent and performs, improve if it matches but underperforms.
4. Several URLs per cluster: decide whether they really serve the same intent. If they do, keep the stronger one, move the useful content from the others into it and redirect them (status merge). If the intents differ, split the cluster and give each page its own row.
5. A URL that matches no cluster: it either serves an intent you missed (add a row) or it has no search purpose. Pages for users only can stay; outdated pages can be updated, redirected or removed.
6. A cluster with no URL: status create.

For merges, Google's guidance on canonical URLs is direct: “Use permanent redirects to tell Google that a redirected URL is a worse version than the URL it redirects to.” and “Use this only when deprecating a duplicate page.” Update internal links to point straight at the URL you kept. The step-by-step handling of pages that already compete is covered in "Keyword cannibalisation".

Gap analysis

Once every cluster has a row, the gaps are visible. There are three kinds:

• Missing pages: clusters with status create. Your site has no page for a need your audience searches for.
• Wrong pages: clusters whose URL does not match the intent, such as a product page that collects impressions for a how-to query, or a blog post ranking for a query where the SERP shows only shops. The fix is a better-matched page, not more keywords on the wrong one.
• Competitor gaps: clusters your competitors rank for and you do not cover at all. Collect the pages that bring them traffic, cluster those queries the same way, and add the clusters that fit your audience and business to the map.

Filter the gaps through the question Google asks in its helpful content guidance: “Does your site have a primary purpose or focus?” A competitor's cluster belongs in your map when your audience needs it and you can write it with real knowledge, not just because it has search volume. Topic coverage as a whole, and how it builds trust in a subject, is the subject of the "Topical authority" lesson.

Hub-and-spoke organisation

Clusters are not isolated. Most of them belong to a few larger topics. The hub-and-spoke model turns that into site structure.

• The hub is a broad page on the topic, such as "Espresso at home". It gives the overview, answers the broad query, and links to every spoke.
• Spokes are the specific pages, one cluster each: grind size, dialling in, tamping, steaming milk. Each one links back to the hub and to the spokes a reader is likely to need next.
• Commercial pages related to the topic, such as the grinder category, are linked from the spokes where they help the reader, in context.

A few rules make the model work:

• The hub has its own intent. It targets the broad query, not every spoke's query. If the hub tries to answer grind size in full, it competes with the grind size page.
• Links are real <a href> links with descriptive anchors. Google's link guidance says “Good anchor text is descriptive, reasonably concise, and relevant to the page that it's on and to the page it links to.” Link to the grind size guide with words like "espresso grind size", not "read more".
• No orphans. Every spoke is linked from the hub at least, in line with Google's advice that every page you care about should have a link from another page on the site.
• Directories follow the topics where it makes sense. /espresso/ for the hub and /espresso/grind-size/ for a spoke gives readable URLs and breadcrumbs. Google notes that “using directories (or folders) to group similar topics can help Google learn how often the URLs in individual directories change.” Do not move existing pages just to fit a new folder plan: URL changes need redirects and carry risk, while Google says it will likely understand your pages as they are.
• There is no link quota. Google says “There's no magical ideal number of links a given page should contain.” Link where the reader benefits.

Prioritisation and a publishing plan

The map usually holds more work than you can do in a quarter. Score each row and work from the top.

• Demand: the total searches across the cluster, not just the main keyword.
• Business value: how directly the page can lead to a sale, lead or sign-up. A transactional cluster with modest demand often beats an informational one with high demand.
• Difficulty: how strong the pages in the current top 10 are, and whether your site can match their depth.
• Head start: where you already rank. Clusters where an existing page sits roughly in positions 5–20 are usually the quickest wins: the page is relevant, it needs to be better.

Combine them into P1, P2 and P3, then sequence the work:

1. Fix merges first. Two pages competing for one cluster hold each other back; consolidating is quick and makes later work count.
2. Improve P1 pages that already rank. Updating content that exists is usually faster than ranking a new URL. Google's own advice: “Check in on previously published content and update it as needed, or even delete it if it's not relevant anymore.”
3. Publish new hubs together with their first spokes, so the links between them exist from day one and no page launches as an orphan.
4. Add the remaining spokes in batches by hub, linking each new page from the hub and from related spokes as it goes live.
5. Review the map every month or quarter with Search Console data: new queries to assign, pages moving in or out of their cluster, URLs that start competing.

Write each page to answer its cluster fully, not to a length target. Google is explicit: “Are you writing to a particular word count because you've heard or read that Google has a preferred word count? (No, we don't.)” A good test for every spoke is another question from the same guidance: “After reading your content, will someone leave feeling they've learned enough about a topic to help achieve their goal?”

Common mistakes

• ✕One page per keywordTen pages for ten wordings of one need are thin and compete with each other. Group by intent.
• ✕One page for everythingA single page that tries to sell, compare and teach serves none of those intents well. Split by intent.
• ✕Clustering by words onlyShared words are not shared intent. Check the SERP for the main and the borderline keywords.
• ✕Ignoring existing URLsBuilding a new page for a cluster that an old page already ranks for creates a competitor on your own site.
• ✕The wrong SERPClustering on another country, language or device than the one you target gives clusters that do not match your market.
• ✕Hub that answers every spokeThe hub summarises and links; the detail lives on the spoke.
• ✕Orphan spokesA page published without links from its hub and related pages is hard for users and Google to find.
• ✕Keyword stuffing secondary termsSecondary keywords guide sections and headings; they are not a list to repeat. Google calls keyword stuffing “against Google's spam policies.”
• ✕A map that never changesResults, demand and your own pages change. A map you do not review stops describing your site.

FAQ

Sources

• Google Search Central — SEO Starter Guide
• Google Search Central — Creating helpful, reliable, people-first content
• Google Search Central — Link best practices for Google
• Google Search Central — How to specify a canonical URL

Quotes checked against the official pages on 30 October 2026.