Keyword clustering for large sites: methods, tools and page mapping (2026)

Keyword clustering means grouping keywords that Google answers with the same pages, so one page on your site can rank for all of them. The dependable method is SERP clustering. You pull the top 10 Google results for every keyword on your list, then group the keywords that share three or four ranking URLs. Semantic clustering groups by meaning instead. It's faster and cheaper, but it merges searches Google keeps apart, and on a large site that turns into pages fighting each other.

Each cluster then gets one URL, ideally one you already own. Here at Searchmaxxed, that's the work we do for law firms, clinics and premium builders. Then we make each page the answer in Google and AI search. That means ChatGPT, Gemini, Perplexity and AI Overviews too.

What is keyword clustering?

Keyword clustering is the step between keyword research and content planning, where a long list of searches becomes groups that one page can win. Out comes a page plan. It tells you how many pages you need and which page owns which search.

The logic comes straight from Google. When Google ranks the same pages for two different searches, it's telling you both searchers want the same thing. So "ztna" and "zero trust network access" belong on one page. "ZTNA vs VPN" needs its own.

People mix up keyword clusters and topic clusters all the time, yet a keyword cluster is just the set of searches one page targets. A topic cluster is a set of pages linked around one subject, the hub-and-spoke structure behind topical authority. You cluster keywords first. Those clusters then tell you which pages the topic cluster holds, and which pages it can do without.

SERP clustering vs semantic clustering: which should you use?

Use SERP clustering for every decision about pages, and semantic clustering only to clean and pre-sort a raw list. SERP clustering copies Google's view of intent. Semantic clustering copies a language model's view of meaning, and the two disagree more often than most content teams expect.

SERP clustering, also called SERP-based or SERP-overlap clustering, compares the top 10 results for each keyword. Enough shared URLs, and two keywords join one cluster. Semrush's own explainer from May 2026 says its Keyword Strategy Builder works this way: "not word similarity, but actual search engine results."

Semantic clustering uses natural language processing (NLP). Most tools turn each keyword into an embedding, a list of numbers that captures its meaning. Keywords whose numbers sit close together share a group. It needs no search data, so it costs little and runs in minutes.

The catch is that meaning and intent are different things. In Keyword Insights' clustering guide (updated July 2025), "vaporizer parts" and "vaporizer accessories" look like synonyms to an NLP tool. Their search results were only 11.8% similar. That's two pages. A semantic tool would have planned one.

How many shared URLs should put two keywords in one cluster?

Three or four shared URLs in the top 10 is the usual line, and it's where we start. Raise the number and you get more, smaller clusters, while lowering it gives you fewer, bigger ones that hide real differences in intent.

The tools publish their defaults. KeyClusters groups keywords with 3 or more of the same pages in their top 10. Keyword Insights defaults to 4 shared URLs from the top 7, according to its clustering docs, and lets you change it. SE Ranking lets you pick anywhere from 1 to 9, as its keyword clustering guide explains.

The second setting matters just as much: how strict the comparison is, which decides whether every keyword in a cluster must match every other.

• Soft clustering, which Keyword Insights calls centroid, compares every keyword with the biggest-volume keyword in the group. Big clusters, fast.
• Hard clustering, also called agglomerative, compares every keyword with every other keyword in the list. Clusters come out smaller and tighter, and the run takes longer.

Wikipedia's keyword clustering entry adds a middle "moderate" type, though most tools offer only the two above. Here's how we set them.

On money pages, a wrong merge costs more than a wrong split. Split pages are easy to merge later. One page trying to serve two intents, though, usually wins neither search, and you won't see why until you check the results.

What does keyword clustering look like on a real B2B keyword set?

Messier than the tool demos, and far more useful: ten zero-trust searches that look like one topic turn out to need five pages. On 1 November 2026 we pulled the US top 10 for each one. Then we counted the URLs each pair shared. A semantic tool would likely put all ten in one group, since they're all about zero trust.

That gives five pages:

1. A ZTNA explainer for "zero trust network access", "ztna" and "what is ztna".
2. A ZTNA buyer's page for "ztna solutions", "best ztna solutions", "zero trust network access software" and, most likely, "ztna vendors".
3. A ZTNA vs VPN comparison.
4. A SASE vs ZTNA comparison.
5. A zero trust security page, which shared no URLs at all with any of the nine ZTNA searches.

Three things stand out. First, "zero trust network access software" straddled two clusters, sharing 4 or 5 URLs with members of both. Its results mixed vendor definitions from Zscaler, Palo Alto Networks and Cloudflare with "best tools" lists and a Reddit thread. When that happens, check which page types rank. Then send the keyword to whichever page sits closest to a purchase.

Second, "ztna vendors" is a threshold call. Soft clustering keeps it with "best ztna solutions", since they share 4 URLs. Hard clustering splits it off, because it shares only 2 with "ztna solutions". Both results pages leaned on provider shortlists. We'd keep one page.

Third, the broad head term had no overlap with anything. "Zero trust security" is its own subject. Folding it into the ZTNA pages would have weakened every one of them, and handed the head term to a competitor with a focused page.

Which keyword clustering tool should you use?

If you already pay for Semrush or Ahrefs, start there. Big lists that need full control over thresholds call for a dedicated SERP clustering tool such as Keyword Insights or SE Ranking. For a quick check of a few hundred keywords, ContentGecko's free version is enough.

Prices are as each vendor's site listed them on 1 November 2026.

Ahrefs works differently. Its Parent Topic takes the number-one page for a keyword and finds the query that sends that page the most traffic. Keywords with the same Parent Topic land in one cluster. That's one URL of evidence rather than ten. Treat it as a fast sort, and check money clusters with a full overlap run.

Got engineers? A results API and a short script do the same job at any scale, and you keep the thresholds, the raw data and the history.

How do you cluster keywords, step by step?

Build one clean list, cluster it by search results in one market at a time, then map every cluster to a single URL. On a large site, that runs to eight steps.

1. Build one master list. Export from Semrush or Ahrefs, then add your Search Console queries and the words prospects use on sales calls, which most lists miss. Our B2B keyword research guide shows how to find them.
2. Clean it. Strip out duplicates and off-market terms. A quick semantic pass lets you delete whole junk groups at once, instead of filtering row by row.
3. Pick one market. Results differ between countries and languages, so run each one on its own.
4. Run the SERP clustering. Soft at 3 shared URLs sizes the market.
5. Tighten the money clusters. Rerun anything near a sale, such as product, comparison, pricing and alternatives searches, with hard clustering at 4 shared URLs.
6. Label each cluster. Note the intent and the page type that ranks: vendor page, shortlist, glossary entry or comparison.
7. Map each cluster to one URL, using Search Console to find the page that already owns it. The next section shows how.
8. Rank and brief, putting revenue clusters first. Each brief carries its whole cluster: the main keyword in the title and H1, the rest in the headings.

How do you map clusters to pages without cannibalising your own site?

Give each cluster exactly one URL, check Search Console before you create anything, and send every new page request through the same map. Most of the damage on big sites comes from pages that nobody ever mapped to a cluster, published by teams that never saw the plan.

For each cluster, open Search Console's Performance report, filter by the cluster's main queries and open the Pages tab. What you see there decides the move.

Already split? Our keyword cannibalisation guide walks through choosing the owner page and fixing the rest without throwing away traffic you've earned.

Skip this step and every variation gets its own page, which leaves Google choosing between near-identical pages for the same search, again and again. Enterprise sites run the same risk with templated product, industry and use-case pages. A single clustering run before launch shows which variations deserve their own page and which belong together on one.

How does keyword clustering work at enterprise scale?

At tens of thousands of keywords, cluster in two passes and run every market separately. Then treat the keyword-to-URL map as a live record with one owner, because the method stays the same while the discipline around it changes.

Run two passes. Keyword Insights' docs suggest putting 200,000-plus keywords through the centroid algorithm first, to find the broad topics you aren't covering. The agglomerative algorithm then breaks those topics into pages, so you pay for tight clustering only where it counts.

Price the results data. Every keyword needs its own results pull. Keyword Insights charges one credit per keyword on every plan, so a semantic pre-pass that deletes junk first cuts the bill.

Cluster every market on its own. Different sites rank in the US, the UK and Australia, so a cluster that holds in one country can split in the next. Compare the maps side by side.

Recheck money clusters every quarter. A cluster that held in January can split by June, because competitors publish and Google keeps testing its results pages. So rerun commercial clusters on a schedule, plus any whose page slips.

Give the map one owner. On a big site, product marketing, docs, the blog and regional teams all publish. Cannibalisation usually starts when one team briefs a page that another team already owns. Make the map the first stop for every brief. Our SEO content strategy work starts there too.

Does keyword clustering still matter for AI search?

Yes, arguably more. Google's AI features answer one question by running several related searches, so a page that covers a whole cluster gets more chances to be picked. Clustering also keeps you clear of the spam policy that catches sites publishing a page for every variation.

Google calls this query fan-out. Its guide to generative AI features (updated 10 July 2026) defines it as "a set of concurrent, related queries generated by the model". Those related queries are your cluster. A page with a clear section for each one can serve several at once.

The same guide warns against the opposite. Separate content for every variation or fan-out query, made "primarily to manipulate rankings or generative AI responses," breaks Google's scaled content abuse policy. One strong page per cluster is both the safe shape and the one that wins.

So write each question in the cluster as a heading. Answer it in the first two sentences. That way ChatGPT, Gemini, Perplexity or an AI Overview can lift the answer whole and name you as the source. Our guide to showing up in AI search covers the rest, from crawler access to citations on other sites.

Is Searchmaxxed right for you?

We're a strong fit if you run a large site for a law firm, clinic or premium builder and your pages already overlap. Several teams publish, and nobody owns the map. You want those clusters turned into pages that rank, get named in AI answers and bring in pipeline.

We give every search one clear page. We find the searches you're missing, build the pages that should exist and fix the ones competing with each other. We make those pages the answer in Google and in the AI engines your buyers ask, from ChatGPT and Perplexity to Gemini and AI Overviews. We track which searches and AI answers turn into revenue. You talk directly to the people doing the work, and you own your site, content and data from day one.

One client came to us with a website that wasn't working. Three months later he was on page one for one of the most competitive searches in his industry. Now he has more high-intent leads than he can handle, and we're building the systems to handle them.

The bigger picture:

Searchmaxxed's work has generated more than $15 million in revenue through search for its clients.

"I can't believe this worked." That's what we hear. Clients tell us they finally found a growth partner in search that keeps its word and gets things done.

Only need the clusters? A tool from the table above will cost you less. Pick Searchmaxxed when you want every search that matters to land on one page that wins it, so buyers find you before a competitor.

What should you do next?

Export your top non-brand queries from Search Console and run them through a SERP clustering tool at hard 4. Then count the clusters where two or more of your URLs already earn impressions. That count is your cannibalisation backlog, and it tells you which pages to fix before you brief a single new one.

Want us to do it with you? Get your free search analysis. We'll benchmark you against your top competitors across Google, ChatGPT and the other AI engines. We'll show you where they're named instead of you, and the few moves that will matter most.

Frequently asked questions