Keyword Mapping

Keyword mapping is the process of assigning keywords to the topics, clusters or pages they belong to, so every query in your research has a clear home. MLforSEO uses it in two linked ways: mapping keywords to topics, to organise a keyword universe, and mapping groups of keywords to the pages or page types that should serve them.

How it works

MLforSEO’s guide to keyword-to-topic mapping compares four approaches, from lightweight to more robust:

• Fuzzy matching scores how closely the characters of a keyword resemble a seed topic. Quick and easy, but it ignores meaning.
• Sentence-BERT turns keywords and your topic labels into embeddings and assigns each keyword to the label with the highest cosine similarity. It can map a keyword to a topic even when they share no words.
• K-means clustering splits keyword vectors into a chosen number of groups. It does not name the groups.
• BERTopic discovers topics on its own and labels them with their most representative terms.

Fuzzy matching and Sentence-BERT need a topic list from you (supervised mapping); k-means and BERTopic find structure without one (unsupervised). Use supervised mapping when you already have a taxonomy that mirrors your site, unsupervised when you don’t know the topics yet. Mature keyword programmes often run both: unsupervised first to discover topics, then supervised to map new keywords onto them.

Mapping topics to pages adds a SERP check. Keywords that Google answers with the same URLs usually belong on one page; keywords that share no ranking URLs almost always need separate pages, even when they look similar.

How it’s used in SEO and marketing

• Organising a large keyword list into topics for briefs, dashboards and reporting.
• Planning site architecture and topic clusters, with one primary page per keyword group.
• Finding cannibalisation, where two pages target the same group, and content gaps, where a group has no page.
• Supporting internal linking and redirect mapping, which use the same matching techniques.

Example

In the MLforSEO notebook, you upload one file of topics and one of keywords. The fuzzy matching version returns each keyword with its best-matching seed term and a match score; the Sentence-BERT version returns the closest topic and a similarity score. For the page level, the SERP analysis guide clusters keywords by shared URLs in the top 10, starting at a threshold of three shared URLs and tuning it by checking pairs by hand. In that data, the US and UK top 10 for the same English keyword shared no URLs at all, so each market needed its own check.

Limitations and things to watch

• Similarity is not the same as “Google serves these with one page”. Confirm page-level decisions with SERP overlap and search intent.
• Scores need thresholds, and thresholds need hand-checking on a sample.
• Supervised mapping is only as good as your topic list; unsupervised topics need a human to name and sense-check them.
• Maps go stale as SERPs and your site change, so refresh them with your keyword data.

Learn it on MLforSEO

• Different ways to Map Keywords to topics explains all four approaches with code.
• Map Keywords to Topics: Supervised vs. Unsupervised (Notebook) and Fuzzy Matching: Map Keywords to Seed Terms or Topics (Notebook) are ready to copy.
• 10 Ways to Automate SERP Analysis with AI covers clustering keywords by shared URLs.
• The Semantic ML-enabled Keyword Research course puts mapping into a full keyword universe workflow.