Most keyword research produces a spreadsheet nobody uses. Someone exports a few hundred phrases from a volume tool, sorts by search volume, highlights the top twenty, and the project stalls there, because a volume figure does not tell you what to publish. Growth Bully, a Malta performance marketing agency, runs this exercise mostly on businesses that already spend money on search, and the pattern repeats: the evidence they needed was sitting in their own account, unread.
A keyword list is not a research output. It is a publishing plan, and it is only finished when every phrase on it has a page assigned and a reason to be built first.
What is keyword research actually for?
It is for deciding which pages to build and what each one has to answer. The output is not a list of phrases, it is a map of searcher jobs to pages, ordered by commercial value. Any research that ends in a sorted spreadsheet has stopped one step short of being useful.
That reframing changes what you collect. You stop hunting for the biggest numbers and start looking for repeated wording, because repeated wording is what tells you a page is needed. A phrase with modest demand that nobody has answered properly is worth more than a popular phrase answered by ten stronger sites.
Where should a keyword list come from when volume data is thin?
From evidence you own before estimates you rent. Four sources beat any tool in a small market: your search terms report, the questions your sales team answers every week, your site search box, and the wording of enquiries that arrived and converted. All four are observed demand rather than modelled demand.
Work through them in this order, because the first one is already paid for:
- Your search terms report. Every query that triggered one of your ads is a real search by a real person, with a click cost and sometimes a conversion attached. No tool can match that. The same report drives your negative keyword list, so you are reading it anyway.
- Sales and support conversations. Write down the exact phrasing of questions people ask before buying. Those phrasings become question headings, and they read as human because they are.
- Your site search and contact forms. People tell you what they could not find. Each unanswered search is a page brief with the demand already proven.
- The wording of enquiries that closed. Not just what people searched, but how buyers described the problem. That language belongs in headlines, not in a backlog.
Tools still earn their place after this, for discovering phrasings you never thought of and for checking what already ranks. They are the second pass, not the first.
How do you tell a buyer search from a browsing search?
By what the searcher has to do next. A buyer search names the thing being bought and usually adds a qualifier: a location, a brand, a price, the word service, or a question about who can do it. A browsing search asks what something is, or how to do it without help.
Both are worth having, but they are worth different pages. Buyer searches belong on service and industry pages that can take an enquiry. Browsing searches belong in articles that answer the question honestly and then show the reader what doing it properly involves. Putting a browsing search on a service page wastes the page, and putting a buyer search in an article wastes the demand.
How do you group keywords without building duplicate pages?
Group by the answer, not by the words. If two phrases would be satisfied by the same page, the same headline and the same proof, they are one target and belong together. If either needs a different page to be satisfied, split them. Similarity of wording is not the test.
Neither is how the phrases would be matched in an ads account, which is a delivery setting rather than a content decision. Duplication is the most expensive mistake in the whole exercise. Two pages written for the same job compete with each other, split whatever authority they earn, and leave search engines to pick one, usually the weaker one. One page per job, every time, and a clear note of which page owns which group.
Which keyword do you build first?
The one with the highest value for the effort, which is rarely the one with the most volume. Score each group on four things: how ready the searcher is to buy, whether you can answer better than whatever ranks now, what one customer from that search is worth, and how much work the page actually is.
- Read the current first page of results for the main phrase. If every result is thin, outdated or beside the point, that is your opening.
- Ask whether you can answer it from genuine experience. If you cannot, the page will be generic and will deserve to lose.
- Multiply roughly: expected enquiries against what a customer is worth. A low demand phrase attached to a high value service beats a popular phrase attached to nothing.
- Estimate the build honestly, including the proof, the images and the internal links. A page you will not finish scores zero.
- Commit the order in writing and work it top down, so nobody renegotiates the plan every week.
Then brief the page properly before anyone writes a word. The grouping work you just did is most of a brief already, and the brief is what keeps the page on target.
How do you know the research worked?
When pages start earning impressions for phrases you never listed. That is the signal that your grouping matched how people genuinely search. Check which queries each page attracts rather than only its positions, and move anything the wrong page is answering to the page that should own it.
Review quarterly, not weekly. Search behaviour in a small market moves slowly, and the discipline that pays is keeping the map current, retiring groups that never produced, and merging pages that drifted into answering the same job. The wider approach sits in how search works for a small local market, and the paid side of the same evidence runs through our Google Ads work and across every digital marketing programme we run.
If you already spend on search, the keyword research you need is mostly done and unread. Send us the account and we will show you which pages your own search data says to build.