What Is Keyword Research in SEO?
Knowing what people search tells you what to write about. It does not tell you where to put words, how often to repeat them or what proportion of a page they should occupy. Two of the industry's most persistent ideas depend on pretending otherwise.
Keywords Are An Input, Not A Recipe
Research tells you which subjects people want answered and in what language. That is genuinely valuable. What it does not provide is a set of instructions about where words go, which is what most keyword advice pretends to supply.
What research legitimately gives you. Subjects and language.
Which questions people are asking, roughly how many, plus the words they use to ask them. That tells you what to write and helps you write it in terms a reader would recognise.
What it cannot give you. A construction method.
No amount of search data tells you how many times a term should appear or where it should sit. Those are not questions the data answers. It never did.
Why the recipe idea took hold. Search once worked that way.
Early search engines counted term occurrences fairly literally, so a page mentioning something more often looked more relevant. Techniques built for that model outlived the model by decades.
What replaced it. Understanding rather than counting.
Search engines now assess whether a page covers a subject, using far more than the presence of a phrase. That change is why the formulas stopped working and why they are still sold.
What this page therefore is. Mostly a correction.
Two of the five subjects absorbed here are ideas that no longer describe anything. Saying so is more useful than restating them carefully.
What survives. Less than you would expect.
One genuine placement principle, one useful way of thinking about scope and one commercially important observation about longer searches. Those are blocks four, five and six.
Research For A Page
Before writing one page you need three things. How to find them is a discipline of its own, covered in our keyword research material rather than here, because this page is about what happens once you have them.
The search itself. One, named.
The specific thing somebody types that this page should answer. Not a topic and not a category. A search you could write down and recognise.
What that search wants. The intent behind it.
Whether somebody is learning, comparing or ready to act. A page mismatched to that will fail regardless of everything else, which search intent covers properly.
What already answers it. The current results.
What exists, what it covers and whether you can realistically better it. That assessment prevents most wasted writing. It takes minutes rather than tooling.
What you do not need. A list of phrases to insert.
Research tools produce related terms in volume. Those are useful for finding subjects and are not a checklist to work through inside the text, which is the misuse this page exists to correct.
Why the boundary is worth stating. The title is misleading.
This page is named for research because that is what people search for, while being about applying keywords to a page. The methodology lives elsewhere and the two should not be conflated.
Keyword Density
There is no target keyword density. There never was one published by anybody, the figures in circulation have no source other than each other, then writing to a proportion produces worse pages than writing normally.
What the idea claims. A proportion.
That a term should occupy some share of a page's words, then that hitting that share improves performance. Stated plainly, it is a claim that writing can be tuned by measurement.
Where it came from. Literal counting.
Search engines of thirty years ago assessed relevance substantially by counting term occurrences. In that environment the idea was not unreasonable. It described how those systems actually worked.
Why it stopped applying. The systems changed entirely.
Modern search assesses whether a page covers a subject, using context, related concepts, structure and a great deal more. Counting occurrences describes almost none of that.
The strongest evidence available. An absence.
No search engine has ever published a target proportion. Every figure quoted traces back to other guidance quoting a figure, which is a chain with nothing at the end of it. That is checkable by anybody.
What writing to a figure does. Damages the page.
Repeating a phrase to reach a proportion produces text that reads as written for software. A reader notices immediately, which costs you the thing the page existed for.
What to say to somebody quoting one. One question.
Ask where the number came from. Not whether it works, since that invites an argument. Its source. The answer is always another piece of guidance, so following that chain is the fastest way to settle it.
Why we state no figure at all. Even to refute it.
Publishing a number inside a correction puts it into circulation anyway, because people remember figures rather than context. Nowhere on this site will you find one.
Why it persists despite all that. It is measurable.
A proportion can be calculated, reported and shown to have improved. That makes it an appealing deliverable in a way that judgement about whether a page is any good never will be, which is the whole of its survival.
What the tools are measuring. Something real and useless.
Density calculators compute a genuine proportion accurately. The arithmetic is fine and the significance is invented, which is why the output looks authoritative.
Placement
It remains sensible for the subject to appear in the title, the main heading and early in the text. That is because those places describe the page rather than because of any formula. Beyond them there is no map.
The title. The page's own summary.
If the subject is not in the title, the title is not describing the page. That is the reasoning. It happens to satisfy the placement advice as a by-product.
The main heading. The same logic.
A heading naming the page's subject is a heading doing its job. Where the subject is absent, something has gone wrong with the heading rather than with an optimisation.
Early in the text. Confirming arrival.
Somebody landing on a page wants immediate confirmation they are in the right place. Naming the subject early is a reader courtesy first and a search convention second.
What that means for the rest. Nothing to follow.
There is no advice about the middle of a page, the closing paragraph or subheadings. Those are governed by whether the writing is any good rather than by placement.
Why the reasoning matters more than the rule. It generalises.
Somebody who understands that these three places describe the page can apply that anywhere. Somebody following a list cannot, then asks where else the phrase should go.
The version to avoid. Forcing it.
Contorting a title or opening sentence to accommodate an exact phrase produces awkward text for a benefit that does not exist. Describe the page and the words follow.
Primary Against Secondary
A page should have one main subject and may naturally cover related ones. That is a statement about scope rather than a hierarchy of phrases to insert at different frequencies.
What the useful version says. One subject, some neighbours.
This page is about applying keywords to a page. It necessarily touches research, intent and scope, because those are adjacent. That is one subject with natural neighbours.
What the unhelpful version says. A ranked list.
A main phrase to use often and supporting phrases to use less often, distributed deliberately. That is the density idea in different clothing, with the same problem.
Why the reframe is more useful. It is a planning decision.
Scope is decided before writing and determines what belongs on the page. A phrase hierarchy is applied afterwards to text that was already written, which is decoration.
How to identify the main subject. The one-sentence test.
If you cannot say what a page is about in one sentence, it has more than one subject. That is a scope problem and it is solved by splitting rather than by rearranging phrases.
What related coverage does. Happens by itself.
Covering a subject properly means addressing what a reader needs, which produces the neighbouring language naturally. That is block seven.
Long Tail
Longer, more specific searches. Fewer people make them, far fewer businesses compete for them, then the people making them are closer to acting. For most of the businesses we work with, this is the entire strategy.
What the term describes. Specificity.
A broad search names a category. A long tail search names a situation: a particular problem, in a particular place, with a particular constraint attached.
Why the volume is lower. Fewer people share a situation.
Many people search the category and few share the exact circumstance. That looks like a disadvantage in a research tool and is frequently the opposite.
Why competition is lower. Nobody wrote the page.
Broad searches are contested by everybody. Specific ones are frequently answered by nothing written for them, which means a modest page can win where an excellent one could not.
Why intent is higher. Specificity reveals progress.
Somebody describing their exact situation has moved past browsing. They are closer to a decision, which makes their visit worth considerably more than a general one.
Where this becomes a whole strategy. Trades and services.
For a specialist business, the winnable searches are almost all specific. Our trade guides are built on that observation. It is why they concentrate on situations rather than categories.
What it looks like in practice. More pages, each narrower.
Rather than one page for a broad term, several pages each answering a real situation completely. More work, though the only approach available to a business that cannot outcompete on the broad term.
Write For The Subject, Not The Phrase
Covering a subject properly produces the language naturally. That is the observation underneath every technique on this page. Understanding it makes the techniques unnecessary.
Why it works. Subjects have vocabulary.
Anybody genuinely explaining a subject uses the words that belong to it, because those are the words for the things involved. Nobody has to be told to include them.
What that means for the formulas. They describe a symptom.
Pages covering a subject well contain related terms. Somebody observed that and concluded terms cause the coverage, which reverses the relationship.
How to apply it. Answer the question completely.
Write what somebody needs to know, including the parts that are inconvenient. The vocabulary arrives because you are describing the thing.
What still needs attention. The title and the heading.
Those are the two places where deliberate wording matters, because they are summaries rather than prose. Block four covers why.
The test for whether you have done it. Read it aloud.
Text written for a formula sounds wrong when spoken, because the repetitions that look acceptable on a screen become obvious in the ear. That takes a minute and catches almost every instance.
Where this argument continues. The semantic material.
Our advanced material covers how search engines assess subject coverage rather than term matching, which is the fuller version of this block.
One Subject Per Page
The scope rule in block five has a failure mode in both directions. Too broad and a page competes for nothing. Too many pages on one search and they compete with each other.
The first failure. A page about everything.
Six services on one page, else a guide answering four unrelated questions. Nothing is the subject, so nothing is the best available answer to any search.
The second failure. Several pages, one search.
Pages written at different times, by different people, targeting the same thing. Each dilutes the others, so the site divides its own effort without anybody deciding to.
Why the second is harder to see. Every page looks fine.
Nothing about any individual page indicates a problem. It is visible only by looking at what the site as a whole targets, which almost nobody does.
Where it comes from. Publishing without checking.
Somebody writes a useful page without establishing whether one already exists. Reasonable each time and cumulative over years.
Where this is covered. Its own page.
Keyword cannibalisation covers detecting and resolving it, including which fix to prefer. The full series is on the on page SEO guide.
Ask where that
density figure
came from.
Every figure in circulation traces back to other guidance quoting a figure. No search engine has ever published one. If somebody has given you a target proportion to write towards, that is a chain with nothing at the end of it.
How we approach a page:
You will not find a density figure anywhere on this site, including inside a correction.
Every guide.
One practice.
What a page says about itself, headings, keywords, search intent, images, page types, user behaviour, credibility and how to review an existing site.
