Keyword research is a collected and cleaned list of the search queries a site should be found for. Clustering groups those queries by meaning and intent so that one group maps to one page. The list is built from Wordstat, search suggestions, Yandex Webmaster data and competitor queries, then cleaned and checked for volume.
01What keyword research is and why a site needs it
A familiar situation: the site has plenty of pages, the texts are written, and search keeps sending people to three random addresses. Almost always the reason is that the structure was invented in someone's head instead of being derived from demand. Keyword research restores the order of steps: first you find out what people actually search for, and only then decide which pages should exist.
A keyword set is a map of demand. The site's sections, the headings, the article topics and the order of work for the coming months all come out of it. A list of words to sprinkle through the text gives you none of that.
Step by step from here: where to get the queries, how to read search volume, what to discard, how to group what is left and how to turn groups into pages.
02Sources of queries: where to collect them
A single source always skews the picture. Wordstat shows what people ask Yandex but knows nothing about your site; Webmaster knows your site but cannot show you demand you have never appeared for.
- Yandex Wordstat: the base source of volume and wording, including the refining phrases.
- Search suggestions and the related-searches block — the live language of the audience, usually with no volume attached.
- Yandex Webmaster and Google Search Console: the real impressions and clicks your site already gets.
- Internal site search: the things people could not find in the menu.
- Competitor queries from visibility services — as hypotheses, not as the truth.
- Correspondence with customers and the questions your sales team hears: wording that appears in no tool at all.
03Search volume: broad, phrase and exact match
The classic beginner's mistake is planning the work from the broad-match number. It sums up every phrase that contains your words in any combination, so the figure next to a high-volume query is almost always inflated several times over.
Check the refined volume and take the decision on the whole group of queries, not on one attractive number in the first row.
| Type of volume | How it is requested | What it shows |
|---|---|---|
| Broad match | The query with no operators | Every phrase containing these words in any order and form |
| Phrase match | The query inside quotation marks | Only these words with no additions; word forms are still allowed |
| Exact form | An exclamation mark before each word | The exact word form, with no declensions |
| Webmaster data | The site's own query report | Real impressions and clicks, not an estimate of demand |
04Cleaning the set: what to throw out
Cleaning is the most underrated stage of all. When I built the keyword set for this site, roughly 627,000 raw rows of exports came down to about 254,000 unique queries after cleaning and deduplication — and that was before filtering out other people's intent.
The Russian word for promotion turned out to be ambiguous: the export pulled in military dispatches, sewing machine parts and a clinic that happens to carry the same name. Job seekers formed a separate layer — vacancies, salaries, training. Not one of those phrases will ever bring a customer, yet every one of them inflates a report about potential.
- Homonyms and unrelated topics that matched on a single word.
- Job seeker queries: vacancies, salary, courses, how to become one.
- Do-it-yourself intent, if you sell a service rather than training.
- Competitor brands and navigational queries aimed at other people's sites.
- Locations where you do not operate.
- Fragments, broken phrases and single words with no meaning.
- Services you do not offer: traffic collected for them will not convert anyway.
05Keyword clustering: which methods work
Clustering answers one question: which queries can be promoted on the same page, and which ones need pages of their own. How similar the words look decides nothing here. What decides is a match of intent and the pages search already shows for those phrases.
An error in cluster boundaries costs more than an error in collection. Merge two intents and the page turns vague and answers neither query properly. Split one intent across two pages and they begin competing with each other — that is what cannibalisation means.
| Method | How it works | When it fits | Risk |
|---|---|---|---|
| Manual grouping by meaning | A specialist reads the queries and sorts them into topics | A small set, a narrow topic | Slow and subjective |
| By overlap in search results | Comparing which sites rank at the top for each query | Competitive commercial niches | Depends on the quality of the results scrape |
| Automatic, by text similarity | A model groups the queries by meaning | Large sets of tens of thousands of phrases | Cluster boundaries still need a manual check |
| Mixed | Automation plus a manual check of intent | Practically always | Takes specialist time |
06From queries to site structure: the point of the work
The rule that saves the most money: one intent, one page. If two queries need different answers, they need different addresses. If they need the same answer, two pages will get in each other's way and search will not know which one to show.
So the output of keyword research is a map of pages: what to keep, what to create, what to merge and what not to build at all. A spreadsheet of keys settles none of that on its own.
| Type of cluster | Intent | Type of page |
|---|---|---|
| «Order», «cost», «services» | Ready to buy | A service page |
| «What is», «how it works» | Getting to grips with the topic | A blog article |
| «Compare», «versus», «reviews» | Choosing between options | A review, cases, a hub |
| Queries with a city name | Looking for something nearby | A local page, if there is something to fill it with |
| Vacancies, training, other people's brands | Not your customer | Do not build a page |
07Tools: what to collect with, and whether free is enough
You can build a keyword set for free: Wordstat, search suggestions, Webmaster and a plain spreadsheet cover the needs of a small site. Paid tools are bought for speed and for automatic grouping of large volumes — the data inside them is the same.
- Yandex Wordstat — free, the main source of search volume.
- Yandex Webmaster and Google Search Console — free, the real impressions your own site gets.
- Desktop collectors such as Key Collector — they speed up volume scraping and stop-word cleaning.
- Competitor visibility services — a source of hypotheses and of other people's landing pages.
- Spreadsheets and scripts — a perfectly sound working tool, provided you understand what you are counting.
- Clustering services built on search results — they save days of manual work on large sets.
08Timing and cost: what drives the volume of work
Timing depends on the number of products, regions and languages rather than on the topic of the business: a one-page site and a store with a thousand items are different jobs. Scraping refined volumes and checking cluster boundaries by hand each take their own time on top.
In my price list, keyword research and clustering costs $340, and SEO site structure design based on that set costs $490. The exact timing and scope are fixed in the proposal before the work starts, and they change only together with an agreed change of scope.
09In short: the order of building a keyword set
The order runs like this: collect queries from several sources, scrape the refined volume, clear out other people's intent and homonyms, group what is left by intent, assign a page type to every group, and only then write the texts. Keyword selection without that step produces a list of phrases with nowhere to apply them.
If you already have a site and it is unclear which pages compete with each other, it is more logical to start with diagnostics: a full SEO audit costs $580 and includes a review of the structure and of how well the pages match demand. Describe the task in an enquiry and I will tell you whether you need a full keyword set or simply order in the pages you already have.
