Faragiri

Learn Keyword research SEO · Finding keywords, Search intent, Volume and difficulty and 3 more

Finding what people type is the easy half. Deciding what they meant, and which one page of yours answers it, is the work that changes rankings.

6Branches
6Nodes below
3Depth

Keyword research is not a list of words. It is the work of finding out what a person wanted when they typed something, and then deciding which single page of yours is going to satisfy that want. The list is a by-product. Teams that treat the list as the deliverable end up with four hundred rows in a spreadsheet and no idea which page to write first.

Start from the question, not the phrase. A phrase is what got typed; a question is what was meant. Search engines have been answering the second one for years, and since generated answers arrived at the top of results they answer it out loud, which makes the gap between the two visible for the first time.

The six moves, worked on one shop

Cloudberry Bikes in Sheffield has 1,400 monthly visits and one blog nobody reads. Its keyword work took an afternoon and produced six decisions rather than four hundred rows.

Cloudberry Bikes, Sheffield

Six moves from an afternoon of keyword work

Starting from 1,400 monthly visits and an unread blog, one afternoon replaced four hundred rows with eleven pages worth building.

  1. 01Read live results firstFive shop service pages for bike fitting showed a blog guide would not rank.
  2. 02Separate four intentsInformational saddle tips and local commercial bookings need different pages.
  3. 03Merge competing URLsCombining three half-aimed pages lifted bike fitting from position 19 to 6.
  4. 04Split by modifierBike fitting, cost, and near-me map to service, pricing, and local pages.
  5. 05Pick winnable volumeA 90-search query you can win beats a 4,000-search term you cannot.
  6. 06Group rows by answerCollapses four hundred raw keywords into eleven distinct target pages.

Every step removes guesswork before a writer drafts a single paragraph.

  • read the current results before writing. Search the query and look at what is already ranking. For bike fitting the first page was five service pages from bike shops, which says the query is commercial and a guide will not rank.
  • informational, commercial, transactional, navigational, the four intents. How to adjust saddle height is the first, bike fitting Sheffield is the third, and no single page serves both.
  • one query, one page. Cloudberry had three pages half aimed at bike fitting, splitting their own signals. Merging them into one moved that page from position 19 to position 6 in eleven weeks.
  • a modifier splits one query into three. Bike fitting, bike fitting cost and bike fitting near me want a service page, a price page and a map listing respectively.
  • volume is a guess and difficulty is a competitor count. Both numbers are modelled rather than measured, so a 90 volume query you can win beats a 4,000 volume query you cannot.
  • group by answer, not by word. Sort the list by which page would satisfy each query, and the four hundred rows collapse into the eleven pages actually worth writing.

What a keyword actually is

Three things have to be true at once before a query is worth targeting. People have to be typing it. It has to mean something specific enough that one page can answer it. And you have to be able to answer it better than what is already ranking. Miss any one and the row in your spreadsheet is a wish.

Search

A query is only worth targeting where three things overlap

People have to be asking it, it has to mean one thing, and you have to be able to answer it better than what is there now.

  • People ask it
  • It means one thing
  • You can answer it
Worth a page
  • People ask itThe only one a keyword tool measures
  • It means one thingSpecific enough that a single page can satisfy it
  • You can answer itBetter than whatever is ranking today

The overlap is where a row in the spreadsheet becomes a page. Outside it, the row is a wish.

Most keyword tools only measure the first. That is why a list built from a tool alone reads like a list of things other people have already written about, which is exactly what it is.

The second condition is the one people misread. A query that means three different things needs three pages, and a page that tries to serve all three serves none of them: someone searching for a price, someone searching for a definition and someone searching for a login are not one audience. Splitting them is a decision you make while building the first list, not afterwards.

The two numbers, and what neither of them knows

Every tool gives you a volume and a difficulty score, and both are estimates built from what happened before. Volume is a smoothed average of past months, so it is wrong for anything seasonal and wrong for anything new. Difficulty is mostly a function of how many links point at the pages currently ranking, which is a statement about the past rather than about the page you are going to write. See how badly both of those numbers mislead before you sort a spreadsheet by either of them.

The two numbers

What volume and difficulty actually know about a query

Both are estimates built from what already happened, which is why sorting a list by either of them ranks nothing.

  1. YesHow many people typed this beforeSmoothed over past monthsYes
  2. YesHow many links point at the pages ranking nowThis is most of what difficulty meansYes
  3. NoWhether the searcher wanted to buyIntent is a judgement, not a measurementNo
  4. NoWhether a generated answer already ends the searchNew since 2024 and not in either numberNo
  5. Not checkedWhether the query will still be asked next yearNobody checked, and the tools do not claim toNot checked

The third state is the honest one. A checklist with only yes and no would file the last row under one of them and hide that nobody looked.

The practical effect is that the highest-volume row in a list is usually the worst place to start. It is the query the biggest sites in the field have been answering for a decade, and the specific, longer questions underneath it are where a new site can actually appear. Those are long-tail queries, and they convert better for the same reason they are easier: the person asking has already narrowed down what they want.

Neither number knows whether the searcher wanted to buy, to learn, or to find a page they had already seen. That is search intent, it is the thing that actually decides what your page has to be, and no tool reports it because it is a judgement rather than a measurement.

How this is done in 2026

Google Search Console is still the only source that tells you what people typed to reach your pages rather than what a panel of users typed somewhere. It is free, and it replaced the keyword data Google removed from analytics in 2011. Start there, always, before opening anything paid: the queries you already half-rank for are the cheapest rankings you will ever get.

2026 workflow

Three checks in order before paying for or writing anything

First-party search data comes first, paid databases fill gaps, and AI answer checks test whether clicks still exist.

  1. 03Assistant answer checkAsk the assistant your readers use; target queries where its answer is thin, wrong, or hedged.Since 2024
  2. 02Ahrefs or SemrushSurfaces terms you do not rank for yet; Ahrefs Webmaster Tools covers your own site free.Gap discovery
  3. 01Google Search ConsoleShows real queries reaching your pages since analytics hid keywords in 2011.Free first-party

Queries where you already sit on the edge of page one are cheaper to win than chasing cold volume.

For everything you do not yet rank for, Ahrefs and Semrush remain the two serious keyword databases, both paid, both with a free tier that is enough to learn on. Ahrefs Webmaster Tools gives site owners free access to their own site's data, which is the honest free path if the subscription is out of reach.

What is new since 2024 is that a query can be answered without a click. Ask the assistant your readers use the question you are targeting, and read what it says. If the generated answer is complete, the click is gone and the row is worth less than its volume suggests. If the answer is thin, wrong, or hedged, that is a page worth writing, and it is the mechanism underneath all of this that decides which of the two you are looking at.

The mistake: one list, many pages

The failure that costs the most is not choosing a bad keyword. It is letting several of your own pages chase the same one. Two pages on the same query split the links and the signals between them, and search engines pick one of the two, usually not the one you would have picked. It looks like a ranking problem and it is a filing problem, and it is called cannibalisation.

Filing before writing

Unmapped keyword lists versus one-query-one-page mapping

Most ranking losses on multi-page topics are filing failures rather than writing failures.

Unmapped list (cannibalisation)Multiple pages chase one query
Keyword map (1-to-1)Written query assignment per URL
Signal distributionLinks and relevance split between pagesAll links and signals focus on one URL
Who picks the winnerSearch engine guesses, often the wrong pageYour team decides on purpose
Downstream clarityLooks like an unexplained ranking dropGuides on-page copy and site structure

Mapping produces no public page on its own, which is why teams skip it and pay for it later.

The fix is boring and it works: one query, one page, decided on purpose and written down. That decision is keyword mapping, and it is the step most teams skip because it produces no visible output.

Once the map exists, the next question is what a single page can actually control about its own position, which is on-page, and how those pages relate to each other, which is site structure.

Inside this, 6

Alongside this

Above this