This tool groups a keyword list by the words the keywords share. Paste a few dozen phrases, choose how strict the match should be, and it returns clusters you can use to plan pages. It runs in the tab, keeps duplicates out, and exports the full grouping as a CSV.
It is a lexical clusterer, not a semantic one. It knows that keyword research template and keyword research tool share two words. It does not know that link profile and backlink audit mean similar things. That limit is the most important thing to understand before you plan anything from the output.
How two keywords are compared
Each keyword is reduced to its words, with optional stop-word removal and stemming. The comparison score is 70% containment and 30% Jaccard similarity. Containment asks how much of the shorter keyword appears in the longer one; Jaccard asks how much the two sets share relative to everything they contain. Weighting containment higher means a short phrase can absorb its longer variants without every variant needing to match.
A keyword joins an existing cluster when its best score against any member is at or above the threshold. Otherwise it starts a new cluster. The comparison is against members, not against the cluster label, so a group can grow through a chain of overlapping phrases.
Choosing a threshold
The default is 35%, and the slider runs from 10% to 90%. A lower threshold creates fewer, broader clusters; a higher one creates tighter groups and more singletons. There is no correct setting: 35% is a reasonable first pass for a mixed list, and 60% to 70% is better when you already know the topics and want to split them.
Stemming is on by default and merges plurals and regular verb forms. Stop-word removal is on by default and stops for, the and with from making unrelated phrases look similar. Turning either off makes the grouping more literal.
A worked example
Twelve keywords at the default 35% produced three clusters: keyword research with five members, a second group of five, and backlink checker with two. The largest cluster held 5 of 12 keywords, the average size was 4, and no keyword was left alone.
The second group is the interesting one. It was labelled local seo, but its members were seo audit, seo audit checklist, technical seo audit, local seo and local seo services. The shared token seo pulled two different topics together, and the label came from the shortest member. It is a good example of why the label is a starting point rather than a page title.
The same list at 90% produced 11 clusters, 10 of them singletons. Only free keyword research tools and keyword research tool stayed together, which is close to an exact-match setting.
Read the label, then read the members
The label is the member with the fewest words, then the shortest text, then alphabetical order. That rule is predictable but not semantic. A cluster can end up with a label that describes one member and not the rest, especially when a generic word such as seo, tool or free appears across topics.
Before planning, open the cluster and read the members. If the group contains two ideas, raise the threshold or split the list by topic and run each part separately. A cluster of 30 keywords is usually several pages, not one.
Caps and blind spots
The tool takes up to 500 keywords per run. Anything beyond that is dropped silently, so a 700-line export produces a 500-line result with no warning. Remove duplicates first if you are near the cap, because the tool also deduplicates case-insensitively before clustering.
It does not use search volume, SERPs or click data. Two keywords can share no words and still deserve the same page, and two keywords can share a word and belong on different pages. The CSV gives you the grouping; the judgement about intent is still yours.