Semantic core in 2026: collection, clustering and automation through AI
Semantic core manually in 2 weeks - this is 2022. In 2026, a basic kernel for 500-1000 requests is assembled per day through Wordstat, Key Collector and Claude. Analysis of clustering methodology, calculation of priorities, automation through AI.
The semantic core is the foundation for two tasks at the same time: SEO promotion and contextual advertising. The difference is how to use it. An incorrectly assembled core means money in Direct, lost on non-targeted requests, and content that no one is looking for.
I've been collecting kernels for clients since 2018. A standard project is an online store or service company with 200–2000 pages. The largest core I worked with was 84,000 keys for an ecom project in the construction niche. Of these, 9,000 were actually used in advertising. The remaining 75,000 are garbage that did not undergo clustering and filtering by intent.
Semantics is not a table with keys. It's a map of what people want to find and buy.
1. What is the semantic core and why is it needed?
The semantic core is a structured list of search queries for which you want to attract traffic. For SEO, this is the basis for creating content and optimizing pages. For contextual advertising - the basis for grouping ads and managing bids.
Without a core, you are working blindly: writing articles “on a whim,” running ads for broad keywords, wasting your budget on irrelevant traffic. With the core, you know exactly what content is needed, what queries convert, and where the competition is too high.
2. Types of requests: intent is everything
Before assembling the kernel, you need to understand the types of queries by intent - the user’s goals when searching:
Information (“how to do”, “what is”, “why”). The user is looking for knowledge. In SEO - ideal for blogs and expert content. In context, it almost never converts into a purchase, the money flies away.
Commercial (“buy”, “price”, “order”, “cost”). The user is ready for a transaction. In context, the main type. In SEO - categories, product cards, landing pages.
Navigation (brand or website name). The user is looking for a specific company. They are not needed for advertising - this is your audience who already knows you.
Transactional (“download”, “register”, “checkout”). The user is ready for a specific action. Relevant for SaaS, marketplaces, banks.
3. Collection tools: free vs paid
| Tool | Price | What does it give? | When to use |
|---|---|---|---|
| Yandex.Wordstat | Free | Frequency in Yandex, associative queries | Always - a basic tool |
| Google Search Console | Free | Real queries that are already generating traffic | To optimize existing pages |
| Key Collector | 1800 ₽ one-time | Wordstat parsing, frequency, clustering | Cores from 1000 keys |
| Semrush / Ahrefs | from $100/month | Competitor keys, positions, backlinks | Competitive intelligence |
| Yandex Direct (forecast) | Free | Traffic forecast, bids, competition | Pre-launch potential assessment |
4. Collecting the core: step-by-step logic
Step one - marker requests. These are 10-20 high-frequency phrases that describe your product or service. For a furniture store: “buy a sofa”, “living room furniture”, “custom sofa” and so on. Markers are the skeleton around which everything else is built.
Step two is expansion. For each marker in Wordstat, we look at the left-handed column (query variations) and the right-handed column (associative queries). Plus - search suggestions in the browser bar. They show what people actually type, not what we think they type.
Step three - competitors. Using Semrush or SimilarWeb, we look at what queries bring traffic to the top 5 competitors. Some of these requests will end up in our core, some will be new ideas.
Step four - clustering. We group requests by meaning and intent. One cluster - one page. You can’t optimize one page for 50 different clusters: Yandex doesn’t like this and the traffic gets smeared.
5. Negative keywords: this is more important than collection
For contextual advertising, negative keywords are critical. Without a proper backing track, you will be paying for clicks from people who are looking for anything but your product.
The standard list of negative keywords for most commercial projects: free, do it yourself, DIY, instructions, do it yourself, video, reviews (if you don’t sell through reviews), foreign brands, irrelevant regions.
On one of the food delivery projects, after adding 200 negative keywords, CTR increased from 4.2% to 7.8%, CPL decreased from 680 rubles to 390 rubles. This is not an optimization - it's just a normal setting.
6. SEO Core vs Context: The Difference
Same semantics, different applications. In SEO, we work with all types of content: informational queries bring traffic to the blog, commercial queries bring traffic to categories and cards. Positions have been growing for months.
In context - only commercial requests with direct intent. Information traffic will give, but there will be no conversions: a person is looking for an article, not a purchase. The money will go away.
Learn more about setting up Direct campaigns for specific clusters: Yandex Direct in 2026 - step-by-step guide.
7. Kernel update: when and how
Semantics is not static. New requests appear, frequency changes, competitors occupy niches. For SEO - update once every 6-12 months or when new directions are launched. For context, a quarterly review plus constant work with negative keywords.
A good signal that the core is outdated: conversion is falling, although the CTR has not changed. This means that the requests in the campaign no longer correspond to the buyer’s intent.
About how semantics affects visibility in AI search: AI Overview and Perplexity SEO - how to get to the sources of AI answers.
If you need help with semantics for your project, write to Telegram @dipustovalov or through form. Starting consultation - 0 ₽.
Related materials: CPC benchmarks in Yandex Direct, landing page template, a complete guide to performance marketing, query clustering.