The Keyword Funnel
Keyword research is the work of turning what buyers actually type into a ranked list of search terms, with every term attached to the page that should own it. The dependable way to do it is a funnel: start wide with every phrase your market might use, filter to mid-tail, keep only long-tail questions you can answer better than the current results, then approve a short list where one keyword maps to one page. Wide lists feel productive. Filtered lists earn rankings.
Each layer removes candidates. Nothing reaches the page list without passing a written filter.
Run four stages, in order. Pull 200 to 500 broad terms from your catalog, sales notes and competitor pages. Remove anything irrelevant or unsearchable. Narrow to 60 to 120 mid-tail phrases with a clear modifier. Keep 30 to 50 long-tail questions you can answer better than the current results. Approve 20 to 40 keywords, each tied to a single page.
Why the shape matters
Most teams skip the filtering and call a bulk export research. They finish with four thousand rows, no owners, and a content plan that dies in week three. A funnel forces a decision at every layer. A query is relevant or it is not. A page exists or it does not. A phrase is worth a paragraph, a section, or an entire page of its own.
Decisions are what turn a spreadsheet into a build order. When each layer has a written rule, two people on the same team reach the same answer without a meeting. That matters more than any volume number, because the expensive part of SEO is not finding keywords.
The shape also matches how demand behaves. Head terms carry huge volume and vague intent, so they are expensive to win and easy to misread. Long-tail questions carry thin volume but explicit intent, which means a smaller page can win them and convert on the first visit.
The funnel keeps both without confusing them. Broad terms tell you which market you are in. Long-tail terms tell you which page to write next week.
The four layers
Collect everything a buyer might type: product categories, materials, processes, problems, synonyms, misspellings, and the words your sales team hears on calls. Use search suggestions, competitor category pages, marketplace filters, Reddit threads and your own support inbox. At this layer, judgement is off. Quantity is the point. You are drawing the boundary of the market, not choosing pages.
Filter Relevance Filter Searchable Target 200–500
keep only what passes
Group the survivors by theme and add the modifiers buyers use at the moment of comparison: service, supplier, manufacturer, price, cost, near me, for small business, best, alternative, vs. This layer is where intent becomes visible. A phrase that pairs a category with a commercial modifier is a page candidate. A phrase that pairs a category with an informational modifier is a section inside that page, not a page of its own.
Filter Commercial intent Filter Page exists? Target 60–120
keep only what passes
Now read the actual results. For every mid-tail phrase, check whether the top ten pages answer the question or only mention it. Long-tail wins come from gaps: a size chart nobody publishes, a compatibility list nobody maintains, a process step nobody explains, a comparison nobody dares to make. If the first page already answers it completely, you need a different angle or a different query.
Filter Gap in results Filter Can we prove it? Target 30–50
keep only what passes
This is the only layer anyone outside the research work should see. Each row holds one keyword, one page, one intent, one owner and one quarter. If two keywords want the same page, keep the higher-intent one as the target and treat the other as a supporting phrase inside the copy. If a keyword has no page and no budget, cut it now instead of carrying it for a year.
Filter One page per keyword Filter Owner assigned Target 20–40
Filter criteria
Write these rules down before you open a spreadsheet. A criterion that lives only in someone's head cannot be audited, and it cannot be handed to a writer.
| Layer | Question it answers | Keep if | Drop if | Output |
|---|---|---|---|---|
| 1 · Broad | Is this our market at all? | A buyer could type it and expect us | Wrong industry, wrong language, zero demand | 200–500 raw terms |
| 2 · Mid-tail | Are they shopping, or just reading? | Modifier shows comparison or purchase | Only academic or DIY curiosity | 60–120 phrases |
| 3 · Long-tail | Can we beat the current results? | Clear gap we can fill with proof | Results are complete and authoritative | 30–50 questions |
| 4 · Approved | Who builds it, and when? | Page, owner and quarter assigned | No page, no owner, no evidence | 20–40 keywords |
A term that fails a layer is not deleted, it is parked. Parked terms become next quarter's pipeline, which is why the export stays useful after the sprint ends.
Keyword to page
The last mile of keyword research is a mapping table, and it is the part most teams leave unfinished. A keyword without a destination page is a wish. The mapping decides whether a phrase becomes a new page, a section inside an existing page, or a line of copy on a product page.
| Keyword | Layer | Page type | Destination | Supporting phrase |
|---|---|---|---|---|
| stainless steel filter housing | 2 | Category page | /collections/filter-housings | ss filter housing |
| filter housing supplier for OEM | 2 | Service page | /pages/oem-filter-program | custom filter housing |
| how to size a filter housing | 3 | Guide | /blogs/guides/sizing | filter housing sizing chart |
| 10 inch vs 20 inch filter housing | 3 | Comparison | /blogs/guides/10-vs-20 | which filter housing length |
When a page targets two competing keywords, both lose. The title has to choose, the opening paragraph has to choose, and the internal links have to choose. Pick the phrase with the clearer buying intent and let the other appear naturally in the body, in a subheading, or in the FAQ. If the two phrases truly need separate answers, they need separate pages.
Sharing is allowed when the queries are variations of the same question: singular and plural, with and without a qualifier, two spellings of one term. In that case, write one page that answers all of them and vary the subheadings, not the promise. The test is simple. Would a reader be annoyed to land on this page after typing that phrase? If yes, split it.
Common mistakes
A database only returns what you asked for. If the seed list is thin, the export is a confident-looking dead end.
High volume with vague intent produces traffic that never contacts you. Intent first, volume second.
Forty thin pages compete with each other. Group questions that share an answer and build one strong page with clear sections.
A keyword you cannot win is a keyword you should not build for. Read the top ten before you commit a writer.
No owner, no date, no page. The funnel is only finished when the last layer has a build order attached.
What we deliver
Our Keyword Research Services produce the four layers as four files: a raw collection sheet, a themed mid-tail map, a long-tail gap list with result notes, and an approved list with page, owner and quarter. You also get the filter rules in writing, so the next sprint does not start from zero.
Once the approved list exists, the bottleneck moves to production. That is where SEO Content Writing Services take over: one page per approved keyword, answer first, proof second, internal links third. Research and writing stay separated on purpose, because a writer who is still deciding which keyword to target writes slower and ranks worse.
The sprint
FAQ
Twenty is enough for a new site with a small team. Forty is realistic for a store with a catalog and a writer producing four pages a month. Beyond that, the list gets ahead of production and starts to rot.
No. Search suggestions, competitor category pages, marketplace filters and your own support inbox will fill layer one. A paid database saves time on volume estimates, not on judgement.
Two weeks for a focused sprint, with about two hours of your team's time for the relevance pass. Anything longer usually means the filter rules were never agreed.
Only when the answer needs a full page. If the question can be answered in a section, put it in a section of a stronger page and win the cluster together.
Twice a year for the full cycle, and once a quarter for a light layer three pass. Markets shift, competitor pages get rewritten, and new product lines create new mid-tail phrases.
They go to a parked tab with the reason attached. Parked terms are cheap to revisit and often become viable when a new page or a new product line appears.
