What the Topic Cluster Mapper does
The Topic Cluster Mapper groups a keyword list - or a list of page titles - into clusters of closely related terms, picks a pillar for each cluster and marks the rest as supporting topics, then exports the whole content map as CSV or Markdown. You control how tight the clusters are with a similarity threshold, and every result can be traced back to the words the items share.
The method is plain word overlap (Jaccard similarity) applied in a fixed order, so the same list and threshold always give the same map. It runs entirely in your browser; nothing about your keyword research leaves the page.
How to use it
- Paste keywords one per line, or a CSV export with a Keyword column and, ideally, a Volume column. Page titles with a URL column work too, for mapping existing content.
- Set the similarity threshold. 0.3 is a sensible start; move towards 0.2 for broader clusters or 0.5 for tighter ones. The map redraws as you move the slider.
- Select Build clusters. The cards show the six largest clusters, each with its pillar in the centre and supporting items around it; line thickness is similarity to the pillar.
- Read the table for every item's cluster, role and similarity, and review the unclustered items: some are genuinely separate topics, others are synonyms that share no words.
- Download the CSV or Markdown map and adjust it by hand before briefing content.
Reading the results
A pillar is the item most similar to all the others in its cluster - the natural broad page. Supporting items are narrower topics that link to it. Unclustered items matched nothing at the current threshold.
Similarity to pillar is shared terms divided by all distinct terms of the two items, after removing stop words and folding plurals. 1.00 means the same terms ("how to prune roses" and "when to prune roses" both reduce to prune + rose); 0.33 means one term shared out of three.
Cohesion is the average similarity between every pair in a cluster. Low cohesion with many members suggests the threshold is too loose for that group.
Worked example: twenty keywords for a garden centre
The example list has twenty keywords with monthly volumes. At a threshold of 0.30 it forms six clusters covering 13 keywords (65%): compost bin with two supporting items, and five pairs - growing and planting tomatoes, lawn weed and moss killers, pruning roses, rose black spot and climbing roses.
"how to prune roses" and "when to prune roses" have a similarity of 1.00, since how, when and to are stop words. "compost bin for small garden" joins "compost bin" at 0.50 (two of four terms). Seven keywords stay unclustered, including "growing tomatoes in pots": the light stemmer does not reduce growing to grow, so it shares only tomato with its neighbours - exactly the kind of case to merge by hand.
Formulas and scoring rules
- Terms
terms(item) = lower-cased words - stop words, with plurals folded (shoes -> shoe, policies -> policy)With word pairs on, adjacent pairs such as "running shoe" are added.- Similarity
J(A, B) = |A n B| / |A u B|Rounded to two decimals for display, three in the CSV.- Assignment
an item joins the cluster whose seed it is most similar to, if J >= threshold; otherwise it seeds a new clusterItems are processed by volume (high first), then fewer terms, then alphabetically.- Pillar
pillar = member with the largest sum of J to the other membersTies go to higher volume, then earlier input.- Cohesion
cohesion = mean J over all member pairs
Limitations: what the result does not prove
- It measures shared words, not meaning. Synonyms ("sofa" and "couch") and different wordings of one need never cluster; words with two senses can cluster wrongly.
- The stemmer only folds plurals and possessives, on purpose, so "marketing" and "market" stay apart - but "growing" and "grow" do too.
- It does not look at search results. Whether two keywords can share a page is best judged by whether the same pages already rank for both.
- Clustering 5,000 items compares each against every cluster seed; very large lists with many clusters can take a few seconds in the browser.
Privacy: where your data goes
Everything you paste, type or drop is processed in this browser tab. It is not uploaded, logged, stored or sent to analytics. Session recording and tag-manager scripts are switched off on this page.
Standards and sources
Frequently asked questions
What is a topic cluster?
A group of pages on one subject: a broad pillar page and several narrower supporting pages that link to it and to each other. Planning content in clusters helps you cover a subject thoroughly without writing several pages that compete for the same searches.
What threshold should I use for keyword clustering?
Start at 0.3. If unrelated keywords end up together, raise it; if obvious variants stay apart, lower it. Short keywords share proportionally more words, so lists of two- and three-word keywords usually suit a slightly higher threshold than long questions.
Why did two keywords with the same meaning not cluster?
Because they share no words after stop words are removed, or share too few for the threshold. This tool compares words, not meanings - that is what makes it transparent and repeatable. Merge such pairs by hand in the exported CSV.
How is the pillar chosen?
It is the member most similar to all the other members combined, not simply the one with the most volume. Volume only decides the order items are processed in and breaks ties.
Can I cluster existing pages instead of keywords?
Yes. Paste page titles, or a CSV with Title and URL columns. The URL is carried into the export so the map shows which existing page is the pillar and which pages support it.
Will I get the same clusters if I run it again?
Yes. The order of processing is fixed by volume, length and alphabet, so the same list at the same threshold always produces the same clusters, whatever order you paste it in.
Last reviewed by the A2Z.Tools team against the sources listed above.
