How to do keyword research for a focused website
Updated · By SEOAction
To decide how to do keyword research for your website, begin with the audience question and the page you can actually provide. Collect candidate phrases, record their sources, check intent and fit, then add measurements with their limitations. Choose a page purpose before treating a large number as an opportunity.
This is our suggested research workflow for a bounded website task. It is not a tool ranking, a promise of traffic or a universal minimum search-volume rule. Use current instructions for the measurement service you select.
Define the reader and the job
Write down who the page is for and what they should be able to accomplish after reading it. Separate an informational question from a request to purchase a service or use a tool. List capabilities the website does not offer so you can recognize an attractive phrase that would send the wrong visitor.
Use the language in your own authorized customer questions, support notes or public material you have permission to examine. Keep the original source and a short explanation of the need beside each candidate. Do not present a phrase suggested by an assistant as a quotation from a customer.
Our suggestion is to begin with one small topic where you understand the intended outcome. Record uncertainties about the audience rather than filling the list with additional terms to make the research appear complete. A short list with traceable reasons is a useful starting point.
Build a candidate pool with provenance
Collect wording from relevant page content, reader questions and your own proposed descriptions of the task. Label those origins separately. Expand a candidate when it describes a different reader need, not merely to create more rows that look independent.
For niche vocabulary, compare your phrasing with the actual products, processes or problems the page will discuss. Keep unfamiliar terminology unresolved until you can verify its meaning. Our method does not assume a general tool's suggestions are a complete picture of a specialist audience.
If you ask an assistant to organize a list, review its additions against the underlying sources. Remove invented evidence and unsupported feature descriptions. Keep the human decision about whether your page can serve the phrase separate from any automated grouping.
Review intent before buying more data
Inspect the current search results for a candidate using a method you can describe. Record the date, context and visible result types. Note whether they appear to emphasize product use, purchasing, explanations or another job, while keeping the limits of that observation clear.
Use that inspection to ask a concrete question about your planned page. Could it satisfy the task being sought, and what would it need to cover? Do not treat a title or snippet as a full audit of another page, or infer that every reader has the same intention from a single result set.
Where the fit remains uncertain, keep the phrase in a review group. Our suggestion is to clarify the page's purpose before paying for a larger suite or committing to a publishing schedule. The workflow does not require a particular subscription or tool bundle.
Add measurements without replacing the question
Record the service, geography, date, wording returned and any grouping shown with each measurement. Preserve an absent value as absent. If a service combines variants, keep that relationship visible instead of adding every row as separate demand.
Official planning guidance distinguishes historical metrics from forecasts that account for factors such as bids and budgets. Use the kind of data you actually obtained in your notes. A historical number in a keyword list is not our suggested basis for promising future visits or acquisition cost.
Choose the next step using relevance, evidence and your capacity to make a useful page together. For paid research, decide what additional measurement would change that decision before spending more. Leave an uncertain phrase unresolved when another estimate would not answer the real fit question.
Try the synthetic selection exercise
The keyword example CSV contains fictional phrases, intent notes and deliberately synthetic values. Our local CSV check preserved zero separately from a missing value and rejected an unsupported service phrase despite its larger example number. No row is a measured opportunity or a real customer query.
Open the example and write a reason for each decision using the intended page purpose. Then apply the same record format to your own sourced candidates. Add the evidence you need before moving an unresolved phrase into a page plan.
Finish with one clear primary question for each proposed page and a brief describing the related questions it should answer. Review existing pages for overlap before creating another one. Our suggestion is to revisit the plan when real reader feedback or performance evidence becomes available, rather than treating the first list as permanent.