What keyword research is for
Three decisions depend on it:
- What to build. Which pages the site needs, which it has, and which it has too many of.
- What each page is about. The query cluster a page owns, which drives its title, structure and content — the inputs to the on-page section.
- What to measure. The queries whose positions, impressions and clicks matter, which the measurement section tracks.
Keyword research is not the same as search intent analysis: that guide reads one query's results page to decide what one page must be; the research process operates on hundreds or thousands of queries and decides how many pages there are. The two meet at the keyword map.
Where query data comes from
| Source | What it gives | Caveats |
|---|---|---|
| Search Console (GSCGoogle's search performance and indexing tool.) performance report | Real queries the site already appears for, with impressions, clicks, position and click-through rate (CTRShare of impressions that result in a click.) | Only queries the site already ranks for; anonymised long tail omitted |
| Google Ads Keyword Planner | Volume ranges and related ideas from Google's own data | Ranges unless the account spends; groups close variants together |
| Third-party keyword tools | Large candidate lists with estimated volume and difficulty | Estimates modelled from samples; difficulty scores are each tool's own invention |
| Autocomplete, "People also ask", related searches | How queries are phrased and what adjacent questions exist | Not volumes; personalised and regional |
| Internal site search | What visitors ask the site in their own words | Small samples; reveals gaps |
| Customer-facing teams | The vocabulary customers use before they learn the industry's | Anecdotal; invaluable for new categories |
| Competitors' ranking pages | Which queries the market already serves | Tools' estimates again |
The number next to a keyword is an estimate of monthly searches, usually averaged over a year, often grouped with variants, and never a prediction of what a page will receive. Position, click-through, intent and the results page's features decide that. Treat volumes as relative — this cluster is larger than that one — and never as targets.
The process
- Seed. Start from the business: the products, services, problems solved, audiences, locations, and the words customers use. Add every query the site already appears for from Search Console.
- Expand. Run seeds through the tools and the results-page features to build a long candidate list. Include questions, comparisons, modifiers and the vocabulary from support and sales. A language model can draft candidates here but never volumes; the AI for keyword research guide covers what to take from one and how to check it.
- Classify intent. Label each query by what the searcher wants — the intent classification guide gives the rules and the modifier patterns for doing this at scale.
- Cluster. Group queries that one page can satisfy — by results-page overlap and by meaning — per the keyword clustering guide. The cluster count is the page count.
- Prioritise. Score clusters by relevance to the business, demand, the site's chance of competing, and the stage of the journey, so the work starts where it pays.
- Map. Assign each cluster to an existing or planned URL, one owner per cluster, per the keyword mapping tutorial.
- Brief and build. The map feeds content briefs and the topic cluster plan.
- Revisit. Quarterly, against Search Console: new queries, shifting intent, clusters that merged or split on the results page (SERPThe page of results a search engine shows.).
Further guides
Beyond the core process, further guides cover particular research jobs:
- Competitor keyword gap analysis — a tutorial for finding the queries search competitors rank for and the site does not: picking the right competitors, exporting and comparing rankings, filtering by intent and fit, and prioritising.
- Keyword research when volumes are low — researching search demand when tools show little or no volume, in niche business-to-business markets, new categories and small countries, and validating what the research finds.
Difficulty and opportunity
Tool difficulty scores estimate how strong the pages currently ranking are, usually from their links. They are a rough sort, not a verdict. A better read of whether a site can compete is to look at the results page itself: who ranks, how specialised they are, what formats appear, and whether the top results are weak on the intent. A query where the top results are generic pages from large sites can be winnable with a specific, better page; a query where every result is a specialist with a decade of links is not, whatever the score says.
Keyword research for different site types
- Local businesses add location modifiers and "near me" patterns; the local section covers the profile side.
- Stores research at category, refinement and product level; the e-commerce section decides which refinements become pages.
- Publishers and B2B sites research questions and problems more than products, and cluster into topics for the content section.
- International sites research per market, because the queries differ by language and by country even in the same language.
The keyword research expertise page describes the author's working method and includes a clustering demonstration.
