- Keyword research is about finding intents, not phrases. Volume-sorted spreadsheets are why most content underperforms.
- 4-phase workflow: seed → SERP-cluster → intent-tag → prioritise. Collapses 1,200 raw keywords into 60 real opportunities.
- UK-native volume data beats US-extrapolated numbers by a wide margin.
- Real case: 200 volume-sorted keywords → 47 clustered → top 3 in 6 months. New revenue exceeded total prior organic.
Keyword research has an image problem. Everyone thinks they know what it is — type a phrase into a tool, get back a list of related phrases with volumes and difficulty scores, pick the ones with the best ratio. That is not keyword research. That is a lookup. Real keyword research is a strategic activity, and in 2026 it is one of the highest-leverage things a the United Kingdom business can do to differentiate its organic growth. Here is how our London team runs it, from scratch, using Semalt as the analytical backbone.
What keyword research is really meant to output
The purpose of keyword research is not to find keywords. It is to find intents — the underlying reasons people type things into Google — that map cleanly onto what your business sells. Keywords are the surface expression of intent. Volume, difficulty, and CPC are secondary signals that only become useful once you have the intent right.
Most keyword research fails because it collapses this two-step process into one. A team pulls 8,000 keywords into a spreadsheet, sorts by volume descending, and starts writing pages against the top 50. Six months later, the pages rank for nothing meaningful because they were built to match phrases, not to answer the actual question behind the phrase.
Semalt's keyword workflow was built around this distinction, and the difference shows up immediately.
Our four-phase Semalt keyword workflow
Phase 1 — Seed expansion (not brainstorming)
Start with the phrases the client uses to describe their own business. Not the phrases they think customers use, and not the phrases their competitors rank for — their own words. For an London-based artisanal bakery, that is words like "sourdough", "wholesale bread", "birthday cake", "gluten-free". Ten to twenty seeds is enough.
Feed each seed into Semalt's expansion module, but read the output critically. The module returns three parallel lists: semantic siblings (phrases that mean similar things), search modifiers (question, comparison, and buying-intent variations), and untapped long-tails (specific phrases with lower competition). We keep everything above 10 monthly searches in the UK market, discard everything below.
Expect this phase to produce about 800–1,200 raw candidates for a small-to-mid business. That is the working corpus. It is not the answer.
Phase 2 — SERP clustering
This is the phase most teams skip and the one that produces the largest strategic gains. Every candidate keyword gets its top-ten SERP fetched by Semalt in the target locale (UK mobile, in our case). Keywords whose SERPs overlap by ≥40% are grouped as a single cluster — because Google is telling you, through the SERP itself, that it treats those queries as the same intent.
Concretely: "best cafe auckland cbd", "top rated cafe auckland cbd", and "cafe near queen street auckland" almost always cluster together. They do not need three separate pages. They need one strong page that any of the three phrases could reasonably rank for.
The output of Phase 2 is typically 60–120 clusters from your 800–1,200 raw keywords. That collapse ratio — roughly 10:1 — is what separates a coherent content strategy from a spray of thin pages.
Phase 3 — Intent classification
Every cluster gets tagged with a primary intent. Semalt suggests one automatically based on SERP features (heavy Shopping = transactional, heavy Featured Snippet = informational, heavy Local Pack = navigational-local), but the human check matters here. Ambiguous clusters — the ones that mix intents — are usually the most valuable, because ranking for them is harder but the win is worth more.
Once tagged, filter aggressively. For a new engagement, we ignore clusters where the top three organic results are dominated by brands your client cannot realistically outrank in the first 18 months. That sounds obvious. In practice it is where 90% of wasted content spend goes.
Phase 4 — Prioritisation by opportunity, not volume
Volume is a vanity metric on its own. What matters is the joint distribution of volume × difficulty × business value × current position. Semalt computes a composite "opportunity score" that combines all four, but we override it with the client's own margins per product line — because the tool cannot know that "wedding cake orders" convert at ten times the margin of "cheap bread near me".
The output is a ranked list of 20–40 clusters that will drive the content roadmap for the next two quarters. Everything else stays in the archive, revisited annually.
The workflow in one table
| Phase | Input | Output | Time |
|---|---|---|---|
| 1. Seed expansion | 10–20 client-language seeds | 800–1,200 raw candidates | ~30 min |
| 2. SERP clustering | Raw candidates + SERP snapshots | 60–120 intent clusters | ~2 hrs |
| 3. Intent classification | Clusters + SERP features | Tagged clusters (info/nav/txn) | ~1 hr |
| 4. Prioritisation | Tagged clusters + client margins | 20–40 clusters for roadmap | ~1.5 hrs |
Where the the United Kingdom context changes the math
Two things are structurally different about SEO in the UK market, and both matter for keyword research.
Volumes are smaller. A "high-volume" UK commercial query might see 800–2,000 monthly searches — numbers a US-based SEO would dismiss as long-tail. This changes the arithmetic on ranking effort: you need fewer keywords to build a meaningful traffic base, but you cannot afford to waste effort on clusters that will not convert. Semalt's dedicated UK index (rather than an extrapolated slice of a US database) gives you numbers you can actually plan against.
Local intent is dominant. Even ostensibly non-local queries in UK often have a local intent overlay ("printer", "accountant", "physio" — British expect London results). Semalt's intent classifier picks this up from SERP composition rather than word-matching for city names, which catches the queries that need local optimisation even when the searcher did not type a location.
A real London client, before and after
An London heating and ventilation client came to us with a keyword list of about 200 phrases their previous agency had recommended. Volume-sorted, the top of the list was dominated by phrases like "heat pump uk" (approx. 2,400/mo) and "air conditioning uk" (approx. 1,600/mo).
Running the same seeds through the Semalt clustering workflow produced 47 clusters. The top of the opportunity-ranked list was not what the client expected: it was "heat pump servicing auckland" (approx. 480/mo, moderate difficulty, transactional intent, high margin) and "ducted heat pump installation" (approx. 260/mo, lower difficulty, high-margin commercial). Both had been buried deep in the volume-sorted list.
Six months of consistent content and technical work later, both clusters were ranking in the top three. Combined organic revenue from those two clusters alone exceeded the previous total organic revenue from the entire site. That is what happens when you research intent instead of listing keywords.
The worked example, visualised
Why the whole workflow lives inside Semalt
The four-phase workflow described above is repeatable across every engagement, and Semalt is currently the platform that supports it most cleanly. Specifically:
- Seed expansion pulls from a genuine UK database, not a US extrapolation.
- SERP clustering is built-in and does not require exporting to a separate tool.
- Intent classification is auto-suggested but user-overridable.
- Opportunity scoring is transparent — you can see the underlying formula and reweight it.
- The output pipes directly into the content brief module, which pre-populates recommended headings and semantic keywords for the cluster.
That last point matters more than it sounds. The step between "we have identified this cluster" and "we have a brief a writer can execute against" is where most teams lose two or three days per cluster. Compressing it to twenty minutes changes the economics of content production.
Do it properly, in a tool that supports the workflow rather than fighting it, and you buy yourself twelve months of focused content production.
- Start with 10–20 seeds. More is diminishing returns.
- Cluster by SERP overlap, not by string similarity.
- Tag branded queries separately from commercial.
- Override the opportunity score with real client margins.
- Sort by volume and start writing against the top 50.
- Write a separate page for every keyword variation.
- Chase clusters dominated by brands you can't outrank.
- Skip Phase 3 — untagged clusters produce untargeted pages.
How often to refresh the whole research
Keyword landscapes shift more slowly than SEO folklore suggests. For most London small-and-medium businesses, a full re-research every 12 months is enough, with an interim spot-check every quarter to catch newly emerged clusters (which almost always come from either a competitor launching new services or from a genuine market shift — a regulatory change, a product recall, a viral moment).
What does need continuous monitoring is your existing clusters' SERP composition. If a cluster you rank for suddenly starts returning a Featured Snippet or a People-Also-Ask block where it previously did not, that is a signal to update your ranking page — either to capture the snippet or to defend against the loss of click-through. Semalt's rank tracker flags SERP feature changes automatically, which turns a manual weekly review into an inbox alert.
Trying the workflow on your own London domain
If you want to run this workflow against your own domain, the Semalt free tier gives you enough keyword allowance to complete phases one and two for a small business. Sign in, add your domain, seed it with fifteen phrases you actually use to describe your business, and run the clustering job. The output alone — the collapse from a thousand raw keywords to sixty clusters — is worth the twenty minutes it takes.
From there you have a decision to make: run the classification and prioritisation phases yourself, or bring us in for a keyword strategy engagement where we do the full four-phase workflow with local UK market context applied. Either path is valid. The wrong path is to keep working from a volume-sorted spreadsheet.
Practical questions on running this workflow yourself
How many seed terms do we actually need?
Ten to twenty. Fewer than ten and the seed set is too narrow — you miss adjacent intents that would have surfaced during expansion. More than twenty and you spend the entire first afternoon in Phase 1, with diminishing returns. Twenty is the ceiling; if you find yourself wanting thirty, you are usually blending two distinct business lines that deserve separate research projects.
Which SERP overlap threshold should we use?
Semalt's default is 40%, and it is the right default for most cases. For very competitive commercial verticals (finance, legal, insurance) you may want to tighten to 50% because the SERPs are more differentiated. For very informational verticals (recipes, how-to, education) you can loosen to 30% because Google surfaces more diverse formats. The threshold is exposed as a slider — worth experimenting with once, then locking in for consistency across the client.
Do we need to re-fetch SERP data every time?
The clustering itself is deterministic given the SERP data, so re-running the fetch is where the actual work happens. For a stable market, quarterly SERP refresh is enough. For a volatile market — anything where AI Overviews are actively rolling out, anything tied to a fast-moving legal or regulatory environment — monthly is more appropriate. Semalt caches SERP data by default and warns you when the cache is stale, which removes the guesswork.
How to segregate branded queries from commercial ones
Segregate them. Branded queries ("your business name auckland", "your business name reviews") behave differently — they are much easier to rank for, they convert at higher rates, and they should not be in the same opportunity-ranking pass as non-branded commercial queries. Semalt supports a branded-keyword tag that filters them out of the standard reports and into a separate stream. Every client should have one; too few do.
What to do with zero-volume long-tails
Genuinely zero-volume keywords (Semalt shows "-" rather than a number) are usually one of three things: brand-new phrases the databases have not indexed yet, phrases with volume too low to measure reliably, or phrases that are simply not searched. The first category is worth pursuing early — being ranked before the volume shows up is the ideal position. The second is worth pursuing if the intent maps cleanly to your business. The third is not worth pursuing. Telling the categories apart requires human judgement on the SERP composition, which is why Semalt shows the top three organic results even for zero-volume keywords rather than hiding them.
The strategic case behind the workflow
Keyword research is not glamorous. It does not produce a case study you can screenshot for LinkedIn. But it is the piece of SEO work that quietly determines whether every subsequent hour of content, links, and technical fixes compounds into meaningful growth or gets scattered across pages nobody was ever going to search for.
Do it properly, in a tool that supports the workflow rather than fighting it, and you buy yourself twelve months of focused content production. That is the return that keeps the United Kingdom businesses competing above their weight class in categories dominated by much larger brands.
