Intent clusters
Terms grouped by what the searcher wants rather than by shared words, so one page targets one cluster instead of dozens of near-duplicate pages.
Why volume is the least useful column
Because it says nothing about whether you can win the term or whether the searcher buys anything. A high-volume head term you cannot rank for is worth nothing, and a term with no measurable volume that three buyers search each month can be worth more than everything above it.
The export problem is worth naming directly. Most keyword research delivers thousands of rows sorted by volume, and the client is left to decide which matter. That is the hard part being handed back, and it is why so many keyword deliverables are opened once and never used again.
Winnability has to be assessed before anything else, because it removes most of the list. If the results page is dominated by publishers with authority you will not match this decade, that term is not a target however attractive its volume. Establishing that early saves a year of content aimed at nothing.
Clustering matters more than the individual terms. Search engines resolve intent, not strings, so dozens of phrasings resolve to one page. A flat list of two thousand keywords implies two thousand pages; the same research clustered properly implies perhaps sixty, which is a completely different content plan.
The last step is ownership. Each cluster needs exactly one page assigned to it, or you rebuild the cannibalisation problem from scratch as content gets produced. Research that stops before this point reliably produces a site competing with itself within eighteen months.
What the engagement produces
The engagement produces clusters rather than a term list, a winnability assessment filtering out what you cannot realistically rank for, intent classification per cluster, an assignment of one owning page to each, and a ranked production order based on commercial proximity rather than volume.
Winnability filter
Which clusters you can realistically rank for given your current authority, assessed from who holds the results rather than from a difficulty score.
Intent classification
Informational, commercial or transactional per cluster, since the page type that wins differs and a mismatch cannot be optimised away later.
Page ownership map
One page assigned per cluster, existing or planned, so production has a destination and the site does not start competing with itself.
Commercial ranking
Production ordered by proximity to a buying decision rather than by volume, which routinely inverts the order a volume-sorted export implies.
Existing coverage audit
Which clusters your current pages already serve, so effort goes to genuine gaps rather than producing something that competes with what you have.
How the engagement runs
The engagement runs over four to six weeks: gather terms from your own conversations as well as the tools, filter aggressively for winnability, cluster by intent, map clusters against existing coverage, then assign one owning page each and order production by commercial value.
- 01
Gather beyond the tools
Sales calls, support tickets and answer surfaces alongside keyword tools, because the phrasing buyers use with people is closer to what they type than tool output.
→ A raw term corpus from multiple sources
- 02
Filter for winnability
Who currently holds each results page and whether you could plausibly displace them. This removes most of the list and prevents a year aimed at nothing.
→ A winnability-filtered term set
- 03
Cluster by intent
Group by what the searcher wants rather than by shared words. Two thousand terms usually resolve to a few dozen clusters, which changes the content plan entirely.
→ Intent clusters with representative terms
- 04
Map against existing coverage
Which clusters your current pages already serve, well or badly. Frequently reveals that the priority is improving what exists rather than producing anything new.
→ A coverage map with gaps identified
- 05
Assign owners and order
One page per cluster, and a production order set by commercial proximity. Research that stops before assignment reliably produces cannibalisation later.
→ An owned, ordered production plan
What a keyword deliverable should contain
The difference between a useful deliverable and an export is whether the hard decisions have been made. A list of terms sorted by volume hands the judgement back to the client, which is precisely the part they were paying to have done.
| A typical export | A useful deliverable | |
|---|---|---|
| Structure | Thousands of rows, flat | A few dozen intent clusters |
| Sorted by | Search volume | Commercial proximity and winnability |
| Winnability | A difficulty score | Who holds the results and whether you can displace them |
| Implied page count | One per row, implicitly | One per cluster, explicitly |
| Ownership | Undecided | One page named per cluster |
| Who decides | You, afterwards | Decided as part of the work |
Signals you need this now
You need this when you hold a keyword export nobody has used, when content is commissioned by volume, when several pages target the same term, or when your research predates conversational search and no longer reflects how buyers phrase things.
- You hold a keyword export nobody has opened since delivery
- Content is commissioned on volume rather than winnability
- Several pages target the same term
- Your research predates conversational search
- Nobody can say which page owns a given term
- You are ranking for terms that never convert
- You are planning content for the next two quarters
What the research typically changes
The most common outcome is a much shorter content plan than clients arrive expecting. Aggressive winnability filtering and intent clustering routinely reduce a two-thousand-row export to a few dozen clusters, most of which map onto pages that already exist and need improving.
Where this fits
Related SEO services
The adjacent disciplines
Deciding between options
Questions about keyword research services
Each answer is written to stand alone in 40 to 60 words — the shape an AI Overview or Perplexity citation lifts. Ships with FAQPage schema.
Why is search volume not the main criterion?
Because it says nothing about whether you can win the term or whether the searcher buys. A high-volume term you cannot rank for is worth nothing, and a low-volume term three serious buyers search monthly can outperform everything above it.
How do you decide what is winnable?
By looking at who currently holds the results page and whether you could plausibly displace them, rather than by trusting a difficulty score. If the page one is publishers with authority you will not match this decade, it is not a target.
Does this mean we need a page per keyword?
The opposite. Engines resolve intent rather than strings, so dozens of phrasings resolve to one page. A properly clustered set of two thousand terms usually implies a few dozen pages, not two thousand.
How is this different from prompt research?
Keyword research covers what people type into search engines, with volume data available. Prompt research covers what people ask assistants, in full sentences with no volume data. They overlap in language and differ in surface and measurement.
Will you give us the raw keyword export?
Yes, alongside the clustered deliverable — it is your data. But the export is the input rather than the output. Handing over rows sorted by volume returns the judgement you were paying to have made.
How often should keyword research be redone?
Every eighteen months or so for most categories, sooner if your product or market shifts. Buyer language changes more slowly than tools imply, so annual rewrites are usually unnecessary — a coverage review is more useful than fresh research.
Find out whether this is your constraint.
Thirty minutes with a senior strategist. We pull your live visibility while we talk and tell you plainly whether a keyword research services is what you need — or whether your problem sits somewhere else.