Free Semantic Keyword Clustering Tool

How this tool computes its result

Despite the name, clustering here is lexical, not embedding-based. Keywords are deduplicated, then for each not-yet-used keyword the tool walks the remaining list and pulls in any other keyword that either scores >=60% on a Levenshtein-distance similarity ratio OR shares at least one whitespace-split word (case-insensitive) with it -- whichever condition fires first is enough. The shortest keyword in each resulting cluster becomes its "primary" label, and the displayed similarity percentage is the average Levenshtein similarity of every cluster member against that primary keyword (100% for singleton clusters). Clusters are sorted largest-first.

Worked example

Pasting "AI marketing", "marketing with AI", "SEO tools", "SEO software", "content marketing", "content strategy" produces three 2-item clusters: "AI marketing"/"marketing with AI" merge on the shared word "marketing", "SEO tools"/"SEO software" merge on shared word "SEO", and "content marketing"/"content strategy" merge on shared word "content" -- none of these pairs necessarily clear the 60% Levenshtein threshold, they cluster purely because they share one literal word.

When not to use this tool

There is no synonym or embedding awareness at all: "cheap flights" and "flight delays" would cluster together on the shared word "flight" despite unrelated intent, while true synonyms sharing no substring (e.g. "affordable" and "cheap") never cluster. Treat clusters as a rough dedup/grouping pass, not a semantic-intent analysis.

Common mistakes

• - Reading the displayed similarity percentage as topical relevance -- it is the Levenshtein-distance similarity to the cluster's primary keyword only, not a measure of shared search intent.
• - Not realizing a single shared low-value word (e.g. a generic modifier) is enough to force two otherwise-unrelated keywords into one cluster, independent of the 60% edit-distance threshold.
• - Pasting well beyond the suggested ~50-keyword limit -- clustering is an O(n^2) pairwise comparison with no cap, so large lists get noticeably slower in-browser.

Related tools