Semantic Core for a Website: How to Build It, Clean It, and Actually Use It

Most clients who come to us asking for a "semantic core" have already ordered one before. They have an Excel file with thousands of queries from last year. Nobody explained what to do with it. The contractor collected everything, handed over the file, and disappeared. This is the most common story in SEO projects: a semantic core as an artifact instead of a tool.

Let's walk through the entire process — from collection to implementation — and explain why each step matters not on its own, but only in connection with the next one.

What Is a Semantic Core and Why Do You Need It

A semantic core is a complete list of search queries that your target users type when looking for your products or services. Not "keywords in general," but specifically those with real search volume and relevance to your business.

Here is what it is used for in practice:

Understanding how to structure your website — and how many pages it should have
Knowing what to write on each page (which words to use, which questions to answer)
Prioritising effort: which queries will bring traffic faster, which will pay off in the long run
Measuring SEO results through rankings for specific queries

Without a semantic core, SEO work becomes guesswork. With one, it becomes a manageable process with clear metrics.

Where to Find Queries

A good semantic core is assembled from 5–6 sources in parallel. Each source contributes its own piece of the picture.

Yandex.Wordstat — the fundamental tool. It provides monthly impression volume for queries, related queries, and queries with similar words. The main limitation: it only shows what people are already searching for — it will not surface new queries that will emerge six months from now. It is used as the primary base: enter 2–5 seed queries describing the core service and expand the tree in depth.

Yandex Suggest — the completions the search engine offers as you type a query. These are real patterns of actual user queries in real time. Suggestions often contain "long tails" that have no notable volume in Wordstat but that people genuinely type. Scraping suggestions by letter ("buy air conditioner a", "buy air conditioner b", etc.) yields hundreds of unique variations.

Google Search Console — an invaluable source for sites with existing traffic. It shows the actual queries for which your site is already receiving impressions and clicks. Especially useful are queries with high impressions but low CTR — they are already close to the top but the page is not optimised. These are quick wins.

Competitors — analysis via Megaindex, Ahrefs, or SEMrush: examine which queries the top 5 competitors in the niche rank for. These tools show queries where a competitor holds a position but you do not. This is a ready-made gap list. There is no need to invent anything — just take what is already working on other sites in the same niche.

Megaindex — provides visibility data for sites in Yandex and Google, and keyword rankings for competitors. Particularly useful for the Russian market, where Western tools have incomplete data for Yandex.

Internal site search — if your site has a search function, the query logs are gold. Users tell you themselves what they are looking for. We often find queries that appear nowhere in Wordstat but actually convert, because the users are already on the site.

How to Clean the Semantic Core

A raw semantic core after collection contains 3,000–10,000 queries, half of which you do not need. Cleaning is critical.

Negative keywords — queries with clearly irrelevant modifiers: "free", "DIY", "download", "used", "reviews", "forum". These should be removed first. For an air conditioner shop, the query "air conditioner repair DIY" is not a target: the person is fixing, not buying.

Geo queries for other cities — if you operate in Rostov, queries like "buy air conditioner Moscow" are not for you. Yandex's algorithm already accounts for the user's geolocation, and a page targeting another city without a real presence there will not generate traffic. This is one of the most common mistakes — leaving "moscow", "saint petersburg", "krasnodar" in the semantic core of a regional business.

Marketplace queries — "air conditioner Ozon", "air conditioner Avito", "air conditioner Wildberries". The person has already chosen a platform; your site is not what they want. Remove them.

Informational vs. commercial queries — this is not about removing, but about separating. The query "how to choose an air conditioner" is informational: the person is in the research phase. "Buy 9000 BTU inverter air conditioner" is commercial: the person is ready to purchase. They belong on different pages with different content. Mixing them together is not acceptable.

Irrelevant queries — those that formally contain the keyword but are not about your topic. "Hair conditioner" in the semantic core of a climate equipment shop. It sounds absurd, but it happens in automated collection.

After cleaning, a good semantic core loses 40–60% of its queries. This is normal — 500 precise queries are better than 2,000 vague ones.

Clustering: What It Is and How to Do It

Clustering is the process of grouping queries so that each group is promoted through a single page.

The main mistake is clustering by word forms: "buy air conditioner" and "air conditioner buy" placed in one cluster because the words are the same. This is an outdated approach. It ignores the fact that the search engine may show different pages for different queries.

The correct method is SERP-based clustering. The logic: if two queries share the same URLs in Yandex's top 10, those queries can be promoted with a single page. If the top results differ, the algorithm treats the queries as meaningfully different, and separate pages are needed.

Example: "air conditioner for an apartment" and "wall-mounted air conditioner for home use" may seem like synonyms. But if the first query surfaces general catalogue categories in the top results and the second surfaces specific selections of wall-mounted models, these require different pages.

SERP clustering is done with tools: services like Just Magic, Rush Analytics, or KeyAssort retrieve the top-10 positions for each query and automatically group them by overlap. Manual clustering takes days; tool-assisted clustering takes hours.

Intents within a cluster — an additional layer. Within a single cluster, queries may carry different intentions: "12000 BTU air conditioner price" (commercial — wants to buy), "12000 BTU air conditioner which to choose" (informational — wants advice). They can coexist on one page if that page addresses both intentions. But the content must answer both.

Mapping: Each Cluster to a URL

After clustering, you have 50–200 groups of queries. For each one, you need to decide: promote it on an existing page or create a new one.

The mapping rule: one cluster = one page. If there is no suitable page for a cluster — create one. If a page exists but the content does not address the cluster's queries — optimise it.

How to decide: look at the query volume in the cluster and its competitiveness. High-frequency clusters with high competition are typically the homepage or a top-level category. Mid-frequency specific clusters are subcategory pages. Low-frequency narrow queries are product cards or individual blog posts.

A typical map for a climate equipment shop:

Cluster "buy air conditioner" → catalogue page /catalog/kondicionery/
Cluster "Daikin air conditioner" → brand page /brands/daikin/
Cluster "9000 BTU air conditioner for a 25 sq.m. room" → a selection page or blog article
Cluster "air conditioner recharge Rostov" → service page /services/zapravka/

Mapping is recorded in a separate Excel column: query → cluster → URL. This becomes the technical brief for content work.

New pages are only created for clusters with sufficient volume. There is no point creating a page for 3 queries with a combined volume of 50 impressions per month — it is simply duplication with no effect.

The Main Mistake — A Semantic Core for Its Own Sake

This needs to be said plainly: a semantic core that sits in Excel and is never implemented is worth exactly zero. Literally. It will not produce a single ranking, a single click, or a single lead.

Yet this is exactly how most keyword projects end: the contractor collected 5,000 queries, packaged them into a neat file, and handed it to the client. The client said thank you, put it in the "SEO" folder, and returned to their current tasks. A year later they hired a new contractor to "build a semantic core" — because "something was done before but there were no results."

The semantic core is a starting point, not a final deliverable. After the core comes:

Optimising existing pages for their clusters
Creating new pages for uncovered clusters
Writing content that answers the queries in each cluster
Placing internal links with the correct anchor text
Tracking rankings for the queries in the core

Only when all of this is done does the semantic core start producing results. And those results are measurable: 0 queries in the top 10 became 40. 200 monthly visitors became 1,400.

A good SEO contractor does not "deliver a semantic core" — they work with it. The difference is that you are not paying for an Excel file; you are paying for rankings and traffic.

Summary: A 6-Step Checklist

Collect your seeds — 5–10 base queries that precisely describe your service or product. The entire core expands from these.
Run collection from 5 sources — Wordstat, Suggest, Search Console, competitors, Megaindex. A single source gives an incomplete picture.
Clean with negative keywords — remove irrelevant queries, marketplace queries, other cities' geo terms, and informational queries (if they are not needed).
Cluster by SERP — not by word forms, but by overlaps in the top search results.
Map to URLs — assign each cluster to a specific URL. New clusters mean new pages.
Implement and track — optimise pages, create new ones, monitor rankings by cluster. Without implementation, the semantic core is just Excel.

Need help building a semantic core and implementing it? Contact us — we build keyword strategies aimed at real traffic results, not beautiful files.