Keyword research, step by step: find the searches that pay

Most keyword research I see is a spreadsheet of terms sorted by search volume, with the top twenty highlighted. That’s not research. That’s a list, sorted by the one column that tells you least about whether writing the page will make you any money.

Start with what you already rank for

Before touching a keyword tool, open Google Search Console and look at the queries you already appear for. Almost every site is showing up on page two or three for things nobody ever wrote a page about — Google inferred it from a paragraph buried on a service page.

Those terms are the cheapest wins available. You’ve already proved the site can surface for them; it just hasn’t been given a page that deserves to win.

Free tool · Striking-distance finder

Find the searches you’re nearly winning

Export 3 to 16 months from Search Console → Performance → Export → Download CSV, unzip it and choose Queries.csv, or paste the table. Your data stays in your browser.

Queries.csv

Positions from

to

Minimum impressions

Or paste the rows

Find them

How do you find keywords for SEO?

You find keywords for SEO by starting from ten to twenty plain seed terms that describe what you sell, running each through a keyword tool, and keeping the specific variations people actually type. Search Console, competitors’ rankings and Google’s own suggestions fill in what the tool misses.

Seed keywords are the plain terms that describe what you do. Not clever, not optimised — just the words a customer would use. A plumber in Ipswich starts with plumber, boiler repair, emergency plumber. I’d say ten to twenty seeds is plenty.

Each seed is an entry point, not an answer. Put it into a keyword tool and it branches into dozens of specific variations, and those variations are the things people actually type.

How do you find local keywords?

Local keyword research means adding the places you serve to your seeds, and checking how Google treats “near me” versus a named town. For the Ipswich plumber, plumber becomes plumber ipswich, boiler repair ipswich, emergency plumber felixstowe and so on for each town worth serving. Search each one yourself. If the results for a town search are a map pack and local firms, a page for that area can compete. “Near me” searches rarely need a page with “near me” in it, because Google answers them from the searcher’s location, so they’re won by being a relevant local business with a complete Google Business Profile rather than by keyword stuffing.

Intent decides the format before anything else

Two people can search almost the same words and want completely different things. Google reads that difference well, and it won’t rank a blog post for a buying query no matter how good the blog post is.

The practical test costs thirty seconds: search the term yourself and look at what Google is already ranking. If page one is all comparison articles, Google has decided that query is commercial, and your product page won’t displace them. My rule is to build what the results say Google wants.

Judge every term on three things, not one

Volume tells you how many people search. It doesn’t tell you whether you can win, or whether the people searching will ever buy. Score each term on all three and the list reorders itself completely.

What are long-tail keywords?

Long-tail keywords are longer, more specific searches with lower volume each and clearer intent, such as running shoes for flat feet rather than running shoes. They’re easier to win and easier to satisfy on one page.

Short-tail terms are one or two words — SEO, running shoes. Enormous volume, enormous competition, and vague intent that’s almost impossible to satisfy on one page.

Long-tail terms are longer and specific — running shoes for flat feet — with less volume each and far clearer intent.

Ahrefs’ own keyword database shows how lopsided this is: keywords with fewer than 10 searches a month make up almost 93% of its US database. Each one is tiny, but together they’re where most searches sit.

The practical version: a new site targeting SEO is entering a marathon untrained. I’d win the specific terms first, and the broad ones become reachable later because of the authority the specific ones built.

What is keyword difficulty?

Keyword difficulty (KD) is a 0 to 100 score keyword tools give a search term to estimate how hard it is to reach the top ten. Ahrefs builds its score from how many websites link to the pages already ranking, so it measures backlink strength, not content quality or intent. Checking the results page yourself is more reliable.

Ahrefs explains its own KD plainly: it pulls the top ten pages for the keyword and looks up how many websites link to each, and the more links, the higher the score. Semrush and others use their own formulas, so the same term scores differently in each tool.

The scores also wobble within one tool. When I pulled Ahrefs data for this site in September 2026, it gave link building a KD of 0 in the US and 77 in the UK.

The honest alternative takes two minutes per term. Search it in a private window and read page one:

• Who ranks? National brands and big publishers, or sites the size of yours?
• What format? If it’s all comparison articles or all product pages, that’s the format you’d need to build.
• How good are they? A page two years out of date, or one that half-answers the query, is beatable by a better page.

For boiler repair ipswich, a page one of trade directories and two small local firms says a well-built local service page can compete. KD is only part of it: the other part is authority, which is why earning backlinks makes more terms winnable over time.

Keep the pages that will never sell anything

This is the part that gets argued about. A plumber doesn’t make money from someone searching how to fix a dripping tap — that person is going to fix it themselves.

I’d write it anyway.

That page earns links, gets shared, and answers a question thoroughly enough that Google starts treating your site as a source on plumbing rather than a listing for one. When the same person hits a job they can’t do, you’re the name they already have.

And mechanically: the authority those pages accumulate is what lifts the pages that do sell.

The distinction worth holding on to is between a page that sells and a page that is worthless. Those aren’t the same thing.

Plumber jobs is worthless — the people searching it want a job, not a plumber, and no amount of volume changes that. How to fix a dripping tap sells nothing and is still one of the best pages on the site.

Steal the list your competitors already built

Your competitors have done keyword research for you. They just can’t stop you reading the results.

Put three to five rivals into Ahrefs or Semrush and export everything they rank for. The gold isn’t their top rankings — it’s the terms where they sit between positions five and twenty.

That’s Google saying the query is winnable by a site like theirs, with a page nobody has bothered to make excellent yet.

What is keyword clustering?

Keyword clustering is grouping search terms that one page can answer, so each page targets one main search plus its variations. The reliable test is shared results: if Google ranks the same URLs for two terms, they belong on one page. If the results differ, they’re separate pages.

How to do it by hand, for a list of any size:

1. Sort your scored list by the main term each keyword seems to belong to.
2. For each pair you’re unsure about, search both and compare the top ten. Three or more of the same URLs in both is a good sign they’re one cluster.
3. If your keyword tool shows a parent topic (Ahrefs does), terms that share one usually belong together. Use it as a quick first pass, then confirm by checking the results.
4. Pick the highest-value term in each group as the page’s primary, and treat the rest as variations the page should answer.

In the plumber example, boiler repair ipswich, boiler engineer ipswich and boiler repairs near me pull up largely the same local pages, so they’re one service page. How much does a boiler service cost returns guides and price pages, so it’s a different page with a different format.

If you’d rather hand the clustering and writing over, that’s what my SEO content service covers.

The whole process, start to finish

Figure

Seven steps from a blank page to a roadmap, starting with data you already have

1. 01Mine Search Console
2. 02List seeds
3. 03Expand
4. 04Competitor gap
5. 05Score
6. 06Cluster
7. 07Sequence

The tools worth having

I don’t think you need all of these. You do need at least one that shows volume, and Search Console, which is free and which most businesses never open.

Can you do keyword research for free?

Yes. Google Search Console, Google’s autocomplete and People Also Ask boxes, and Google Keyword Planner cover the whole process for £0. Paid tools save time and add competitor data, but a small business with one site can build a solid keyword plan without them.

The free route, in order:

1. Search Console. The Performance report shows the queries your site already appears for, with clicks, impressions and average position. Its default view covers the past three months. Queries with plenty of impressions and an average position between about 8 and 20 are your first list.
2. Autocomplete and People Also Ask. Type each seed into Google and note the suggestions, then open the People Also Ask box and expand it. Every question you open adds more.
3. Keyword Planner. It gives search volume estimates and related keyword ideas. Google’s Keyword Planner help says you have to finish setting up a Google Ads account, including entering billing information, before you can use basic features like “Get ideas for new keywords”.
4. The results page itself. Scoring winnability by reading page one, as described above, costs nothing and, in my view, beats any difficulty score.

What free tools can’t do well is the competitor gap: seeing everything a rival ranks for needs Ahrefs or Semrush.

How do you find the questions people ask?

Question keywords come from People Also Ask, autocomplete, your own Search Console queries and the questions customers put to you directly. Filter any keyword list for words like how, what, why, can and does, and you have the questions a page needs to answer.

Questions are worth mining separately because they map onto sections: a page that answers why is my boiler losing pressure in a clear sentence under a heading that asks it is easy for Google to show in People Also Ask, and easy for AI answers to quote.

In Search Console, filter queries with “Custom (regex)” for question words to see which ones you already surface for. Then check the list against real enquiries: the questions customers ask on the phone are often the ones no tool shows, because the volume is too low to register.

What this looks like in practice

For one topic, expect five to ten pages. For a full site programme, several hundred, worked through one area at a time so the whole thing never becomes an undifferentiated pile of posts.

The research isn’t the deliverable. The sequenced, argued-for list of pages is — and the difference between the two is most of the value.

Common questions

How many keywords should one page target?

One, plus the variations that mean the same thing. If a page is trying to win two genuinely different searches, it's two pages, and splitting it usually makes both rank better than the combined version did.

Is search volume reliable?

Treat it as an order of magnitude, not a number. Tool volumes are modelled estimates and routinely wrong by a factor of two or three. They're fine for deciding whether a term deserves a page and useless for forecasting traffic.

How do I start if the site is brand new?

You have no Search Console history to mine, so start with competitors instead. Put three or four rivals into a keyword tool, export what they rank for, and look for terms where they sit between positions five and twenty — those are winnable by a better page.

What is search intent and why does it decide everything?

Intent is what the searcher actually wants: to learn, to find a specific site, to compare options, or to buy. Google is very good at reading it. Publish a blog post for a buying query, or a product page for a question, and you won't rank however good the page is.

How often should keyword research be redone?

Review quarterly for most businesses, monthly if the market moves fast or you're launching things. What changes is rarely the core terms — it's which ones you can now realistically win as the site gains authority.

Should I target keywords with no buying intent?

Often yes. A page that answers a question you'll never charge for earns links and trust, and that’s what lifts the pages that do sell. The distinction that matters is between a page that doesn't sell and a page that’s worthless — those aren't the same thing.