Keyword clustering for content planning: stop guessing what to write next
You have 400 keywords in a spreadsheet. You've had them for weeks. Every Monday you open the file, stare at it, pick the one that sounds most promising, and write. That's not a content plan—that's a lottery ticket with extra steps.
The fix isn't another keyword tool. It's keyword clustering for content planning: grouping queries by the page that would actually answer them, then letting those groups drive your editorial calendar. That one shift is what turned my own publishing from a scatter of orphan posts into something Google clearly understood—and it took a complete rebuild of how I worked, not a new subscription.
Key Takeaways
- A cluster is a set of keywords that should be answered by one page, not one keyword per article.
- Group by intent first, by semantic similarity second. The reverse order produces clusters you can't write.
- Publish your pillar page before the spokes, or the spokes compete with each other for the same position.
- Cannibalization is a planning failure, not a writing failure: you fix it in the map, not the draft.
- Track clusters, not individual keywords. Position per cluster tells you which hub deserves the next article.
- A free spreadsheet does 80% of what a paid clustering tool does. The tool saves time, not thinking.
What a cluster actually is when you're planning, not just researching
Every guide tells you a cluster is a group of semantically related keywords. True, and useless. Here's the working definition I use:
A cluster is the smallest set of keywords that one URL can rank for without splitting its own authority.
That framing changes everything, because it forces you to decide the URL before you decide the topic. When I first started grouping queries in 2023, I built clusters of 30–40 keywords each and wrote a single 5,000-word monster post for every one. Two of them ranked. The rest flopped—too broad to satisfy any single intent, too long for anyone to finish.
Intent before semantics
Two keywords can be 95% similar in wording and belong to different clusters. "How much does email automation cost" and "email automation pricing page examples" share almost every token. They want completely different pages.
So the order is:
- Sort every keyword by intent: informational, commercial, transactional, navigational.
- Inside each intent bucket, group by the answer the searcher needs—not by the words they typed.
- Only then check semantic overlap to catch duplicates.
- Assign one URL per group. If a group needs two URLs, split it.
Franchement, step 1 is where most plans die. People cluster alphabetically, or by volume, or by whatever the tool's algorithm spits out, and end up with a "cluster" that mixes a research question with a buying question.
The pillar and its spokes
Once your clusters exist, they need a shape. The pattern that works for me:
- One pillar page that covers the head term broadly—usually 2,000–2,500 words, high-level.
- 4 to 9 spoke articles, each answering one specific sub-intent in depth.
- Every spoke links up to the pillar with a contextual, descriptive anchor.
- The pillar links down to its best two or three spokes, not all of them at once.
A notes app I used for years had 60 isolated posts on the same topic, all competing. After restructuring into four clusters with four pillars, average position across the whole topic went from the high 30s to page one over about five months. Same content, mostly. Different architecture.
Building the map: a workflow you can run this week
Start from queries you already have
Don't begin with a fresh keyword dump. Export the queries your site already receives from your analytics or Search Console. That gives you a baseline: which clusters you're accidentally half-covering, and which one deserves investment first. In my experience, most sites discover two or three clusters they never intended to build.
Cluster by hand for the first batch
Sort the sheet, add a column called "page", and assign a URL to each keyword. Yes, manually. It's slow for the first 100 rows and then you're fast. Paid clustering tools exist and they're decent, but they don't know your business model, your margin, or which topic you can actually speak about with authority.
Score each cluster, then sequence
I rank clusters on three things: business relevance, how much you already know about the topic, and how crowded the results already are. High relevance plus zero existing coverage is your first project. A table makes this concrete:
| Cluster | Keywords | Business fit | Existing coverage | Priority |
|---|---|---|---|---|
| Pricing and plans | 22 | Very high | 1 weak page | 1 |
| Setup and onboarding | 34 | High | Nothing | 2 |
| Integrations | 11 | Medium | 3 thin posts | 3 |
| Industry comparisons | 47 | Low | Nothing | Skip for now |
That last row matters. A 47-keyword cluster with low business fit is a trap—it's the biggest group on the sheet and the easiest one to waste a quarter on.
Turn clusters into a calendar
Publish the pillar first, even if it's rough. Then one spoke per week or per two weeks, depending on your capacity. I tried the reverse once—spokes first, pillar later—and the spokes cannibalized each other for three months before the pillar existed to anchor them.
Do you need a keyword clustering tool?
Sometimes. Here's the honest split, based on what I've actually paid for and abandoned:
| Approach | Best for | Weakness |
|---|---|---|
| Spreadsheet + manual grouping | Under 300 keywords, high-stakes topics | Slow, doesn't scale |
| Free clustering tools | First pass on a large list, discovering obvious groups | Groups by wording, not intent |
| Paid suites | Thousands of keywords, multiple markets | Cost, and you still validate by hand |
| SERP-overlap clustering | Competitive niches where intent is ambiguous | Needs result data you may not have easy access to |
For most solo publishers and small teams, a free tool to make the first pass plus a spreadsheet for the final decision is the right combination. The tool saves you an afternoon. It doesn't save you the judgment call.
What to measure once a cluster is live
Individual keyword rankings are noise. Watch the cluster.
- Average position per cluster—a single number that tells you whether the whole topic is gaining.
- Coverage: what share of the cluster's keywords a page actually ranks for at all.
- Cannibalization: two of your URLs bouncing between positions for the same query. When you see it, merge or differentiate—don't wait.
- Internal link clicks: are people moving pillar to spoke, or landing and leaving?
When a cluster's coverage plateaus, that's the signal to write the next spoke, not to rewrite the pillar. Every cluster I've built follows the same curve: fast gains, then a flat stretch that only breaks with a new page in the group.
Where clusters fit in an AI-search world
Conversational assistants answer questions, not queries. That changes cluster design in one practical way: a cluster should cover a complete question-and-follow-up chain, not a keyword set. If your cluster answers "what is X", "how much does X cost", and "X vs Y" in connected pages, a model synthesizing an answer has a coherent source to draw from. Fragmented coverage gives it fragments.
I'll say this plainly: I'm not certain how much this matters yet, and anyone claiming to know the exact playbook is guessing. But the sites I see cited by assistants are the ones with tight, well-linked clusters on a narrow range of topics—not the ones with 800 disconnected posts.
The plan is the product
Here's the uncomfortable part. Clustering doesn't make writing easier. It makes it harder, because it removes your excuse for writing whatever felt interesting that week. You stop being a person with a blog and start being a publisher with a map.
That's the trade. And the next time you open that 400-row spreadsheet, the question won't be "what sounds promising?" It'll be "which cluster is one spoke short of ranking?"—and you'll already know the answer.
