Semantic Core: Finding the Queries That Actually Buy

The classic story: the contractor sends an Excel file with 8,000 rows, an invoice for 60 thousand roubles and the words "the core is collected". Six months later the file is still untouched, because nobody knows what to do with it. A useful core is not a list of queries — it is a map of landing pages with priorities and expected traffic.

Do not start with Wordstat

Start with sales. Export from your CRM the wording clients use to describe their problem, listen to ten call recordings, look at the search queries report in Metrica and at your on-site search. That is where the markers come from — 30–60 base phrases from which everything else expands. A core assembled only from search suggestions is always skewed toward informational queries.

Expanding and cleaning

• Wordstat and suggestions to collect the long tail, plus an export of competitors' queries to check completeness
• Strip negative keywords immediately: "free", "DIY", "employee reviews", other brands' names
• Drop queries with fewer than 5 searches a month, unless this is a narrow B2B niche where even 3 impressions mean a deal
• Check frequency in quotes with an exclamation mark — up to 40% of broad-match volume is empty

Cluster by the search results, not by words

The only reliable way to tell whether two queries can be promoted on one page is to compare the top 10 for each. If 4 or more URLs overlap, the queries group together. "Buy CRM" and "CRM system price" usually merge, while "what is CRM" goes into a separate informational cluster, because the top there is articles rather than product pages. That is exactly why commercial clusters get service pages and informational ones get the blog: the difference in approach is described in the article on how to write articles that bring clients.

How to spot a buying query

Look at the results page, not at your intuition. If the top is aggregators, service pages and ads, the intent is commercial. If it is articles and videos, the person is still researching and will not submit a form today. The beginner's mistake is writing an article for "order website development" and being surprised by zero leads. On how intent shows up in search results it is worth reading the recommendations from Google Search Central.

Prioritization: where to begin

Rank clusters by the formula "search volume × proximity to purchase ÷ competition". In the first quarter take mid- and low-frequency commercial clusters with 2–5 strong competitors in the top — there the result shows in 2–3 months. Put the high-frequency ones in the plan for the second or third quarter, otherwise you will invest for six months with no feedback. On realistic horizons, see the piece on SEO timelines.

From spreadsheet to site structure

The final document should contain: the URL (existing or new), the main query, the list of queries in the cluster, total search volume, the page type and the person responsible. In exactly this form the core becomes a specification for development and content. On online store projects it turns directly into the catalog tree — how that is arranged is covered in the article on ecommerce SEO.

Common mistakes

• One page for two different intents — the site ranks for neither
• Two pages for one cluster — cannibalization, and Yandex picks the worse one
• The core is collected but the site structure has not changed — 90% of the work is wasted
• No rebuilds: demand shifts, so the core is worth refreshing every 6–9 months

A good core for a 20–30 page services site takes 5–8 working days to build and then lives on as a working document for the team. If you need help with this stage, look at how we run SEO projects, or tell us about your task in contacts.