How to Do Keyword Research
Most keyword research produces a list nobody ever uses. The method that works is subtractive rather than additive, because the difficult part is not finding searches. It is removing everything that looks like demand and is not.
What It Is
Working out what people actually search for, in what words, then deciding which of those are worth pursuing. Two halves. The second is where the value sits. Finding searches is mechanical. Deciding which deserve a page is a commercial judgement nobody else can make for you.
What it is not. Running a tool.
A tool produces raw material in minutes. That is the beginning of the exercise rather than the output of it, which is why so much delivered research is unusable.
What a finished piece looks like. Short.
A modest number of subjects, each with the searches it covers and a reason it is on the list. A file of thousands of rows is an input rather than a result.
Who it is really for. Whoever writes the pages.
Research nobody can act on has failed regardless of how thorough it was. The test is whether somebody could sit down and write from it.
How long it takes. Less than people fear.
For a local business with a handful of services, an afternoon of thinking and a couple of hours of checking. The scale problem belongs to large sites.
Where it goes wrong first. Block two.
Most of what any tool hands you is not your customer. Not knowing that is how a business ends up writing for people who will never buy.
Most Of What You Find Is Not Your Customer
Export a keyword list for almost any trade and a large part of it is somebody who will never buy from you. Across the trades we have researched in building this programme, we have routinely found between a third and a half of a raw export falls into that category. We state that as a range from our own client work rather than as an industry figure. We will not sharpen it.
Who is actually in the file. Six groups.
Researchers and students. People intending to do the job themselves. Jobseekers looking for work in the trade. Suppliers and manufacturers. Competitors checking the market. And people in entirely the wrong country.
The one that costs most. The do it yourself searcher.
They search the same words your customer does. A page written to serve them attracts people who by definition are not hiring anybody. It can look like a successful page for months.
Why the tool cannot tell. It sees strings.
Nothing in a keyword export distinguishes somebody about to spend money from somebody writing an assignment. That judgement requires knowing the trade.
How you spot them. Read the phrasing.
Words like how, why, diagram, salary, wholesale, course, training and parts are all signals somebody is not buying a service. So is anything asking what something is made of.
What removing them does. Shrinks the list usefully.
The remaining file looks disappointingly small and describes actual demand. That is the correct outcome and it is the point where most research stops short.
Why we publish the range at all. Expectation.
A business handed a list of thousands should know that a large proportion of it was never theirs, before anybody builds a plan on the total.
Start With What You Are Asked
Before opening anything, write down the questions customers ask you every week. On the phone, at the door, in emails, at the counter. That list is better starting material than any tool produces. It costs nothing.
Why it works. Somebody already searched.
A question asked on the phone was very likely typed into a search box first, either by that person or by somebody who never rang. The demand is demonstrated rather than estimated.
What it captures that tools miss. The real problem.
Customers describe symptoms rather than services. Nobody searches for a diagnostic assessment. They search for the noise their boiler is making.
Who should write the list. Whoever answers the phone.
Not the marketing supplier. The person fielding calls has the material. Half an hour with them is the most productive part of this exercise.
What to record. Their words.
Write down how the customer said it rather than translating into trade language. The phrasing is the valuable part and it is the first thing lost in translation.
How many you need. Fewer than you think.
Twenty genuine questions is a substantial content plan. Most businesses can produce that in a single sitting.
What this list becomes. Seeds.
Each question is a starting point for the checking in block four, which is where research earns its place.
Then Check How It Is Actually Phrased
Research is validation rather than the source of ideas. You know what your customers want. What you do not know is the words they use for it. People very rarely search using the language a business uses about itself.
The gap that catches everybody. Trade terms.
A business offers a comprehensive electrical installation condition report. The customer searches for an electrical safety check. Same service, different vocabulary. Only one of them is typed.
Why the business language persists. It is correct.
The formal term is accurate and it is what the certificate says. That is exactly why it feels wrong to abandon it. Abandoning it on the page is what gets you found.
What the checking reveals. Variations.
One question usually corresponds to several phrasings. Grouping those is the beginning of the subject the page will cover.
Where to check. Cheaply.
The search engine's own suggestions and the related searches at the foot of a results page are direct statements of what people type. Our free tools guide covers those properly.
The compromise on the page. Both, in order.
Lead with the customer's words and introduce the formal term afterwards. That serves the search and the credibility at once.
What not to do. Pick one and repeat it.
Choosing a phrase and forcing it through the page is the mistake our stuffing guide covers. Coverage produces the language.
Winnable
Whether a search is realistically winnable depends on your site as much as on the search itself. That single point is missed by almost every difficulty score, because a number attached to a search knows nothing about who is asking.
What actually determines it. Who is already there.
Look at what currently appears for the search. If it is national brands, directories and publishers, a small local site is not displacing them. If it is businesses like yours, it is available.
Why that beats a score. You can see it.
Two minutes looking at the results tells you more than any estimate. It costs nothing. Our metrics guide covers what those difficulty figures actually are.
What makes something winnable for a small site. Specificity.
Longer, more particular searches have fewer serious contenders. A place, a problem or a qualifier attached to a service narrows the field considerably.
The uncomfortable part. Your list shrinks again.
The searches a business most wants are usually the ones it cannot win yet. Accepting that is what makes the remaining list actionable.
What winnable does not mean. Easy.
It means possible with a good page rather than requiring years of standing. The page still has to be better than what is there.
Never publish a threshold. There is not one.
No difficulty figure below which a search is safe. Anybody offering you that number is offering a rule that ignores your site entirely.
Worth Winning
A search carrying a decision behind it is worth more than ten carrying curiosity. This is where most keyword strategies go wrong, because volume and commercial value frequently run in opposite directions and the larger number is the more attractive one to report.
What intent looks like. A situation.
Somebody with a problem now. A job to arrange. A supplier to choose. Their phrasing usually contains urgency, a location or a comparison.
What curiosity looks like. A question about the world.
How something works, why something happens, what something is called. Worth serving sometimes, though not the same thing as demand for your service.
Why the big terms disappoint. They are broad.
A large general search contains every kind of searcher at once, most of whom want something other than what you sell. A smaller specific one can be almost entirely buyers.
The signal worth trusting. Whether anybody advertises.
If advertisers are paying to appear for a search, somebody has established that money comes from it. That is real evidence. Our metrics guide explains why it is the most useful number available.
What to do with curious searches. Serve them separately.
They belong on informational pages rather than on the pages meant to produce enquiries. Mixing them is how a service page ends up reading like an encyclopaedia.
The test to apply. Would you take a hundred.
If somebody offered you a hundred visitors from this exact search tomorrow, would that be good news. If the answer is no, it fails this filter.
The Two Filters Together
Winnable and worth winning. A search failing either is not a target, however large the number beside it looks. That sentence is the whole method. Applying it strictly is what separates a usable list from a spreadsheet.
The four possibilities. Only one is a target.
Winnable and valuable is where you work. Winnable and worthless is a trap that produces traffic without enquiries. Unwinnable and valuable is a long term ambition. Neither is simply noise.
The trap worth naming. Easy and pointless.
These are the searches that make a report look productive. Rankings appear, traffic arrives and nothing happens commercially, which is the pattern behind most disappointing engagements.
What to do with the ambitions. Keep them, separately.
Searches worth having but currently out of reach belong on a second list, revisited once the site has standing. Deleting them loses the direction.
Applying it in order. Value first.
Filter for commercial worth before difficulty. Otherwise you spend time assessing the winnability of things you would not want anyway.
What survives. Less than expected.
A short, defensible list of subjects with reasons attached. For a local business that is frequently under a dozen, which is a plan rather than a disappointment.
How Many Should You Target
One subject per page, expressed in whatever words fit naturally. No count, no target number of phrases per page and no recommended list length. The question itself comes from an older model of how search engines worked, which is why it has no useful answer.
Where the question comes from. Exact matching.
When engines matched strings, a page could be aimed at a specific number of phrases. Counting made sense then. It has not for a long time.
What replaced it. Subjects.
A page covering a subject properly is matched to every search that subject answers, which is usually far more phrases than anybody would have listed.
Why a count is actively harmful. It shapes the writing.
A writer given five phrases to include produces a page built around inserting five phrases. The reader can tell. Our stuffing guide covers what that costs.
What to hand a writer instead. The subject and the searcher.
What the page is about, who is arriving and what they need answered. That produces a better page than any phrase list.
The number that does matter. Pages.
How many subjects the business will cover, plus in what order. That is a resourcing question with a real answer, unlike the phrase count.
What we would tell a small business. Start with a handful.
Cover a few subjects completely before adding more. A half finished set of twenty performs worse than a finished set of six.
Group Before You Plan
A filtered list is still a list. Nothing useful happens until the surviving searches are grouped into subjects and each subject assigned to a page. That step is where most keyword research stops, one move before it becomes valuable.
The grouping test. One page, all of them.
Phrases belong together when a single page would satisfy anybody making any of those searches. That is the only test worth applying.
The error to avoid. Grouping by wording.
Two phrases sharing words can want completely different things, while two sharing none can want the same. Sorting alphabetically produces pages that serve nobody.
What the grouping produces. A map.
A record of which page owns which group. Our clustering guide covers the exercise and the document it produces.
Why the map outlives the research. It prevents drift.
Without one, a business publishing over years has no way of knowing whether a new page competes with something it already has.
What comes after. Sequencing.
Deciding which subjects to build first, which is a separate judgement covered in our content strategy guide.
The practical shortcut. One document.
Keep the map, the reasons and the ordering in a single place that survives the project. That is what the build then runs from.
It Is Never Finished
Keyword research is usually done once, filed and never opened again. Demand shifts, businesses change what they sell and your own data eventually contradicts the assumptions the original research rested on.
What triggers a revision. Three things.
The business changing what it offers or where. A subject being fully covered. Or enough real search data accumulating to correct the guesswork.
What does not trigger one. The calendar.
Redoing research quarterly because it is on a schedule produces churn rather than insight. There is no interval that makes sense in the abstract.
The most valuable correction. Your own reporting.
After a few months the searches your site actually appears for are better evidence than any estimate. Our guide to using search reporting for research covers turning that into new targets.
What usually changes. Not the subjects.
The subjects a business should own rarely move much. The phrasing, the emphasis and the order do, which is a lighter revision than starting again.
When to genuinely start over. A different business.
A new service line, a new area or a change of customer. Those warrant fresh research rather than an amendment.
Where the maintenance sits. Content strategy.
Our content strategy guide covers keeping the plan alive. Everything sits on the keyword research guide.
Between a third
and a half of
that list is not
your customer.
Researchers, students, people planning to do the job themselves, jobseekers, suppliers and competitors. Nothing in a keyword export distinguishes somebody about to spend money from somebody writing an assignment. The do it yourself searcher types the same words your customer does.
How we build a target list:
The surviving list looks disappointingly short. That is the correct outcome. It is also where most research stops one step early.
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.
