Keyword Research & Strategy
A keyword list is not a strategy, and it is a poor deliverable
The spreadsheet with five thousand rows is easy to produce and almost impossible to act on. What changes anything is a map: which intents matter commercially, which single page owns each one, and where your existing pages are already fighting each other.
The argument
The deliverable people expect is a spreadsheet of five thousand phrases. The deliverable that changes anything is a map saying which page owns which question.
A list is a description of demand. A map is a set of decisions about your site: this page exists, it answers this, it is the definitive one, these other pages support it, and nothing else is allowed to compete with it.
The difference shows up immediately in what happens next. A list gets forwarded to a writer who picks a row. A map gets used by whoever is building the navigation, whoever is briefing the content, and whoever has to decide in eight months whether a proposed new page should exist at all.
Why lists fail
What goes wrong when the research stops at a spreadsheet
- Nobody knows which page a row belongs to.
- Without an assignment, every writer makes their own judgement, and six months later three pages exist for one question. The row is not the unit of work. The page is, and a row that has not been assigned to a page has not yet been made useful.
- Priority was set by volume.
- The largest terms are usually the least commercially specific and the most contested. A phrase with a fraction of the volume but obvious purchase intent is frequently worth several times more, and a list sorted by estimated volume actively hides that.
- The head term has no plausible owner.
- Some phrases are answered by the result page itself, by marketplaces, or by publications you cannot displace. Putting them at the top of a plan commits budget to a fight nobody has assessed. Checking what actually appears for a term is part of the research, not an afterthought.
- Existing pages were never audited.
- Research done as though the site were empty produces recommendations to create pages that already exist in some form. The overlap between what is proposed and what is published is the most valuable part of the analysis and the part most often skipped.
Reading the modifiers
Phrases that look related and are not
These examples use one industry for clarity. The pattern generalises: the modifier reveals the job, and the job decides which page should own the phrase.
Wants to understand the category
Belongs on an explanatory page. Sending this to a sales page produces a bounce and teaches you nothing about demand.
- what does a bookkeeper do
- bookkeeper vs accountant
- do i need a bookkeeper
- bookkeeping basics for small business
Wants to buy the service
Belongs on the service page, and only one page should be the answer. This is where cannibalisation does the most commercial damage.
- bookkeeping services
- bookkeeper for small business
- monthly bookkeeping service
- hire a bookkeeper
Wants a price
Wants numbers, ranges and the variables that move them. A page that refuses to discuss price will lose to one that does, however good it is otherwise.
- bookkeeping cost per month
- how much do bookkeepers charge
- bookkeeping fees small business
- average bookkeeping rates
Wants a specific solution
Each of these is a candidate for its own page, but only where you genuinely do that work and can write something specific about it.
- xero bookkeeper
- catch up bookkeeping behind on books
- bookkeeper for ecommerce sellers
- construction bookkeeping services
These are examples of how customers in this market search, drawn from keyword research and from the questions that come up on sales calls. They are illustrative, not a volume claim — the actual demand in your area is something we size before recommending anything.
On the numbers
Where volume figures come from, and what each one is
Nobody outside Google counts searches. Every published figure is a model, and knowing which model you are reading changes how much weight it deserves.
| Dimension | What it actually is | What limits it |
|---|---|---|
| Google Keyword Planner | Advertising planning data from Google, grouped into ranges and intended for budgeting ad campaigns. | Bucketed and rounded, close variants are grouped together, and the numbers behave differently depending on account spend. |
| Third-party SEO tools | Models built from clickstream panels and other signals, extrapolated to a national figure. | Panel composition varies by tool and country, which is why two tools disagree and neither is simply wrong. |
| Search Console impressions | The only first-party data here: how often your pages were actually shown, for which queries. | Limited to queries you already appear for, anonymised at the long tail, and shaped by where you currently rank. |
| Autocomplete and related searches | Evidence of real phrasing and of questions people follow up with. | No volume attached at all, and personalised and localised in ways you cannot see. |
| Your own site search | What visitors could not find using your navigation, in their own words. | Only reflects people already on the site, so it tells you nothing about demand you never reached. |
How we work
What we hand over, and what we will not pretend
We will
- Deliver a page-level map with one nominated owner page per intent
- Show the working, including terms we considered and rejected and why
- Present volume as ranges and relative ordering rather than as counts
- Check what actually appears for a term before recommending you pursue it
- Tell you when demand for something you sell is too small to build a plan around
We will not
- Hand over a five-thousand-row export and call it a strategy
- Recommend a separate page for every phrasing of the same question
- Build a traffic forecast by multiplying estimated volume by an assumed click rate
- Sort a plan by volume and present the ordering as commercial priority
- Ignore the pages you already have and write the research as though the site were new
Method
From raw demand to a map somebody can build from
Harvest from several sources
Tool exports, Search Console, site search, autocomplete, the questions your sales team is asked, competitor coverage, and the language used in reviews and support tickets. Tools alone miss how customers actually speak.
You get: A raw demand set with sources recorded
Cluster by intent, not by string
Group searches that want the same outcome, then name the dominant intent for each cluster in a sentence. Phrases sharing no words often belong together; near-identical phrases sometimes do not.
You get: Named clusters with the dominant intent stated
Check what the results are made of
For each cluster that matters, look at what is currently shown: the format, who occupies it, how much of the page is advertising or features, and whether a company page can plausibly compete there at all.
You get: A viability note per priority cluster
Assign one owner page per intent
Existing page, page to be improved, or page to be created. Everything else in the cluster becomes a supporting page or a section, and is explicitly forbidden from competing with the owner.
You get: The map: intent, owner page, supporting pages
Audit the site for overlap
Find where two of your pages are already plausible answers to the same question, and resolve each case by merging, differentiating or demoting. This is usually where the fastest gains are.
You get: A cannibalisation register with a decision for each conflict
Order the work by value
Weighted by commercial value, existing position, competitive difficulty and effort — with the reasoning visible so you can disagree with the ordering rather than having to accept it.
You get: A prioritised roadmap and a maintainable working file
Questions
What clients ask about research and mapping
How accurate are search volume numbers?
They are estimates, and they should be read as ranges rather than counts. Tools model volume from sampled clickstream data and from advertising platform figures that are themselves bucketed and rounded, which is why two respected tools can differ substantially for the same phrase.
We use them for relative ordering — is this bigger than that — and never as a forecast. Anyone presenting a traffic projection built by multiplying an estimated volume by an assumed click-through rate is presenting arithmetic, not evidence.
What is keyword cannibalisation, in practical terms?
It is when two or more of your own pages are plausible answers to the same question, so neither becomes the obvious one. You usually see it as positions flickering between two addresses for the same query over weeks.
The remedy is a decision rather than a technical change: pick which page is the answer, rewrite the other to serve a genuinely different intent, or merge them. Adding a canonical tag without changing the content just hides the argument.
Can one page target several keywords?
Yes, and it should. A page that serves one intent well will be shown for many phrasings of that intent, including ones nobody thought to research.
What a page cannot do is serve two different intents at once. A page trying to define a term, compare five products and sell a service will do all three worse than three pages would.
Do we need this if we already have a keyword list?
It depends on what the list is. If it says which page owns which intent, what that page has to do, and how the pages relate to each other, you have a map already and you may not need us.
If it is an export of phrases with volume and difficulty columns, then the work of turning it into decisions has not been done yet. That work is the part that takes judgement.
How long does this take?
Typically two to four weeks depending on the size of the site and the number of markets. Harvesting demand is fast; clustering, checking what the results actually contain, and auditing the existing site for overlap is where the time goes.
For a site being rebuilt, we would rather do it before the information architecture is signed off than after.
Get the map before you commission the pages
If a rebuild or a content programme is coming, this is the cheapest point at which to be right about structure. Tell us what you sell and where, and we will tell you what the research would need to cover.
If we don't deliver the work we agreed to deliver for reasons within our control, you don't pay for the undelivered work. Read our guarantee
Related
Where to go next
- turning the map into pagesThe map decides what exists; the craft work decides how well it performs.
- auditing what is already publishedCannibalisation usually surfaces here first.
- testing demand with paid searchThe fastest way to find out whether a term converts is to buy it for a month.
- the same exercise across languages
- building the structure the map implies
- where each channel earns its place
Last updated · Reviewed by Zubair Afzal