Keyword Research: How to Build a Semantic Core Step by Step
Practice3 minInnovialabs editorial team
In short
A semantic core is built in five steps: seed queries, expansion with tools, cleaning out junk, clustering by search results, and mapping each cluster to one page. The outcome is a site map where every page has a clear job.
What a semantic core is and why you need it
A semantic core is the list of search queries your audience uses to find what you offer, grouped by site pages. It answers two questions: which pages you need and what each one should be about. Without it, a site grows chaotically: pages compete with each other and important topics stay uncovered.
Step 1. Seed queries
Start with seeds — short 1–3 word phrases describing your products, services and customer problems.
- List every category and service.
- Add synonyms and casual phrasings ("web development", "build a website", "make a site").
- Look at how customers phrase things in inquiries, chats and reviews.
- Study competitors' menus and headings.
Step 2. Expansion
Each seed expands into dozens or hundreds of long-tail queries.
| Tool | Type | Use |
|---|---|---|
| Yandex Wordstat | Free | Volumes and expansions for Yandex |
| Google Keyword Planner | Free (needs a Google Ads account) | Ideas and volume ranges for Google |
| Google Search Console | Free | Queries your site already appears for |
| Search suggestions | Free | Real user phrasing |
| Key Collector | Paid | Bulk collection and processing |
| Serpstat, Semrush, Ahrefs | Paid | Competitor keywords, difficulty, volumes |
Free sources are enough for a small site. Paid tools save time on large projects and competitor analysis.
Step 3. Cleaning
A raw list is often half junk. Remove:
- Irrelevant queries: "free", "DIY", "download" if you sell a service.
- Other brands and cities you don't serve.
- Queries with zero exact-match volume — check with operators, not just the broad number.
- Duplicates and word reorderings.
Build a negative keyword list — it is useful for paid search too.
Step 4. Clustering
Queries are grouped so that each group can be targeted by one page. The most reliable approach is SERP-based clustering: if two queries share many URLs in the top 10, the search engine treats them as one topic.
- Hard clustering — queries join a group only if every pair shares enough URLs. More precise, more groups.
- Soft clustering — a link to the group's main query is enough. Fewer groups, higher risk of mixing intents.
Paid services and Key Collector can cluster by SERP. A small core can be grouped manually by meaning, then disputed pairs checked in the results.
Also respect intent: "buy a sofa" (commercial) and "how to choose a sofa" (informational) need different pages even though the words overlap.
Step 5. Mapping to pages
Each cluster gets one page:
- Commercial clusters — categories, services, product pages.
- Informational clusters — blog posts and FAQ.
- The cluster's main query goes into the title and H1; the rest appear naturally in body text and subheadings.
- Clusters without a suitable page become a plan for new pages.
Keep everything in one table: cluster, queries, volume, intent, URL, status.
Common mistakes
- Chasing only high-volume queries — competition is fierce and conversion is often lower than for long-tail queries.
- Several pages for one cluster — cannibalization; they undermine each other.
- Mixing intents on a single page.
- Building the core once and forgetting it — demand changes, so update it.