Building a Keyword Core with AI: Search Data, an Agent and Common Sense
How to build a keyword core faster with an AI agent: collecting queries, removing junk, grouping by page and planning articles. What you can trust to AI and what you cannot.
A keyword core is the list of queries people should find you by, organised by page. It used to take days in spreadsheets. Now an AI agent handles most of the routine: collection, cleanup, grouping. But deciding which pages to build is still a human call.
That is how I built the keyword core for this site: Yandex Wordstat, an agent in the browser and a manual review.
What a keyword core is
It answers the question "what do people search for when they need what I do?" If you implement AI, your clients search for things like "AI implementation for business", "AI agents for business" or "business process automation". Each query has a volume: how many times a month it is searched.
The core tells you:
- which pages your site needs;
- which headlines and copy match what people search for;
- what to write about for months ahead.
Building the core with AI, step by step
- 01Seeds20-40 phrases
- 02Collectkeyword tool and agent
- 03Cleanjunk out
- 04Groupone group, one page
- 05Planpages and articles
1. Seed phrases
Write down 20-40 short phrases that describe your services. AI is great at expanding the list: ask it how else people might phrase the same thing.
2. Collecting data
For each seed, a keyword tool (Wordstat for Russian search, Google Keyword Planner and similar tools elsewhere) shows every query containing those words and its volume. A browser agent can go through all the seeds and collect the tables while you do something else.
3. Removing junk
The most important step. Results always include noise: job searches, "free download", student queries, other brands, ambiguous words. AI filters obvious junk well using rules you set.
A real example: the Russian query for "VK automation" looks perfect for an automation builder, but almost all its volume is speech therapy (exercises for "automating" a sound). You only catch that by looking.
4. Grouping
Queries are grouped by intent: one group, one page. "AI agent development", "build an AI agent" and "AI agents for business" share a page. "AI agent courses" gets another, because that person wants to learn, not hire.
5. Page and article plan
Big commercial groups become landing pages. Informational queries ("what is", "how to", "difference between") become articles.
What to trust to AI and what not
| AI handles | A person is needed |
|---|---|
| Expanding seed phrases | Deciding what you actually sell |
| Collecting tool data | Understanding ambiguous queries |
| Filtering obvious junk | Separating commercial from informational in edge cases |
| Grouping similar queries | Choosing which pages to build first |
| Suggesting headlines | Checking that the copy is honest and useful |
Common mistakes
- Chasing only big queries. Competition is fierce there. Mid and long-tail queries often bring clients faster.
- One page for everything. Search engines struggle to rank a page that tries to answer ten different questions.
- Forgetting location. City-specific queries are small but very precise.
- Never revisiting the core. Demand shifts and new tools appear monthly. Review the core regularly.
FAQ
Can I build a keyword core for free?
Yes. Free keyword tools exist. Paid services speed things up, but for a small site a free tool plus an AI assistant is enough.
How many queries should a core have?
It depends on the niche. A services site usually starts with a few hundred clean queries, then expands the core for each article.
Can AI write SEO copy from the core?
It can draft, but a person must check facts, honesty and usefulness. Search engines are getting better at telling empty text from useful text.
What matters more, the core or content quality?
Both. The core tells you what to write about. Quality decides whether the reader stays and becomes a client.