Learning how to choose the right keywords for SEO comes down to rejecting most of your list. A keyword is worth writing about only when it passes six filters: correct market index, matching search intent, beatable page-one floor, real business value, cluster fit, and no cannibalization with a page you already own. Volume is not on that list, because a high-volume term you cannot win and cannot monetize is worse than no keyword at all.
Run every candidate through the six filters below in order. Most lists lose 70 to 90 percent of their rows, and what remains is a build order rather than a spreadsheet.
Filter 1: Is the data from the right market?
Check this before anything else, because it invalidates everything downstream. Keyword volumes are index-specific. A UK, Australian, or Singapore business querying the US index sees inflated numbers for terms nobody in its market searches, and near-zero data for terms that are actually driving its competitors’ revenue.
Set the country index at the start of the project. If your target market is a single city or a small country, expect much lower absolute volumes and do not compare them against US benchmarks — 200 searches a month in a small market with high commercial intent can be worth more than 20,000 informational searches you will never convert.
Filter 2: Does the intent match what you can publish?
Search intent decides the format, and the format is not negotiable. Open the live results and look at what Google is already rewarding:
- Nine product or category pages — this is a transactional query. A blog post will not rank. Point a commercial page at it or drop it.
- Guides and tutorials — informational. Write the article.
- Listicles and comparisons — commercial investigation. Highest value per visit for most businesses, and usually the least crowded.
- Local pack dominant — you need a location page and a Business Profile, not content.
Ten seconds of SERP inspection per candidate is the single highest-return habit in keyword selection. It also catches the terms where an AI Overview now resolves the query completely, which changes the calculation on click-through even when ranking is achievable.
Filter 3: Can you beat the weakest page-one result?
Difficulty scores sort a list; they do not predict outcomes. Every vendor computes difficulty differently, and none of them know your site. The reliable test is specific: find the page sitting at position ten and assess whether you can outbuild it.
Ask four questions about that page. How thorough is it, honestly? How many referring domains point at that URL, not just the domain? Does it match intent better than what you would publish? And is the site comparable to yours in size and authority? If the tenth result is a 700-word post on a domain no stronger than yours, that term is winnable in a couple of months regardless of the difficulty number. If page one is nine 3,000-word guides from established brands, choose a different term and revisit it in a year.
Filter 4: Would this reader ever buy from you?
Traffic that never converts costs money to produce and maintain. Score each surviving term on how close the searcher is to a purchase decision:
- High: “best X for Y”, “X vs Z”, “X pricing”, “X alternatives”. Low volume, high value, often neglected because the numbers look unimpressive.
- Medium: problem-aware queries where your product is a plausible solution — “how to fix [problem your product solves]”.
- Low: broad definitional terms. Worth some coverage for authority and internal linking, but never the bulk of a plan.
A healthy plan for a business site runs roughly 40 percent commercial and 60 percent informational. If your list is 95 percent informational, you have built a magazine, not a growth channel.
Filter 5: Does it strengthen a cluster?
Isolated pages rank slowly. A term that fits an existing topic cluster inherits context from the pages around it through internal links, and complete coverage of a topic is itself a signal of expertise. Group your surviving terms into clusters of five to eight around a pillar concept, and prefer a term that completes a cluster over a higher-volume term that starts a new island.
This is also where publishing order matters. Build one cluster fully before starting the next; a half-finished cluster gives you neither authority nor a clean internal link structure.
Filter 6: Are you about to compete with yourself?
Cannibalization is the quiet killer of mature content libraries. Before adding a term, check whether an existing page already targets it or something close. In Search Console, filter Performance by that query and see which of your URLs collects the impressions. If two pages alternate, Google is unsure which one to rank, and both underperform what a single consolidated page would achieve.
The symptom is easy to spot once you look for it: a keyword whose average position sits at 14 while two URLs each collect part of the impressions, neither ever settling. Combined, that content would frequently sit inside the top ten, because the links, the engagement, and the topical signals stop being split.
The fix is usually consolidation rather than a third page — merge the thinner page into the stronger one, redirect it, and target the new angle only if it is genuinely a different intent.
What about volume, difficulty, and the other numbers?
They are inputs to the filters, not filters themselves. Understand what each one actually measures before you weight it.
- Volume is a modeled figure, usually a rolling twelve-month average. Seasonal terms therefore read flat when real demand spikes. Use it to compare two terms in the same niche, not to forecast traffic.
- Difficulty is a vendor formula weighted mostly toward the link profiles of the current top ten. Two tools will score the same keyword differently and both are internally consistent. Treat it as a sorting key that tells you which twenty terms deserve a manual SERP check.
- CPC is the most honest number on the export, because advertisers are spending real money against it. A high CPC on a modest-volume term is strong evidence of commercial value.
- Global volume matters only if you sell globally. For a single-market business it is a distraction that makes weak terms look strong.
- SERP features tell you how much of the click is even available. A query with an AI Overview, a video carousel, and four ads above the first organic result has a very different click ceiling than a clean ten-blue-links page.
A useful discipline: never let a keyword into the plan on the strength of a number alone. Every term that survives should have a one-line justification you could say out loud — “three competitors rank for it, the tenth result is thin, and it names our product category.”
Putting the filters to work
In practice this runs fast. Pull 300 to 800 candidates from multi-seed keyword exploration and a competitor content gap report. Kill everything from the wrong index. Sort by intent and drop the mismatches. Spot-check page-one floors for the top 60 survivors. Score those on business value, group them into clusters, and check each against Search Console for cannibalization. You will end up with 20 to 40 terms — one quarter of work — ranked in the order you should publish them.
Keep the shortlist where it feeds the rest of the workflow instead of dying in a spreadsheet. SEO Rocket saves selected terms to a project keyword pool that both the AI writer and the rank tracker read from, with content gap analysis across up to five competitors feeding the same pool. However you store it, revisit the list quarterly: markets move, competitors publish, and a term that was unwinnable in January is sometimes wide open by July.
