What Is Keyword Clustering in SEO?
Grouping searches that want the same answer, so one page can serve all of them. It is the step that turns a list into a plan. It is also the step most keyword research stops one move short of.
Grouping Searches That Want The Same Answer
You have a filtered list of searches. Some of them are different ways of asking the same thing and some are genuinely separate needs. Sorting them into groups, where each group could be answered by one page, is clustering. That is the whole idea and the difficulty is entirely in the judgement.
Where it sits in the order. After filtering.
Removing everybody who is not your customer comes first, which our method guide covers. Clustering an unfiltered list groups people you do not want.
What it produces. Pages, not phrases.
The output is a set of subjects, each one a page you could brief somebody to write. That is what makes research actionable.
Why it gets skipped. It looks optional.
A list already feels like a deliverable. Handing one over ends the engagement tidily and leaves the recipient with nothing they can act on.
What happens without it. A page per phrase.
Businesses faced with a raw list build a page for each promising row. That produces near-identical pages competing with each other, which the industry then named as a separate problem.
How long it takes. An afternoon.
For a few hundred filtered searches, one sitting with a printed list. That is a smaller job than most people fear.
What it is not. Organising your website.
Navigation groups things the way a business thinks about itself. Clustering groups them the way customers arrive. Those two structures overlap and they are not the same, which is why a menu is a poor substitute for this exercise.
What makes it hard. The basis.
Deciding what to group on. Almost everybody reaches for shared words, which is block two and is the reason this page exists.
Why Wording Is The Wrong Basis
Two phrases sharing every significant word can want completely different things. Two sharing none can want exactly the same thing. So grouping on the words is grouping on the least reliable signal available. It is what almost every business does first.
The same words, different needs. Common.
Boiler replacement cost and boiler replacement grants share the subject entirely. One person is budgeting for a purchase and the other is checking eligibility for help. One page cannot serve both well.
Different words, same need. Equally common.
Radiators cold at the bottom and system needs flushing share nothing. Both are the same customer with the same problem, arriving at different levels of knowledge.
What that proves. Words are a proxy.
They correlate with intent loosely and they are not intent. Grouping on them produces sets that look tidy and serve nobody.
The version we find constantly. Alphabetical sorting.
A spreadsheet sorted so similar phrases sit together, then grouped by eye. Fast, mechanical and it puts the budgeter and the grant applicant on one page.
The other version. Grouping by service.
All boiler searches together because the business thinks in services. Customers do not. The page ends up written from the business outward.
The trap for trades specifically. Materials and brands.
Searches naming a particular make or material look like separate subjects because the words differ. Frequently the person wants the same job done and mentioned the brand incidentally. One page handles all of them perfectly well.
What to group on instead. The situation.
What was happening to this person when they typed it. Two people in the same situation belong together whatever words they chose.
The Test That Works
Could one page satisfy somebody making either search. That is the entire test. Applied strictly to every pair, it produces correct groupings without any theory, any software or any understanding of how search engines work.
Why it works. It centres the reader.
The question is about a person's experience rather than about words or algorithms. That is the right frame, since a page serving people well is the goal anyway.
How to apply it. Imagine arriving.
Picture somebody who made search A landing on a page built for search B. Would they feel served. Or would they feel the page is mostly about something else.
The failure signal. Mostly about something else.
If either person would think that, the searches belong apart. It is a low bar and it separates a great deal.
The confirming check. Look at the results.
Search both phrases and compare what appears. If the results are largely the same pages, search engines already treat them as one subject. If they differ entirely, they are separate.
Why that check is worth doing. Free evidence.
It takes seconds and it is a direct statement about how the searches are understood, rather than a guess about it.
When the two disagree. Trust the results.
If your judgement says together and the results say apart, the results are describing what people actually get shown. Follow them.
Doing It Without Software
For a few hundred searches, doing this by hand is faster than learning a tool and produces better results. The judgement involved is about customers rather than about text, which is exactly the judgement a business owner already has and software does not.
What you need. A printout.
The filtered list on paper, somewhere to write and an uninterrupted afternoon. Working on paper is genuinely better here, because moving things around is faster than editing cells.
The first pass. Obvious families.
Go through once putting each search into a rough pile by situation. Do not agonise. Most will be clear and the difficult ones will announce themselves.
The second pass. The awkward ones.
Take the searches you could not place and apply the test properly. These are where the value is, because they are usually the ones revealing a subject nobody planned.
The third pass. Split what is too big.
Look at each pile and ask whether one page could really cover it. Anything that has swollen probably contains two situations.
The fourth pass. Name each pile.
Write a sentence describing who each group is for, in terms of their situation. If you cannot write that sentence cleanly, the group is not a subject yet and needs another look.
Who should be in the room. Whoever answers the phone.
They know which of these are the same question. Half an hour of their time removes most of the guesswork.
When software earns its place. Thousands of rows.
Large sites with continuous research genuinely need it. Below that scale it automates the wrong part, which is block five.
What Automated Clustering Gets Wrong
Clustering software groups searches by measurable similarity. Sometimes that is word overlap and sometimes it is how much the results for two searches resemble each other. The second approach is considerably better and neither knows anything about your business.
What it cannot see. Your customer.
Whether two searches are the same job depends on what you sell and who buys it. A grouping correct for a national retailer can be wrong for a local firm.
The commercial blindness. Value.
Software will happily group a valuable search with a worthless one because they look similar. Nothing in the input tells it which produces work.
The over-splitting problem. Too many groups.
Strict settings produce dozens of tiny clusters, which becomes a plan for dozens of thin pages. That is the cannibalisation problem arriving by a different route.
The under-splitting problem. Too few.
Loose settings merge situations that needed separating. The business then writes one page attempting three jobs.
Why the settings cannot be right. No correct value.
The threshold that works for one subject is wrong for another. Somebody has to look at the output and correct it, which is most of the work anyway.
The fair summary. A first pass.
At scale it is a reasonable starting arrangement to be corrected by somebody who knows the business. It is not a finished map.
The Map It Produces
The output of clustering is a keyword map. A record of which page owns which group of searches, written down somewhere the business can find it. That document is the real product of keyword research. Very few businesses have one.
What each entry contains. Four things.
The subject in a sentence, the searches it covers, the page that owns it and one line saying why it is worth doing. Nothing more elaborate is needed.
Why the reason matters. Memory fails.
In six months nobody will recall why a subject was included. An unexplained list gets quietly ignored. A reasoned one survives a change of staff.
Where it should live. Somewhere shared.
Not in a supplier's system and not in one person's files. The business should hold it, because it outlasts whoever produced it.
What it is not. A publishing schedule.
The map says what exists and what should. Deciding the order and the pace is separate. our content strategy guide covers it.
What to include for pages that exist. The current owner.
Map your existing pages to clusters as well as your planned ones. That immediately reveals subjects with two owners and subjects with none.
What it does for a brief. Removes the guesswork.
A writer handed a subject, its searches and the situation the reader is in can produce a good page without further instruction. That is the difference between a map and a list. It is where the afternoon repays itself.
The format. Whatever you will maintain.
A document, a spreadsheet, a shared page. The format matters far less than whether anybody updates it.
Why The Map Outlives The Project
Without a map, a site publishing over several years has no way of knowing whether a new page competes with something it already has. The map is the only defence against a business quietly building pages that undermine each other.
What goes wrong without one. Overlap.
Somebody writes a page in year three about a subject already covered in year one. Neither writer knew. Now two pages serve one search.
What that costs. Divided strength.
The two pages split whatever standing the subject had. Our cannibalisation guide covers the diagnosis and the repair.
Why it happens so easily. Staff change.
The person who did the research leaves, the supplier changes, the reasoning goes with them. Only a written map survives that.
The check the map enables. One question.
Before writing anything, ask which cluster this belongs to and whether a page already owns it. Ten seconds, which prevents years of accumulated damage.
When to update it. On publishing.
Record the page against its cluster when it goes live. Retrofitting a map onto a site nobody documented is a considerably larger job.
What to do if you have no map. Build one backwards.
List existing pages, assign each a subject and see what collides. That exercise usually finds problems on its own.
When A Cluster Should Split
Groupings are not permanent. A cluster that seemed right can turn out to contain two things people want. Recognising that early saves rewriting later. Three signals tell you and the first appears while you are still writing.
The writing pulls two ways. The earliest signal.
You cannot draft the page without it wanting to become two. That resistance is the subject telling you it is really two subjects.
The results look different. The clearest.
Search the phrases in the group and compare what appears. Completely different results mean search engines already treat them as separate.
The unplanned queries disagree. The late one.
Once live, a page attracting a distinct family of searches it was not written for is carrying a second subject. Our guide to reading your own search data covers spotting that.
What splitting involves. Less than feared.
Usually taking a section out into its own page, then linking the two. The original page gets shorter and clearer as a result.
The opposite case. Merging.
Two thin pages serving one situation should become one good page, with the weaker address pointed at the survivor. That is the same repair from the other direction.
What not to do. Split on volume.
Separating pages because two phrases both look worth having, when a single page would serve both, manufactures competition between your own pages.
The review worth scheduling. Once a year.
Read the map against what the business now sells and against what your own search data shows. Subjects rarely move, though the groupings shift as customers change how they describe things. An hour keeps the document trustworthy.
Where this goes next. The plan.
A finished map becomes a build order. Everything sits on the keyword research guide.
Two searches
sharing every
word can want
opposite things.
Boiler replacement cost and boiler replacement grants share the subject entirely. One person is budgeting for a purchase and the other is checking eligibility for help. Sort a list alphabetically and they end up on the same page, which serves neither of them. Words correlate with intent loosely and they are not intent.
How we group a filtered list:
Before writing anything, ask which cluster it belongs to and whether a page already owns it. Ten seconds prevents years of damage.
Every guide.
One subject.
What the words mean, what the numbers in a keyword tool actually are, how to find the searches worth targeting, how to turn a list into a publishing plan and what to do when a page you relied on slips.
