Keyword clustering is the practice of grouping search queries that share the same intent so a single page can target all of them. Unlike keyword-by-keyword targeting, which produces one thin page per phrase, clustering concentrates authority on fewer, stronger pages that rank for the whole group.
According to Ahrefs’ study of 3 million searches, the average page ranking #1 also ranks in the top 10 for nearly 1,000 other keywords — a median of around 400. That single number is the entire argument for clustering: Google already treats your page as an answer to hundreds of queries, so planning one page per keyword fights the way ranking actually works.
In a 2026 comparison of 16 clustering tools, SERP-based methods scored 70–95 out of 100 on grouping accuracy, against 9–35 for pattern-matching and rule-based tools. SEO agency Victorious found that consolidating cannibalised pages produced an average 37% traffic increase to the surviving page.
What Keyword Clustering Actually Solves
Most teams arrive at clustering through pain, not theory. They have 300 blog posts, traffic has plateaued, and several pages compete for near-identical queries. Clustering is the repair.
The underlying problem is that keyword tools output phrases, and phrases feel like content briefs. Export 4,000 keywords from Ahrefs and the instinct is to build 4,000 pages. What you get instead is keyword cannibalisation — your own URLs splitting link equity and confusing Google about which one deserves to rank.
Clustering inverts the unit of planning. You stop planning against keywords and start planning against SERPs. If two queries return substantially the same ten results, Google has already decided they are one topic. Building two pages is arguing with a decision that has been made.
In my experience auditing content programmes, over half the pages on a mature site are competing with a sibling page nobody realised existed. The fix is rarely more content. It is consolidation, redirection, and a cluster map that stops the problem recurring.
Why SERP Overlap Beats Semantic Keyword Clustering
There are two ways to group keywords, and they are not equivalent.
Semantic clustering groups by meaning. It uses NLP or embeddings to decide that “seo audit” and “seo audit tool” are similar, because linguistically they are. SERP-based clustering ignores meaning entirely and asks a different question: do these two queries return the same URLs?
Run those two examples through Google and the SERPs diverge sharply. “seo audit” returns explainers and consultancy pages. “seo audit tool” returns software. Same words, different intent, different page. A semantic tool merges them and you write one page that serves neither.
This is why the accuracy gap in the 2026 tool comparison is so wide. Semantic similarity is a proxy for intent. SERP overlap is a direct measurement of it — you are reading Google’s own clustering, already computed and published for free on every results page.
The practical rule: use SERP overlap as the primary signal and semantic similarity only as a tiebreaker for low-volume keywords where SERP data is thin or unstable.
How to Choose Your Overlap Threshold
The threshold is the number of shared URLs in the top 10 required to merge two keywords. Nobody documents how to pick it, so most teams accept a tool default and inherit its bias.
Three settings are in common use:
- Soft (1–2 shared URLs). Casts the widest net. Produces large clusters and frequent false merges. Almost always wrong.
- Moderate (3–4 shared URLs). The practical default, equivalent to 30–40% overlap. Balances coverage against precision.
- Hard (5+ shared URLs). Strict. Produces small, high-confidence clusters and more total pages.
Pick based on your authority, not your preference:
New or low-authority sites should run hard clustering. You cannot win broad head terms, so you want tight, specific pages aimed at narrow intents. Fewer keywords per page, more pages.
Established sites with real link equity should run moderate. You have the authority to rank a comprehensive page for a wide group, and consolidation compounds in your favour.
Nobody should run soft clustering. The false merges it creates are the exact over-merging problem covered below.
One caveat worth building into your process: SERP overlap is volatile for queries under roughly 50 monthly searches, where the top 10 shifts week to week. For those, fall back to manual intent review.
The Five-Step Keyword Clustering Process
This is the process I run on every content programme.
- Export the raw keyword set. Pull everything relevant — seeds, long-tail variations, questions, competitor keywords. Do not filter for volume yet. A 3,000-row list is normal and fine.
- Pull the top 10 for every keyword. This is the step people skip because it costs API credits. It is also the step that makes the output trustworthy. Without live SERP data you are guessing.
- Cluster at your chosen threshold. Run the overlap comparison and generate groups. Expect 3,000 keywords to collapse into 150–400 clusters depending on threshold.
- Assign one intent and one page per cluster. Label each cluster informational, commercial, or transactional, then name the single URL that will own it. If a cluster needs two labels, it is two clusters — split it now.
- Map clusters to existing URLs before writing anything. Most clusters already have a page. Match first, identify genuine gaps second. The output is a three-column sheet: cluster, target URL, action (keep, merge, redirect, create).
Step five is where the money is. Teams that skip it commission new content for clusters they already cover, then wonder why the new page underperforms the old one it is quietly cannibalising. Getting this right is core to any serious content SEO programme.
The Over-Merging Trap: When to Split a Cluster
Every guide teaches clustering as a merging exercise. That framing causes the most common failure I see.
Over-merging happens when a cluster grows until it contains multiple intents and the page serving it has to be everything at once. The symptom is a 4,000-word page that ranks position 8–15 for everything and page one for nothing. It is comprehensive and it is invisible.
Split a cluster when any of these are true:
- The intents differ. “how to do keyword clustering” and “keyword clustering tool” are a guide and a comparison page. Never the same URL.
- The buying stage differs. Research queries and vendor-selection queries need different page types, different CTAs, and different lengths.
- The SERP formats differ. If one keyword returns listicles and the other returns tools or videos, Google is signalling two content types.
- The page needs more than one H1’s worth of promise. If you cannot write a single title that honestly covers the whole cluster, the cluster is too big.
The test I use is blunt: write the title tag first. If the title requires an “and” to cover the cluster, split it. A cluster you cannot name in one phrase is not a cluster.
This is also where clustering connects to topical authority. Authority comes from covering a topic across a coherent set of pages that each own one intent — not from a single page trying to own a whole category.
How to Audit Clusters After They Ship
Clusters decay. SERPs shift, competitors publish, intent drifts, and a grouping that was correct in March is wrong by September. Almost nobody re-audits, which is why programmes stall.
Run this quarterly, using Search Console rather than a clustering tool:
- Pull the query report per URL. For each cluster page, list the queries it actually receives impressions for.
- Flag pages receiving impressions for queries outside their cluster. That is drift. Either the page is broader than planned, or a competitor reshaped the SERP.
- Flag any query where two of your URLs both appear. That is live cannibalisation and needs a merge or a redirect. Google’s guidance on consolidating duplicate URLs covers the mechanics.
- Flag clusters where the page ranks 8–20 across the board. That is the over-merging signature. Split it.
Two of these checks take an afternoon and routinely surface more upside than a quarter of new content. It is the cheapest work in the discipline and the first thing I look at in an SEO audit.
Keyword clustering is not a research technique you run once at the start of a project. It is a maintenance discipline: cluster against real SERP data, set the threshold to match your authority, split as readily as you merge, and re-audit every quarter. Do that and the same content library produces substantially more traffic without a single new page. Get the clusters right and the rest of your keyword research finally has somewhere to go.
Frequently Asked Questions
What is keyword clustering in SEO?
Keyword clustering is the process of grouping keywords that share the same search intent so they can be targeted with a single page instead of one page per keyword. The grouping is usually based on SERP overlap — if two queries return many of the same top-10 URLs, Google treats them as one topic. The result is fewer, stronger pages that rank for the entire group rather than many thin pages competing with each other.
How do you cluster keywords?
Export your full keyword list, pull the live top 10 results for each keyword, then group any keywords that share at least three of the same URLs. Assign one search intent and one target URL to each resulting group, and map those groups against pages you already have before commissioning anything new. Tools like Keyword Insights, LowFruits and SE Ranking automate the SERP comparison, but the intent assignment and URL mapping still need a human.
How many keywords should be in a cluster?
There is no correct number — cluster size should be whatever a single page can genuinely satisfy. In practice most well-formed clusters hold 5 to 30 keywords, but a narrow commercial cluster might hold three and a broad informational one might hold sixty. Judge by intent coherence rather than count: if you cannot write one title tag that honestly covers every keyword in the group, the cluster is too big and needs splitting.
What is the difference between keyword clustering and topic clusters?
Keyword clustering is the analytical step of grouping queries by shared intent; topic clusters are the site architecture you build afterwards. Clustering tells you which keywords belong on one page. A topic cluster arranges those pages into a pillar-and-spoke structure connected by internal links. You cluster keywords to decide what pages exist, then organise those pages into topic clusters to signal coverage.
Is keyword clustering still relevant in 2026?
Yes, and arguably more than before. AI Overviews and answer engines synthesise from pages that comprehensively cover an intent, which rewards consolidated cluster pages over fragmented ones. The technique has not changed — what changed is the cost of getting it wrong, since thin single-keyword pages now lose to both classic rankings and AI citation. Clustering against live SERP data remains the most reliable way to decide what a page should cover.
