Keywords and Search Intent

Track 1 · Lesson 2 of 6 · about 15 minutes

Keywords and search intent

A keyword is not a topic. It is a compressed request, and the same words can carry four different requests depending on who typed them. Getting the request wrong is the most common reason a well-written page never ranks.

Four jobs behind a query

Informational — the person wants to understand something. how does canonicalisation work.

Navigational — they want a specific place. search console login. You will not win these for someone else’s brand, and you should not try.

Commercial investigation — they are deciding between options. ahrefs vs semrush. They are not ready to buy, but they are close enough that the comparison is worth serious effort.

Transactional — they want to act now. buy standing desk.

The same phrase can move between these over time, and one phrase can carry two intents at once. seo audit is a person wanting to learn how to run one, and a person wanting to hire someone to run one, sharing a search box.

Read the SERP, do not guess

The fastest way to learn what an engine believes a query means is to look at what it already rewards. Search it and read the result types. All guides? It has judged the query informational. Product pages and shopping results? Transactional. A mix, or a “people also ask” block full of definitional questions? The intent is split, and you get to choose which half to serve.

This matters more than any keyword tool output, because it is the engine telling you its own answer. If the first page is entirely comparison articles and you publish a service page, you are arguing with a decision that has already been made.

What the tools are actually for

Keyword tools give you three useful things: a list of phrases you would not have thought of, a relative sense of demand, and a rough read on how contested each phrase is. Treat volume figures as estimates with wide error bars — they are modelled, not measured, and they differ between tools for the same phrase.

One firm rule: a language model cannot give you search volume. If you ask an LLM for monthly searches it will produce a plausible number, and that number is invented. Pull metrics from a data source; use the model for what it is genuinely good at, which is grouping and labelling a list you already have.

Clustering: one page per intent

Once you have a list, group it by the answer rather than by the words. what is a canonical tag, canonical tag meaning and rel canonical explained are one page. canonical vs noindex is a different page, because it answers a different question — even though the vocabulary overlaps almost completely.

The test is simple: could one page satisfy both searchers without padding or contradiction? If yes, one page. If you find yourself writing “if you are here for X, skip to section four”, you have two pages wearing a trench coat.

Long tail, and why it is worth your time

Specific multi-word queries have less demand each, far less competition, and much clearer intent. For a site without established authority they are usually the only realistic entry point — and they are disproportionately the queries where a generative answer sends someone onward, because the specific question tends to have a specific answer that lives on one page.

Starting at the head term because it has the biggest number is the classic first mistake. The number is real; your chance of touching it this year usually is not.

Going deeper: Content strategy guides · SEO tools guides · try the Rank the SERP exercise.