Keyword research is not a hunt for the longest spreadsheet. It is a decision process: discover how people describe a problem, verify that the demand is real, understand what result they expect, and decide whether an existing or new page should serve it. The useful output is a page map with evidence, not a bag of phrases.
This workflow combines first-party query data, external estimates, search intent, and business value. It also answers practical questions such as how to find Google keywords, how to check keywords of a website, and when search volume should—or should not—change your publishing plan.
Start with a decision, not a tool
Before opening any keyword research tools for SEO, write down the decision you need to make. Are you choosing the next article, improving a category page, separating two mixed intents, or checking whether a competitor covers a missing topic? A defined decision keeps the research bounded.
Use four evidence levels. Observed demand is what your site already earns impressions for. Market hypotheses come from external databases, competitor pages, sales calls, and support conversations. Intent fit asks whether one page can satisfy the searches. Page feasibility considers expertise, product relevance, internal links, and the effort required to be genuinely useful.
This evidence ladder prevents a common mistake: treating keywords volume as a verdict. Volume is an estimate of a market, while Search Console impressions describe your visibility within that market. Neither number alone tells you what to publish.
Build a seed list from the language customers use
Begin with products, problems, jobs, alternatives, constraints, and outcomes. A seed such as “rank tracking” can expand into setup questions, comparison searches, troubleshooting, reporting, and buying intent. Add words from navigation labels, support tickets, sales notes, reviews, communities, and internal site search. They often reveal vocabulary that an SEO team would not invent.
To find keywords on a website, inspect its headings, navigation, title links, glossary, filters, and recurring entities. Do not copy every phrase. Record the topic, page type, audience, and promise. A competing article can prove that a subject exists, but it does not prove that the same angle fits your users.
Then expand the list with query suggestions, related searches, and a keyword checker website. Preserve the source beside each idea. Provenance helps later: a phrase mentioned by five customers deserves different treatment from one generated by a tool.
Use Search Console as the first-party baseline
Google’s Performance report lets you analyze queries, pages, countries, devices, clicks, impressions, CTR, and average position. Start with a useful date range—often three months for active sites and longer for seasonal or low-volume properties. Compare periods only when the business context is comparable.
Export query and page data with the GSC Export tool, then look at queries by landing page. This reveals three valuable groups:
- queries earning impressions but few clicks, which may indicate an alignment or snippet opportunity;
- queries spread across several URLs, which may indicate overlapping pages;
- queries attached to a relevant page but not answered clearly, which may justify a new section.
Google explains that query filters can use exact matching or regular expressions, and that anonymized queries are omitted from query tables. Therefore, Search Console is evidence, not a complete census. Use regex to group variants, but keep the raw query available for review. The Search Console common tasks guide also recommends focusing on clicks and impressions trends rather than relying on absolute position alone.
Interpret volume, ranking, and click opportunity separately
A phrase can show high monthly volume yet offer little realistic traffic because the results page is dominated by maps, videos, shopping units, or a direct answer. Another phrase with modest volume may be commercially precise and easier to serve. Keep at least four columns: estimated volume, current visibility, likely click opportunity, and business value.
The ranking of keywords is also contextual. Google defines average position as the topmost position of your property in results, averaged across impressions. It is not a fixed rank for every person or location. Read the metric with clicks, impressions, device, country, and page. Google’s metrics documentation explains how these values are calculated.
For checking keywords for a website, compare like with like: same country, language, device assumptions, and date window. Mark branded, navigational, local, and seasonal terms. Otherwise, a large number may describe a different market from the one your page can serve.
Classify intent before clustering phrases
Search intent is the result a user expects, not simply a modifier such as “best” or “how.” Inspect the current result set. Are the leading pages guides, product pages, category lists, tools, videos, or definitions? Note recurring subtopics and what the results omit. Similar wording can hide different intents; different wording can lead to the same answer.
Use a simple test: could one strong page satisfy both queries without awkwardly switching audience, task, or conversion goal? If yes, they can share a cluster. If one query asks for a definition and another asks for a software comparison, separate them even when the nouns match.
A useful cluster contains one primary need and supporting variants. Avoid creating a page for every singular, plural, or word-order change. Google’s SEO Starter Guide notes that search engines can understand variations; write naturally for people instead of forcing every exact string.
Map each cluster to an existing or new page
Now turn the research into architecture. For each cluster, assign one action: improve an existing page, merge overlapping pages, create a new page, or ignore for now. Check the Top Pages report before proposing a new URL. If an established page already earns relevant impressions and links, strengthening it is often safer than splitting its signals.
Use the Cannibalization tool to find cases where multiple URLs alternate for the same query group. Alternation is a diagnostic clue, not automatic proof of a problem. Confirm that the pages compete for the same intent before consolidating.
Give every planned page a concise brief: audience, job to be done, primary intent, evidence sources, unique contribution, required sections, internal-link sources, and success metric. A phrase list without a page promise tends to produce generic text.
Prioritize with an evidence score
Score opportunities on a small, explainable scale. One workable model is: demand evidence (0–3), business fit (0–3), authority and experience (0–2), and implementation effort (0–2, reversed so easier work scores higher). Add a confidence label based on the quality of the evidence. The score supports judgment; it does not replace it.
High-priority work usually combines visible demand, a clear audience problem, a credible answer, and a reachable next step. A high-volume topic with weak product relevance can wait. A lower-volume question repeatedly raised by qualified prospects may deserve immediate attention.
Keep a “not now” list with the reason. This prevents the team from repeating research and makes future reviews faster when the product, authority, or market changes.
Write for coverage, not phrase repetition
Once the page is assigned, outline the user journey. Answer the main question early, then cover decisions, edge cases, examples, and next actions. Use the vocabulary from research where it fits, but do not insert every exact phrase. Google’s guidance on helpful, people-first content emphasizes original value, clear expertise, and a satisfying experience—not content created primarily to manipulate rankings.
Connect the page to relevant tools and explanations. For example, query discovery can lead to the GSC Workspace, while CTR diagnosis can continue in CTR Lens. Internal links should help the reader perform the next task, not merely distribute anchor text.
Measure the page as a hypothesis
Record a baseline before publishing: current clicks, impressions, queries, pages, and conversion proxy. After Google has had time to crawl and reassess the page, review whether relevant impressions expanded, whether clicks followed, and whether unexpected queries reveal a missing section or a mismatched intent.
Do not rewrite after a few volatile days. Use comparable windows and account for seasonality, site changes, and search updates. The Search Console aggregation guide explains why totals can differ depending on whether data is grouped by property or page.
Research is complete when the next decision is clear. The durable loop is: observe, hypothesize, cluster, map, publish, and learn. That loop is more valuable than any static list of the “best keywords SEO” teams could collect.