How to Do Keyword Research: From Seed Word to Topic Cluster
Keyword research is the work of turning a few seed words into a mapped list of search terms, where every term has an owner page and a reason to exist. The output is not a bigger spreadsheet. It is a short, ordered set of queries grouped by intent, each assigned to a page that can realistically answer it — and that you can defend in a planning meeting.
Send one product line and one market. We return a sample cluster map.
What is keyword research, in one paragraph?
Keyword research is the practice of finding the words buyers type into search engines, judging what they want when they type them, and deciding which page should answer each one. Done properly, it produces a small map of queries grouped by intent and assigned to URLs, not a long export of phrases nobody has read.
Why a seed list on its own fails
A seed list is a starting point, not a plan. The word "pump" tells you a category exists. It does not tell you whether the searcher wants to buy one, repair one, size one, or compare two types. Teams that stop at seeds end up publishing one generic page per noun, then wonder why nothing ranks for a commercial query.
The advantage lives after the seed: expansion, intent, competition, and mapping. Those four moves are unglamorous, and they are the difference between a research document and a content plan.
What good research answers
Before the file is handed over, it should answer four questions in one screen: which queries we are chasing, what the searcher wants, which page owns each one, and which of those pages already exist.
If any row cannot answer all four, it is a note, not a keyword.
Five steps from seed to cluster
Run them in order. Each step narrows the list, and the order matters: intent judgments made before expansion are guesses, and competition checks made before intent are noise.
-
Collect seeds you can defend
Seeds come from four places: the product catalogue, the sales inbox, the support queue, and the search bar on a marketplace where your buyers already shop. Ten to twenty seeds is enough to start.
Write each seed in the buyer's words, not the engineering team's. If the catalogue says hydrodynamic coupling and buyers type pump coupling, the seed is the second one.
-
Expand along six modifier rails
Run every seed through the same six rails so the list stays comparable: question (how, what, which, why), comparison (vs, best, alternatives, instead of), specification (size, material, rating, capacity), problem (leak, jam, corrosion, downtime), use case (for food, for outdoor, for high pressure), and buying (price, supplier, MOQ, lead time, sample).
One pass per rail. Mixing rails in a single pass is how lists become unsortable.
-
Read intent before you read volume
Open the results for each surviving query. If the page is a guide, the searcher wants an explanation. If it is a category with filters, they want options. If it is a product page with a price and a form, they are close to a purchase.
Volume describes how many people search. Intent describes whether any of them could become a customer. Only one of those numbers belongs in a prioritization meeting.
-
Judge the result page, not a difficulty score
Difficulty scores are a summary of links and authority, and they miss the practical question: can you produce the format that is already winning? Look at the first page and ask what kind of page each result is, how specific its answer is, and whether a specialist site could out-answer it.
If the whole page is marketplaces, a blog post will not win. If it is thin forum threads, a careful guide will.
-
Map one query family to one page
Group variants that share an answer: singular and plural, synonyms, and question forms. Assign the family to one URL, then decide whether that URL is create, refresh, merge, or drop. Two pages chasing one family is the most common self-inflicted ranking problem.
Record the internal link each page will receive. A cluster without links between its pages is a pile of articles.
Send one product line and we will map it to pages that exist.
Four intent types and what each one needs
Classify every surviving query into one of these four. The label decides the page format, the call to action, and how close the reader is to a quote.
Informational
The reader is diagnosing a problem or learning a term. They are not choosing a supplier yet.
how to size a transfer pump
Page: guide or explainer. CTA: a related checklist, not a quote form.
Commercial
The reader is comparing approaches, models, or suppliers. They have a budget shape in mind.
diaphragm vs centrifugal pump
Page: comparison or buying guide. CTA: a spec sheet and a quote.
Transactional
The reader wants a price, a sample, a supplier, or a lead time, and expects to act now.
explosion proof pump supplier
Page: product or category page. CTA: quote form, WhatsApp, lead time.
Navigational
The reader already knows the name they want, including your brand and your competitors'.
webseogrow keyword research
Page: homepage, product line, or support page. CTA: the exact next step.
Free sources that beat a paid export
Before buying a tool subscription, exhaust the sources that describe your own buyers. These six are free, and five of them contain language no keyword tool will invent for you.
- Search Console query reportPages already earning impressions are the cheapest wins. Filter by position 8 to 30 and look for queries where the page nearly answers the question.
- Autocomplete and People Also AskType a seed and read the suggestions as a list of real phrasings. People Also Ask reveals the next question the reader asks after yours.
- Marketplace search barsSupplier directories and marketplaces rank by buyer language, not by editorial style. Their category names are free keyword clusters.
- Sales calls and inboxesEvery repeated question is a page. Ask sales to forward the last twenty inquiries with the customer's own wording.
- Support ticketsPost-purchase questions reveal the objections that block repeat orders, which is where retention content comes from.
- Trade forums and groupsProblem language appears here first. Use it to write the question, then use search data to confirm anyone asks it.
The word list structure that survives a handover
A useful research file has one row per query family and enough columns to make a decision without a meeting. Six columns are usually enough; more than ten and the file stops being read.
| Query family | Intent | Owner page | Action |
|---|---|---|---|
| chemical transfer pump | Commercial | /transfer-pumps/ | Refresh category copy |
| how to choose a transfer pump | Informational | /guides/choose-transfer-pump | Refresh and add table |
| explosion proof pump price | Transactional | /products/ex-proof-pump | Add price band and form |
| pump flow rate calculator | Informational | /tools/flow-calculator | Create |
| diaphragm vs centrifugal pump | Commercial | /guides/diaphragm-vs-centrifugal | Create |
| transfer pump maintenance schedule | Informational | /support/maintenance-schedule | Merge two posts |
Notice the action column. Research that only says "target this" leaves the hardest decision to whoever has time next week. Naming create, refresh, merge, or drop is what turns a list into a calendar.
Five mistakes that quietly waste a quarter
Sorting by volume first
The biggest number usually belongs to the vaguest query, which is also the hardest to convert. Sort by intent and by whether an existing page is close.
Treating every variant as a new page
Plural, synonym, and question forms usually share one answer. Splitting them creates pages that compete with each other and neither ranks well.
Ignoring what already ranks
If the first page is all video or all marketplaces, a text guide is the wrong format, no matter how good the writing is.
Sending every topic to the blog
Transactional queries belong on product pages, not in a post that links to the product. The page that ranks should be the page that converts.
Never writing the no list
Topics you will not cover protect the calendar. Without them, every internal request becomes a keyword, and the cluster fills with noise.
From the map to the page
A cluster is only finished when the pages inside it link to each other in a deliberate order: the guide links to the comparison, the comparison links to the category, and the category links back to the guide. That sequence is what a reader follows, and it is also what tells a search engine which page is the specialist.
When the map is stable, the next job is page-level work: title, headings, internal anchors, schema, and a clear next step. If that is the part your team is missing, Keyword Research Services covers the execution without rebuilding the strategy.
Teams that want the research itself done with the map attached usually start with On-Page SEO Services, then keep the file as a living document rather than a one-off deliverable.
What a pattern looks like
A machinery supplier arrives with a list of 1,400 keywords sorted by volume. The top fifty are generic category nouns, all mapped to the homepage. Nothing ranks, and the team concludes that SEO does not work in their industry.
We rebuild around six query families taken from sales emails, map each to one existing page, and drop 900 rows that describe machines the company does not sell. The list shrinks by two thirds, and for the first time every remaining row has an owner.
That is the whole trick: fewer queries, better placed.
Keyword research questions, answered
How many keywords do I need to start?
Fifty to a hundred qualified terms are enough for a first map. Fewer, better understood terms beat a five thousand row export that nobody has read.
Is free keyword data accurate enough?
Volume estimates are rough at every price point. Use them to rank opportunities, not to forecast traffic. Search Console impressions are the only numbers that describe your own site.
What is a keyword cluster?
A cluster is a group of queries that share one intent and deserve one page. It usually contains a primary term plus close variants, synonyms, and question forms.
Should I target one keyword per page?
Target one intent per page, not one literal string. A page can answer a family of variants as long as every variant leads to the same answer.
How often should research be repeated?
Review the map every quarter and rebuild it once a year. New queries appear slowly, but intent shifts quickly when a market or a product line changes.
Can I do this without paid tools?
Yes for the first pass. Search Console, autocomplete, marketplace categories, and sales notes will produce a workable map. Buy a tool when the list outgrows a spreadsheet.
Send one product line. Get a cluster map back.
We will show you which queries are worth owning, which pages already exist, and which rows you can delete this week. No dashboard demo, no volume forecast you cannot use.