What Is a Keyword in SEO? Types, Intent and How to Use Them

A keyword in SEO is simply the word or phrase somebody types or says into a search engine. The term is a holdover from an era when searches really were one or two words; today most searches are full phrases or questions, which is why professionals increasingly say "query" instead. A keyword matters for one reason: it is the only evidence you have of what a potential customer actually wants, in their own words. Everything else in SEO follows from choosing the right ones and giving each one a page that answers it properly. Choosing keywords is not about collecting a long list. It is about reading the intent behind each phrase and matching it to the right kind of page.

This page covers what keywords are, the types worth distinguishing, how to read intent, what the volume numbers mean, and the one rule that prevents the most common and costly mistake.

The types of keyword, and why each label exists

The long tail is where most sites should start, and the reason is arithmetic rather than modesty. Individually these phrases bring small numbers of visitors, but there are far more distinct long tail phrases than head terms, they are much easier to rank for, and the person searching one has told you exactly what they want. A site that wins two hundred specific phrases is in a stronger position than one trying and failing to win one broad phrase.

Search intent: the part that decides everything

Two phrases with similar volume can require completely different pages. Intent is what separates them, and it comes in four kinds.

• Informational. They want to understand something. "How does underfloor heating work". The right page is an explanation, and a product page will lose to it every time.
• Commercial investigation. They are comparing before buying. "Best underfloor heating system" or "X versus Y". The right page is a comparison, written fairly.
• Transactional. They are ready to act. "Underfloor heating installer quote". The right page is a service or product page with a clear next step.
• Navigational. They want a specific place. "Screwfix underfloor heating". These are mostly branded and they are the highest converting queries any business has.

There is a free and reliable way to confirm intent: search the phrase yourself and look at what kind of pages are ranking. If the first page is all guides, the engine has concluded that searchers want to learn, and a product page will not displace them regardless of how well it is written. If it is all product and category pages, the opposite applies. Matching the format the results already reward is the highest return decision in keyword work, and it takes thirty seconds per phrase.

What the volume number actually is

Keyword tools show a monthly search volume. It is useful and it should be read for what it is: an estimate, usually an average over the previous twelve months, grouped so that close variants and plurals are counted together, and specific to a country and language setting that you choose.

Three practical consequences follow.

• Averages hide seasonality. A phrase averaging a thousand a month might mean eight thousand in November and almost nothing in June. Look at the trend line rather than the single figure before planning around it.
• Grouping inflates some figures. Variants bundled together can make one phrase look bigger than the page you would write for it.
• Volume is not value. A phrase with low volume and clear buying intent is worth more than a high volume phrase searched by people who will never buy. Multiply expected clicks by conversion rate and by what a customer is worth, and the ranking of your list changes completely.

Our page on where keyword ideas come from and what the volume figure means goes into how the numbers are produced and how to turn a list into a plan.

Where to find the keywords your buyers actually use

Start with the sources you already own, because they are real rather than estimated.

• Google Search Console. The Performance report lists the actual queries your site was shown for. It is the only list in this whole field that is observed rather than modelled, and it is free.
• Your own site search. What visitors typed into your search box, which is them telling you what they expected to find and did not.
• Sales and support records. The exact words customers use for their problem, which are usually better than the words your marketing uses.
• Search suggestions and related searches. Type a phrase into Google and read the suggestions, then read the related searches at the bottom of the results page.
• Communities. Forums and question sites where your buyers ask each other, which gives you the phrasing before it becomes a keyword in a tool.

Only after those do paid tools earn their place, and then mostly for volume estimates and competitive comparison. Our guide to finding SEO keywords from the five sources you already own works through each one in order.

The one rule: one query, one page

Each target query should have exactly one page responsible for it. When two or three of your pages chase the same phrase, you have created a problem called keyword cannibalisation, and it has three consequences: the engine has to pick between your pages and may pick the weaker one, the links and engagement that should accumulate on one page are split, and your own reporting becomes impossible to read because the ranking page keeps changing.

The fix is a keyword map. One row per target query, one URL assigned to it, intent noted, current position recorded. When two pages compete, merge them into the stronger one and redirect the other. Consolidation is unglamorous and it is routinely the fastest improvement available on an established site.

The reverse mistake also matters: trying to make one page serve five different intents. "What is X", "best X", "X pricing" and "X near me" are four pages, not four sections of one.

Where keywords go on a page

Use the phrase where it genuinely belongs, and then stop thinking about it. There is no density target worth chasing.

• The title. Near the front, written as something a human would click.
• The main heading and the first paragraph. The direct answer to the query belongs in the opening paragraph, which is also what automated readers quote.
• The URL. Short, readable and stable. Do not change a URL that is already ranking just to insert a word.
• Subheadings, where they describe the section honestly.
• Internal link text, from your other relevant pages. This is how you tell the engine which page owns which subject, and it is the most underused of the five.

The meta description does not determine position, and it does influence whether somebody clicks, so write it as an invitation. Our internal linking guide covers the link text side, which is where a well made keyword map pays off twice.

Tracking without fooling yourself

Position is not one number. The same page shows at different positions depending on the person's location, device and language, which is why your own browser is the least reliable place to check. Use Search Console's average position, segmented by country and device, and watch the number of distinct queries you appear for as the health measure of the whole programme.

Our pages on taking a clean ranking reading and what a position really measures now cover the method and the reporting.

a Keyword in SEO? Types: the parts people ask about twice

How many keywords should one page target?

One primary query, plus the close variants and questions that belong to the same intent. A page about "underfloor heating cost" naturally also answers "how much does underfloor heating cost per square metre", and that is one page. A page about cost and a page about installation are two.

Are keywords still relevant now that search engines understand language?

Yes, with a shift in how they are used. Engines interpret meaning rather than matching strings, so stuffing a phrase achieves nothing. What keywords still tell you is which topics have demand and in what words, and that information is as valuable as it ever was.

What is a good keyword to start with?

A phrase that describes exactly what you sell, with a place or a qualifier attached, where the pages currently ranking are weaker than the page you could write. Specific beats broad for a first win, and a first win teaches you more than months of research.

How many keywords should a website target overall?

As many as you have genuinely distinct pages worth publishing. A ten page local business might target twenty phrases well. A retailer with a thousand products has a thousand natural targets. Counting keywords is less useful than counting pages that are the best answer to something.

What is keyword difficulty, and should I trust the score?

It is a tool's estimate of how hard a phrase is to rank for, usually built from the link profiles of the pages currently ranking. Use it to sort a list rather than as a verdict. The better check takes a minute: open the top results and ask honestly whether you could write something more useful than what is there.

Should I target keywords my competitors rank for?

Some of them. The useful ones are the phrases where they rank with a weak page, and the phrases they have not covered at all. Copying their whole list means fighting on their chosen ground, which is the most expensive way to compete.

This page covers what keywords are and how to use them. For the research process end to end, see the keyword sources guide linked above, and for the beginner path through all of SEO see SEO basics for beginners.

The query type that is not in any tool

There is a growing category of searching that keyword tools do not report, because it does not happen in a search box. People now ask assistants the same questions in conversational form: "which company should I use for X in Y", typed into ChatGPT, Gemini or Perplexity. The answer names two or three businesses, with no list to scroll and no position to measure, and it is assembled from pages those systems can read. The practical consequence for keyword work is that the questions worth tracking are now phrased as questions, and the thing worth checking is which names come back. AI Knows Us runs those questions across the main assistants and shows which sources each answer was built from.