Keyword Clustering Tool

Group keywords into clusters one page can rank for

This keyword clustering tool turns a flat keyword list into groups, where each group is a set of searches that one well-made page can satisfy. Instead of writing a separate article for “sourdough starter”, “how to make sourdough starter” and “sourdough starter recipe”, you see that they belong together — and that “sourdough bread recipe” needs its own page.

Paste up to 200 keywords, with search volumes if you have them. A lexical preview groups them by shared words straight away, in your browser and without AI. Then Cluster with AI asks TypeSafe’s Jev model, pair by pair, whether one page would fully serve both searches — so “bread maker” and “bread machine” can meet even though they share only one word. You can move any keyword by hand, then export a CSV or a ready-to-use content plan.

How to cluster keywords

1. Paste your keywords, one per line. Add a volume after a comma (“bread maker, 22000”) or paste columns from a spreadsheet or keyword export with a header row.
2. Check the lexical preview cards: keywords that share most of their words are already grouped.
3. Press Cluster with AI. The cards are replaced by AI clusters, ordered by total volume, each headed by its highest-volume keyword.
4. Fix anything you disagree with: drag a keyword onto another card, or tap it and press Move here. Open Borderline pairs to review near misses.
5. Press Download CSV for a cluster/keyword/volume sheet, or Content plan (.md) for one heading per page.

Keyword Clustering Tool features

How the clusters are made

Lexical preview. Each keyword is reduced to its meaningful words: common words such as “how”, “the” or “best” are dropped and endings are trimmed, so “feeding” and “feed” match. The similarity of two keywords is the share of words they have in common (the Jaccard index). Groups merge, strongest pair first, only when the average similarity across all their pairs is at least 0.5. It is fast and free, but blind to meaning.

AI clusters. Code proposes each keyword’s four nearest neighbours by word overlap (initials count too, so “bmi” can meet “body mass index”). For each proposed pair the AI answers one question: would one well-made page fully serve both searches? Groups merge when their judged pairs average at least 60% and none is below 30%. Cohesion is the average of those probabilities inside a cluster. Pairs at 40–59% are listed as borderline instead of being merged silently.

Worked example

Take six keywords: sourdough starter, sourdough starter recipe, feeding sourdough starter, banana bread, banana bread recipe and bread maker. The lexical preview makes three groups. “sourdough starter” and “sourdough starter recipe” share two of three words (0.67), and “feeding sourdough starter” joins them because the average across the group stays above 0.5. “banana bread” and “banana bread recipe” form a second group. “sourdough starter recipe” and “banana bread recipe” share only “recipe” (0.2), so the average linkage keeps the topics apart, and “bread maker” — a product, not a recipe — stands alone.

The sample list with volumes previews as six groups: the starter, sourdough bread, banana bread and bread maker families, plus “sourdough starter not rising” and “bread machine” on their own. The AI is then asked about pairs the preview cannot see as related — whether “bread machine” and “bread maker” need one page (they are the same appliance), and whether “healthy banana bread” belongs with “banana bread recipe”. Pairs it scores between 40% and 59% appear under borderline pairs; whether they share a page is your call.

From clusters to a content plan

• One cluster, one page. Use the primary keyword for the title and the main heading, and cover the other keywords as sections or natural phrasing.
• Check intent before merging: a guide and a product page rarely satisfy the same searcher. Label your list with the Keyword Intent Classifier first.
• Write the title for the whole cluster with the SEO Title Generator, then check how often the page uses each phrase with the Keyword Density Checker.
• Single-keyword clusters are not wasted: they are often sections of a bigger page, FAQ entries or ideas for later.