How to Build a Keyword Core from Scratch in One Evening

If you've never built a core, it seems like a week's worth of work. In reality a basic core for one site comes together in one evening — if you know the order of steps and don't waste time on the unnecessary.

Let's walk through it on a real example: the niche "children's dentistry in Moscow." We'll do everything step by step.

Step 1: Find your seed phrases

Seed phrases are the 3-5 most obvious short queries people might use to search for you. They're easy to think up: imagine you're a potential client off the street, and type into Yandex what you'd want to find.

For our dental clinic that's:

• children's dentistry moscow
• treating children's teeth
• pediatric dentist
• baby teeth treatment
• dentistry near me kids

The main rules:

• No brand. No one knows you by name yet, they search with general words.
• No long tails. Right now we're gathering the foundation, not the refinements.
• With commercial intent (if you have a commercial site). "Teeth treatment" and "how to treat children's teeth" are different intents. See about intent.

If you have fewer than 3 seed phrases — expand the niche mentally ("dentistry" → "children's dentistry" → "children's dentistry moscow"). If more than 10 — that's too many, stick to the 3-5 most capacious ones.

Step 2: Expand via Wordstat

You open Wordstat, pick a region (Moscow — geo-id 213, selected in the top-right corner), and enter each seed phrase one at a time.

Wordstat shows two columns:

• Left: queries that include your phrase. That's the long tail.
• Right: queries that similar people search for. Those are alternative options.

You copy both columns into one table/text file. From the left take everything, from the right — only what fits the topic (there's often junk and irrelevant synonyms in there).

After going through all 5 seed phrases you have 200-400 queries. That's your "raw core." More about Wordstat — in a separate article.

Step 3: Expand via AI

Wordstat shows only what has already been searched — it doesn't know new phrasings. AI, on the other hand, can suggest queries you wouldn't think up yourself.

Go to /free/keyword-research (our free tool), enter 1-3 seed phrases and the niche. Claude expands them into 40-60 ideas of different types: commercial, informational, by occasion, by feature, by geo.

For our dental clinic the AI will produce:

• "children's dentistry 24 hours moscow"
• "treating pulpitis in a child price"
• "dentistry for teenagers with braces"
• "baby tooth extraction under anesthesia"
• "how to convince a child to go to the dentist"

And so on. After this stage you have 300-500 queries in raw form.

Step 4: Pull the tail via autosuggestions

Search autosuggestions are the autocomplete that Yandex/Google show when you start typing a query. It's a free source of long tail.

Simple method: type your main query into Yandex, see 5-10 suggestions, copy them. Then type in each suggestion — it has its own suggestions. Go 2-3 levels deep like that.

In our case:

• "children's dentistry moscow" → "children's dentistry moscow free," "children's dentistry moscow 24 hours," "children's dentistry moscow districts," ...
• "children's dentistry moscow districts" → "children's dentistry moscow north-eastern district," "children's dentistry moscow south-west," ...

More about autosuggestions — in a separate article. After this stage the core grows to 500-800 queries.

Step 5: Clean out the junk

The final step — throw out queries that aren't yours. In our dental clinic's raw core there will be:

• From Wordstat's left column: "children's dentistry moscow free" (we're paid), "treating children's teeth video" (an informational query for observing parents), "children's dentistry jobs" (for job-seekers, not clients).
• From the right column: "pediatric neurologist moscow," "pediatric ENT moscow" (a different niche), "teeth whitening" (adults).
• From the AI expansion: general informational ones ("how to brush a child's teeth"), which might go in the blog, but not on the main commercial pages.

Every query flagged "not mine" — into the discard pile. More about cleanup — in a separate article.

After cleanup, 200-400 target queries remain — that's the working core, ready for clustering and distribution across pages.

How much time it takes in the end

That's about 3 hours with a tea break. On a day off — you'll manage it in an evening. On a weekday — better split it into 2 sittings.

What NOT to do at the gathering stage

• Don't cluster right away. First gather more and cleaner, then group. If you try to group right away — you get confused.
• Don't check the frequency of every query. That's slow and unnecessary work at the gathering stage. You'll only check frequencies for the ones that make it into tracking.
• Don't discard the long tail. "80 impressions a month" seems like little, but these are your future first positions. See about the long tail.
• Don't copy someone else's core. Competitors may have built it crookedly, and you'd copy all their mistakes.

Next article: Wordstat — a detailed guide for first-time users.