How to build a semantic core the right way

What a semantic core is

A semantic core is a set of words and phrases that reflect the site’s topic and structure and describe what it does. Those words are keywords. They answer: “What will the user find on this site?”

We cover how many keywords usually go into the working set, how to classify queries, and a step-by-step build.

How many keywords you need

There’s no hard “exactly N keywords” rule: volume depends on niche, region, and how many landings you can actually maintain. The ranges below are order-of-magnitude for a start, not dogma.

Classifying search queries

Keywords split into groups by different parameters. We’ll cover those parameters. If you want practice first — jump to the step-by-step guide below.

Frequency

How do you work with keyword frequency? Promoting on high-frequency queries is hard — competition is fierce. Beginners often start with micro and low-frequency keys: there are more of them, and the long tail makes early visibility more realistic. On commercial pages, refine the offer: guarantees, timelines, delivery, meta tags.

After low-frequency work, add mid-frequency terms. They often hint at section structure. You may need site and content updates. Once LF and MF are covered with landings and copy, high-frequency queries usually lift more naturally — but that is not a guarantee of TOP-1 “by itself”.

Business signals

Here there are two groups. Commercial keys bring visitors who want to order, buy, or use a service — anything involving money.

The second group is non-commercial queries. People want information; they don’t need to order or buy yet.

The first group fits service sites and online stores. Informational resources and topical blogs need the second group.

Geolocation

A query may include the user’s location — or not. For a geo-dependent query the search engine returns results based on where the user is. Results differ by region. If someone wants to order a cake in Moscow, cakes in Tver rarely help.

If the query has no place cue, results tend to look the same across cities.

In site promotion, geo-dependent keys are often smarter — competition stays within the named region. Phrases without location compete with results from everywhere.

User needs

Depending on the content the user wants, there are four query types.

If your site offers that content, you can take this group. Often people prefer Google Images, streaming services, or YouTube.

Transactional. Functionally close to commercial. Used to find products or services to buy. To rank, you need a catalog with product cards — photos, descriptions, prices — plus cart, payment, and delivery options.

Search engines rarely put small stores in the TOP. Large aggregators usually hold the first rows.

If the resource has useful info for such keys, you can use them in promotion. More often they enter the core when the site is also pushed with paid search.

Other query types

Fuzzy — no clear goal; it’s unclear what to show. Engines ask for clarification or return topical articles. Yandex has Spectrum-style tech for fuzzy queries so the SERP can cover different needs on the same topic.

Why you need a semantic core

Working with the core later helps refine site structure, track dynamics, find why rankings are weak, and show whether you need subtopics.

Step-by-step: building a semantic core

Creating a semantic core is several steps: markers → expand → clean → clusters → map to URLs. Download the table template and fill it as you go.

Download CSV template (marker, frequency, type, cluster, URL)

Step 1. Gather information

Put yourself in the potential client’s place. What interests them? What’s the problem? Dive into the topic and collect everything in that direction.

General info often lives on the client’s site. For a full dive, use outside sources too: competitors, the client’s structure, catalogs and price lists. Wordstat and Google Keyword Planner help collect keywords.

Competitor paid-search analysis also helps — how they promote a similar product or service. Manually or via tools like SpyWords / SimilarWeb (depending on access and budget).

Step 2. Expand the keyword list

Now expand the draft query list. Build a table in a separate file and add every word and phrase from the previous step. You’ll need keyword tools: Wordstat and Google Keyword Planner.

Enter each word or phrase into the tools. You’ll get refined variants and associations.

You can also collect keys automatically with programs such as Key Collector.

Then clean the core. Remove queries that don’t fit the site’s specifics or topic — manually or with stop-word tools.

Step 3. Group keywords. Cluster the core.

You now have a huge sheet of every possible query for the site. Split keys into groups or blocks by topic or client needs. That’s clustering.

In paid search, word form matters. One ad may use “water purification”, another “water treatment”, both leading to the same site — so the keyword can appear in the SERP title and match the query. You end up with several ad groups that still send users to one URL. In Google or Yandex Ads, keys for one product are often merged instead, simplifying structure.

What about SEO? More work: each cluster should lead to a specific page. If the page doesn’t exist — create it. Otherwise queries compete or hang without a landing. On a small site clustering is often manual; on large ones tools like Rush Analytics or clustering modules elsewhere help (you set the SERP URL-overlap threshold separately).

Step 4. Map keys to structure. Refine the core.

The previous step gave you ready keyword clusters. Several groups may still share one URL. Then refine the site structure, create extra pages, and move part of the keys there.

When clustering, map queries to pages so URLs don’t duplicate each other by meaning.

What to do after the core

The core isn’t the finish — it’s the entry to site work. Next usually: Title and H1 per cluster, copy and trust blocks on landings, internal links between sections, tech and indexing.

SEO site structure Technical SEO audit

Mistakes when working with keywords

These gaps grow into bigger problems: the wrong competitor set, weak rankings, irrelevant meta tags, low metrics.

Tools for working with a semantic core

Tools teams usually use to collect and group a core. Not a ranking—pick by task and budget. Narrow or outdated panels aren’t included.

Yandex Wordstat

A basic free source of frequency and wording for Russian-language search. Region, seasonality, and dynamics live here. Marker expansion usually starts here.

Google Keyword Planner

Google Ads keyword planner: query ideas, frequency estimates, and Google forecasts. Handy as a second angle next to Wordstat, especially if you also look at Google.

Key Collector

A desktop suite for parsing suggestions, collecting and cleaning a core, negatives, and soft-duplicate removal. Saves time on mid and large lists.

SlovoEB

A free, simplified relative of Key Collector. It can’t do everything; Wordstat access needs a Yandex account — with limits and risk of blocks under aggressive parsing.

Titlo

A platform with modules: query collection and clustering, list comparison, relevance, rankings, meta tags, and monitoring. Modules can be used separately or in one account.

Rush Analytics

Cloud clustering and related SEO tasks. Handy when the core is already large and manual grouping by SERP URL overlap takes too long.

Keys.so / Serpstat

Paid systems with data on keys, competitors, and SERP pages. Help expand mid- and low-frequency keywords and see who already owns the niche. Choice depends on the team’s usual stack.

Just Magic

A tool for semantics and LSI text surroundings (including “Aquarelle”-style scenarios). Useful when the core already exists and you need to strengthen copy for a cluster.

Ahrefs

Stronger on backlinks and English/global markets, but also shows competitors’ organic keys. For a purely “Yandex” core it more often complements Wordstat and local suites than replaces them.

Any toolset only speeds the work. How you pick wording and avoid mixing intents on one page — that comes down to practice and common sense.