Keyword Clustering

Keyword clustering is the process of grouping a keyword list into sets that belong together, usually because they express the same need or are satisfied by the same page. It turns thousands of rows of keyword research into a manageable set of topics you can map to pages, prioritise and report on.

How it works

There are three broad approaches, and they answer slightly different questions:

• Semantic clustering. Each keyword is turned into an embedding with a model such as Sentence-BERT, then grouped with an algorithm like k-means, hierarchical clustering or HDBSCAN. Keywords close in meaning end up together.
• SERP-based clustering. Keywords are linked when their top 10 results share several URLs. If Google shows the same pages for two queries, one page can usually target both.
• Keyphrase or rule-based grouping. Tools such as KeyBERT extract the core unigram and bigram from each keyword, and those shared terms become cluster labels. N-gram analysis works in a similar spirit.

How it’s used in SEO and marketing

• Deciding which keywords one page should target and where you need separate pages (keyword mapping).
• Prioritising topics by summing search volume, averaging keyword difficulty or counting competitors per cluster.
• Spotting cannibalisation, where several of your URLs compete for one cluster.
• Planning topic clusters and briefs in a semantic keyword universe.
• Comparing markets: the same cluster can need one page in one country and two in another.

Example

In MLforSEO’s SERP analysis guide, keywords are linked when they share at least three of the same top-10 URLs, then grouped. The threshold is a judgement call: start at three and tune it by checking a few pairs by hand. In the KeyBERT tutorial, the clusters are then visualised in Looker Studio with search volume, difficulty and CPC per cluster.

Limitations and things to watch

• Fuzzy string matching is a poor way to cluster at scale, because it compares characters rather than meaning. MLforSEO’s fuzzy matching guide says so explicitly.
• Semantic and SERP-based methods can disagree. Two keywords can mean similar things but need different pages because the intent differs.
• Most algorithms need a parameter you choose (number of clusters, similarity threshold, minimum cluster size), and results change with it.
• Always review a sample of clusters by hand before building a content plan on them.

Learn it on MLforSEO

• How to do keyword clustering with KeyBERT
• 10 Ways to Automate SERP Analysis with AI (clustering by SERP overlap).
• What Machine Learning Clustering Is, and Where It Fits in SEO
• KeyBERT for SEOs: Google Colab Template (resource).
• Introduction to Machine Learning for SEO course