Topic planning
Keyword Clustering: Group Searches Into Pillar Posts
Keyword clustering groups the searches one post can answer. How to cluster by shared words and top-10 overlap, and what our own repeated SERP pulls showed.
Keyword clustering is just sorting your keyword list into groups that a single page can answer, so you end up writing one strong post for each group and your near-duplicates stop competing with each other. Start by grouping keywords by the words they share. Then, for the few pairs you can't call by eye, look at Google's top 10 for both and see how much overlaps.
TL;DR
- A common rule says two keywords belong in one cluster when they share three or more of the same top-10 results.
- In our own test, one look at the results wasn't enough. The same pair scored 0, 1 or 4 shared URLs depending on which pull we used.
- Merge the plain variants first, and don't add their search volumes together, because you'll be counting the same searches more than once.
- Each cluster becomes one post, and those posts get arranged as a pillar with supporting posts that link to it.
What is keyword clustering?
Keyword clustering is the step that sits between having a keyword list and having an actual content plan. You take every search phrase you've collected and sort the phrases into groups, and each group should be something one post can fully answer.
So "seo audit tool" and "audit seo tool" go in the same group. "seo audit checklist" probably goes somewhere else, because the person typing that wants a list of steps to follow, and a roundup of software won't help them much.
What you get at the end is a short list of posts to write, where each post has one main keyword plus a handful of close phrasings it should also cover.
Small blogs feel this the most. If you're shipping two or three posts a week, you can't spare the hours for near-twins, and writing "keyword clustering tool" and "best keyword clustering tools" as two separate articles just sets them up to fight over the same spot.
Why does keyword clustering matter for a small SaaS blog?
When two of your posts go after the same intent, they end up competing for the same results, and whatever links you've earned get split between them. Clustering catches that before the second post ever gets written.
Clusters also give the blog a structure. A pillar post links out to the narrower posts under it, and each of those points back to the pillar.
Google says AI Overviews and AI Mode may use a "query fan-out" technique, which means running several related searches across subtopics before writing one answer. A good cluster lines up closely with those subtopics, so a pillar and its supporting posts give more of those background searches something of yours to land on. Google doesn't publish which sub-queries a given answer used, though, so treat this part as a guess.
How do you cluster keywords, step by step?
1. Collect the list and merge plain variants
Pull keywords from your research tool and from Search Console into one sheet, then add whatever your competitors' best pages rank for. Merge the rows that are really the same search written a different way, like plurals or swapped word order.
Keep an eye on the volume column while you do it. In SupaCite's keyword list for our own site, five different phrasings of "seo audit tool" each show 2,900 monthly searches and the same difficulty score of 78. If you add them up you get 14,500, which is the same 2,900 searches counted five times. Google's Keyword Planner reports average monthly searches for a keyword and its close variants as one figure, and tools built on Google Ads data show that same figure next to every variant, which would explain the identical numbers.
2. Make a fast first pass by shared words
Sort whatever's left by the words the keywords have in common. "keyword clustering tool" and "free keyword clustering tool" will land together, and a few hundred keywords is pretty manageable in a spreadsheet with a text filter.
Shared words are only a guess about intent. "keyword clustering" and "keyword clustering tool" share two words, but the top 10 for the tool version is almost all roundups and vendor pages. A how-to guide isn't going to match that.
3. Check the top 10 for pairs you can't call by eye
This step is only for the pairs that could go either way. You search both keywords and compare their top 10 results, then count how many URLs show up for both. HubSpot's guide puts the bar at three or more shared URLs for one cluster, and says an overlap of zero or one likely calls for separate pages. Clustering tools do the same comparison, just across thousands of keywords at once.
4. Give each cluster one main keyword and one URL
Each cluster gets exactly one URL. Pick the phrasing with the clearest intent as the main keyword and put it in the title and the first couple of lines, then work the other phrasings into headings and the FAQ wherever they read naturally. If two of your existing posts already sit in the same cluster, fold the weaker one into the stronger one and redirect its URL.
5. Arrange the clusters into pillars and supporting posts
Now put all the clusters side by side and pick out the broad ones, like "keyword research". Write that pillar first. Then hang the narrower posts off it, things like "keyword clustering", "search intent", "long-tail keywords" or "keyword difficulty", and link each of them to the pillar in both directions.
We built the SupaSidebar blog this way, with the pillar first and then supporting posts every few days. Across about 200 posts it went from 1,300 to 25,400 monthly visitors in four months (measured in PostHog), though clustering was only one part of that, next to publishing steadily and refreshing older posts.
Where does the overlap test fall short?
On 16 November 2026 we pulled Google's US top 10 for eight keywords through DataForSEO's SERP API. We pulled "keyword clustering" three times and "seo audit tool" twice, a few minutes apart and with the exact same settings. We expected at least one pair to cluster cleanly, and none did.
| Keyword pair | Shared URLs, one number per pull of the first keyword |
|---|---|
| keyword clustering vs what is keyword clustering | 0, 1, 4 |
| keyword clustering vs keyword clustering tool | 0, 2, 0 |
| keyword clustering vs keyword grouping | 1, 0, 0 |
| keyword clustering vs topic clusters | 0, 0, 0 |
| seo audit tool vs seo audit tools | 0, 0 |
| keyword grouping vs how to cluster keywords | 1 |
Each keyword returned seven to nine organic results. The first row is the interesting one. Depending on which pull we used, "keyword clustering" and "what is keyword clustering" scored 0, 1 or 4 shared URLs, so they were under HubSpot's bar twice and over it once. And "seo audit tool" and "seo audit tools", which pretty much anyone would put on one page, didn't share a single URL in either pull.
Even the repeated pulls didn't agree with each other. The first pull of "keyword clustering" shared no URLs at all with the second or third, while those two shared two. The two pulls of "seo audit tool" also shared two.
We can't tell how much of that is Google reshuffling results and how much is the scraping itself. Clustering tools build their groups from the same kind of pull, so we'd treat any single overlap count, whether it's yours or a tool's, as a rough read.
- If a pair could go either way, pull it more than once, ideally on different days, and only count the URLs that keep coming back.
- A low count on a broad keyword doesn't tell you much. Every pull of "keyword clustering" mixed how-to guides with tool pages, which spreads the results thin.
- For plain variants like plurals, trust common sense over the count.
- When one site ranks a different page of its own for each keyword, that's a small hint Google sees two different jobs. In our pulls, a clustering vendor ranked its homepage for "keyword clustering" and a blog post for "keyword grouping".
Which clustering method should you use?
| Method | How it groups | Cost and speed | Where it goes wrong | Best for |
|---|---|---|---|---|
| Shared words | Keywords with a word or phrase in common | Free and instant, works in a spreadsheet | Misses synonyms, merges different intents | The first pass on any list |
| Top-10 overlap (SERP) | Keywords whose top results overlap | One search per keyword, paid at scale | Noisy pulls, mixed-intent results | Doubtful pairs and high-value posts |
| Meaning (embeddings) | Keywords a language model scores as similar | Cheap once set up | Groups by topic and ignores what format the searcher wants | Big lists you'll still read by hand |
| Hybrid | Words or meaning first, then a SERP check on doubtful groups | Moderate | Takes two passes | Most small SaaS blogs |
For most small SaaS keyword lists we'd go hybrid. The shared-words pass does most of the sorting for free, so you only spend money or time on SERP checks where getting it wrong would actually cost you a post.
If you already pay for a big SEO suite, check what it has before buying anything else. Ahrefs, for example, groups keywords by "Parent Topic" right inside its Keywords Explorer.
In SupaCite, the Keywords page does that shared-words pass as soon as you add keywords, and each planned topic carries its cluster and its role, pillar or supporting post. The top-10 check from step 3 is still yours to run on the pairs that need it.
Quick decision: how should you cluster your list?
- A few hundred keywords or fewer: a spreadsheet is fine. Sort by common words and only run the top-10 check on the handful of pairs you're unsure about.
- Thousands of keywords: that's when a clustering tool is worth paying for, ideally one that re-checks overlap over time, but read every cluster yourself before anything gets written.
- Product pages or roundups for one keyword and how-to guides for the other: give them separate posts, whatever the overlap count says.
- A pair that lands on exactly two shared URLs: check it again on another day before you decide.
So what should you actually do?
For the borderline pairs, write the post only once the overlap holds up on a second look, since one pull can mislead you, and the rest of your list rarely needs more than the shared-words pass. It also helps to keep the sheet afterwards, with the URL each cluster maps to written next to it. That way, when a new keyword turns up later, you'll see straight away whether it belongs to a post you already have.
If you'd like the first pass done for you, get started free on SupaCite.
Frequently asked questions
What is the difference between keyword clustering and topic clusters?
A topic cluster is a set of linked pages, usually a pillar and its supporting posts. Keyword clustering is how you figure out which of those pages to write, and which searches each one should cover.
How many keywords belong in one cluster?
There isn't a fixed number. A how-to about one task may only need three or four phrasings. Pillars run a lot bigger, sometimes dozens of long-tail questions, and once a post can't answer everything in its cluster without sprawling, split the cluster.
Is SERP-based clustering better than grouping by meaning?
For deciding whether two keywords can share a URL, we'd lean on top-10 overlap, since it's closer to what Google shows. Our own repeated pulls disagreed, though, so treat it as a check you run twice. Grouping by meaning is cheaper and fine for a first sort, but it can't tell someone learning a concept from someone shopping for software.
Do you need a paid tool to cluster keywords?
Not for small lists. The part you pay for is the SERP lookups, so the cost grows with the number of pairs you check, and doing a shared-words pass first keeps that number small.
How often should you re-cluster your keywords?
Look again whenever you add a batch of new keywords, and every few months anyway. It's also worth a check when two of your own pages start trading places in Search Console for the same query.
Does keyword clustering help with AI search?
Probably a little. Each fan-out search goes after a subtopic, so each supporting post should answer its own question in its first few lines. There's no public data yet on how much that helps.
Supacite writes a researched post every day and tracks you in ChatGPT, Gemini, Perplexity and Claude.
Get started free