Keyword Research for a B2B Services Site: Build and Map the Core

How to build a semantic keyword core for a B2B services website targeting Russia and CIS: Yandex Wordstat, clustering by service and segment, mapping to pages.

A semantic core for a services website is a list of search queries your site should be found for, mapped to pages: every query is attached to one page, and every page answers one intent group. If you sell into Russia or the CIS, the job takes a couple of days: write down the base phrasings of your services, expand them through Yandex Wordstat, search suggestions and competitor results, cut the noise, and lay the rest into "service–segment" and "service–geo" clusters.

B2B rules differ from e-commerce: query volumes are small, sometimes a few dozen impressions a month, and that is normal. Your goal is not thousands of impressions but a precise match with the intent of the person who signs the contract. A core of three hundred exact queries beats a thirty-thousand-line parser dump where ninety percent is noise.

Step 1. Base phrasings come from you, not from a tool

Start with a blank sheet. Write down how you name your services and how prospects name them on first calls — these are often different words. A production company says "maintenance of process lines"; the buyer searches for "equipment repair outsourcing". Both go into the core.

Take 5–15 seed phrases: the service itself, service + "cost", service + "company", service + city. This is the skeleton the tools will flesh out. If your team works with a Russian partner or reseller, collect their phrasing too — local buyers rarely use the vocabulary of an English-language pitch deck.

Step 2. Expansion: three sources, and the main one is Yandex

Here is the part that trips up most foreign teams. In Russia the main source of query volume is Yandex Wordstat, not Google Keyword Planner. Yandex holds the majority of the Russian search market, especially in desktop B2B usage, and Keyword Planner gives blurred ranges for Russian-language niches and tells you nothing about Yandex results, where your audience actually sits. If you only plan in Google tooling, you are optimizing for a market share your buyers do not represent. Three working sources, all free:

The third source is underrated. If three companies in the top results keep a separate page for "accounting outsourcing for sole proprietors", that is not a coincidence — it is confirmed demand. Those pages are a ready draft of your site structure.

Step 3. Cleaning: throw away more than half

A raw export after step two easily gives you thousands of rows. Most of it is noise: queries about free options, training and jobs, "do it yourself", adjacent services you do not offer, and plain broken word forms. Cut hard. The mark of a good query for a services site: behind it stands a company that wants to buy, not a person who wants to figure it out alone.

Aim at intent, not volume. "Workplace safety outsourcing price" with twenty impressions a month belongs in the core. "What is workplace safety" with thousands does not — that is student traffic.

Step 4. Clustering to fit the site structure

Now the remainder is mapped to pages. The main rule: one page answers one query cluster. A cluster is a group of queries with the same intent that a person expects to close with one answer. "HR records outsourcing", "outsourced HR record-keeping", "hand over HR records" — one cluster, one page. "HR records outsourcing for construction companies" is already a different cluster, because the intent is narrowed by segment.

For a B2B services site, two clustering axes work:

• Service–segment. "Legal support for developers" and "legal support for medical clinics" are different pages with different arguments, even if the service is one. A developer's owner will not trust a page titled "legal services for business".
• Service–geo. "Logistics outsourcing Yekaterinburg" — only if you genuinely operate in that city and can show it on the page. Publishing two hundred "service + city" pages with the same text is not allowed in practice: the search engine merges them as duplicates, and it adds nothing to site trust.

If you have nothing to fill a cluster's page with — a case, segment specifics, process numbers — skip the cluster for now. An empty page is worse than a missing one.

Step 5. A table instead of chaos

The output is not a text file but a table: query, Wordstat volume (quoted, exact form), cluster, target page, page status. This is a working document for years: copy is written from it, cannibalization is checked against it (are two pages aiming at one cluster?), and site expansion is planned in it.

Move informational queries — "how", "why", "what to do if" — into a separate block of the table. That is semantics for the blog, not for service pages. It works on the same audience: a person reads a breakdown of why cold emails land in spam long before they are ready to discuss a contract. But mixing informational and commercial clusters on one page is a mistake: a service page should sell, an article should explain.

Where it breaks most often

• Perfectionism. Trying to collect "every query in the niche" stretches the work over months, and the core is outdated before launch. You need a minimal working scope: main services covered by pages, noise removed, then the core grows from live data in Yandex Webmaster.
• One page for everything. A "Services" page with a twenty-item list ranks for no cluster, because it answers no specific intent.
• Chasing volume. In B2B, a query with five impressions a month can be your most profitable — if behind it stands a production director with a concrete task.
• A core without a site. The table was assembled and shelved. Semantics only works when pages are shipped for the clusters and given time to index.

Checklist before calling the core done

• Every service in the price list is covered by at least one page with its own cluster.
• No query is attached to two pages at once.
• "Service–segment" clusters are shipped only where there is real content for the page.
• Geo pages exist only for cities where you actually operate.
• Volume is taken from Wordstat in exact form, not broad match.
• Noise and informational queries are moved out of the commercial core.
• The table is accessible to whoever writes the copy — it is an instruction for production, not an archive.

And remember: semantics decides how you get found in search, but it does not replace direct channels. The same "reconnaissance first, contact second" logic drives cold outreach — building a list of decision makers looks a lot like building a core: precision over volume, the same way it works in finding company owners' emails.

FAQ

How many queries should a services-site core have?

As many as needed to cover your real services across real segments — usually hundreds, not thousands. A core of 200–400 exact queries mapped to pages is almost always more useful than a 20,000-row dump where nobody can tell what to do with it.

Can I build the semantics with Google Keyword Planner?

Not for the Russian market — the primary source is Yandex Wordstat, because most Russian B2B search traffic comes from Yandex, and Keyword Planner returns wide ranges for Russian-language niches. Google sources make sense only if you are targeting an audience outside Russia and the CIS.

What if my service queries show almost no volume?

Nothing wrong — that is the norm for narrow B2B. Take queries with volume from one impression up, as long as the intent is exact, and verify demand through suggestions and results: if competitors rank for a query with dedicated pages, the demand exists regardless of the number in Wordstat.

Do I need a separate page for every city?

Only for cities where you have a real presence and can show it on the page: an address, cases, regional specifics. Mass-producing identical geo pages is treated as duplicate content, and instead of coverage you get a trust drop across the whole site.

How often should the core be revised?

Every few months, check Yandex Webmaster for the queries your site actually appears for and extend the table. A full rebuild is needed only when your service line changes or you enter new segments — the core is a living document, not a one-off project.

If you want a second pair of eyes, send us your task for a teardown: we will look at the niche, the channels and the current funnel and point out the bottlenecks — whether that is search demand or cold touches. The unit's working pipelines — email, LinkedIn and Telegram — are described in the capabilities section, and your specific situation can be discussed through a teardown request.