Best keyword strategy: how to decide what to chase, in what order, and when to stop adding

The best keyword strategy is the one that commits your next ten pages to queries you can realistically win, in an order that pays for the one after it. It has three parts: a portfolio view of what you already hold, a shortlist of what to enter next, and a stopping rule. A list of three thousand keywords with volume beside each one is not a strategy, because it contains no decision. The strategy is what gets crossed out.

Most teams reach for volume first and discover that the biggest queries are held by sites with ten years of links. The sequence that works runs the other way: start with the queries where you are already close, use the traffic and the links those pages earn to buy credibility, then move up. Each stage funds the next.

Part one: sort what you already hold

Before a single new keyword is added, split the queries you already rank for into four buckets. Search Console gives you all of this for nothing, and its performance report is the only source here that counts your actual impressions rather than modelling them.

Bucket	What it looks like	The move
Defend	Positions one to three, commercial intent	Keep the page current, watch for a competitor's new page, do nothing clever
Nearly there	Positions four to fifteen, real impressions	Improve the existing page. This is the cheapest traffic on the board
Wrong page ranking	Impressions split across two or three of your own URLs	Pick one page, merge or redirect the rest
Enter	No impressions at all	Needs a new page, and needs justifying against the three buckets above

The third bucket is the one that quietly costs the most. Two of your own pages aimed at the same query divide the clicks and the links between them and neither ends up strong. Google's guidance on consolidating duplicate URLs covers the mechanics of picking the one that survives. A useful internal rule: one query, one page, and the page's title says which query it is.

Part two: pick what to enter, on four tests

• Can we answer it better than what ranks now? Open the top five results and write down what each one is missing. If the honest answer is "nothing", skip the query, however good the volume looks.
• Does the searcher want what we sell? A query with ten thousand searches and no buying intent is a traffic number, not a pipeline. The intent types matter more here than the volume figure.
• Is the competition a site like ours? If the first page is all national publishers and aggregators, the honest entry point is a narrower version of the same query.
• Does it fit a cluster we are already building? A query that sits next to five pages you already own costs less to win than an identical query standing on its own, because the internal links and the topical overlap are already there.

Long tail entries are where most sites should spend their first two quarters. Low competition queries are findable in Search Console, in your own site search logs, in sales call notes and in the questions people ask under competitors' videos. Keyword Planner then gives you a demand figure for the ones you found, which is the right order to use it in.

Part three: build clusters, not lines

Group the shortlist into page sets. A cluster is one hub page on the broad term plus five to twelve pages on the specific questions inside it, all linked to each other. The reason to work this way is practical rather than theoretical: specific pages rank faster, each one links up to the hub, and the hub becomes winnable six months later because of them.

A worked shape, for a company selling bookkeeping software to tradespeople:

• Hub. Bookkeeping for tradespeople.
• Spokes with buying intent. Bookkeeping software for electricians. Invoicing app for plumbers. CIS deductions in bookkeeping software.
• Spokes with research intent. How to price a job and keep the margin. What receipts to keep and for how long.
• Spokes with comparison intent. Spreadsheet against software for a two van business.

Each spoke targets one query and links to the hub with words that describe the hub. The hub links out to every spoke. That is the whole mechanism, and it is worth more than any individual keyword choice.

Sequencing: what order to do it in

1. Weeks one to four. Fix the bucket three overlaps and improve the five nearly there pages with the most impressions. No new pages at all.
2. Weeks five to twelve. Build one full cluster. One hub, six to eight spokes, properly interlinked.
3. Month four. Measure the cluster as a unit, not page by page. Impressions, average position, clicks and the one conversion that matters.
4. Month five onward. Start the second cluster, and fold the lessons from the first into the brief. Re-run the bucket sort every quarter, because the nearly there list refills itself.

The stopping rule, which most strategies are missing

Decide in advance how many queries you will hold. A good number for a small team is one query per published page plus a waiting list of twenty. Everything beyond that goes in an archive file, not the plan. The reason to cap it: an uncapped list makes every planning meeting a re-litigation of the whole list, and the pages that would have won get delayed by the discussion. A capped list means the meeting is about two decisions, which takes ten minutes.

The companion artefact to all of this is a sheet with the right columns, and the keyword analysis template covers which ones turn a list into a decision. If the question is which page to build first rather than which query to own, the picking order is the sharper read.

Keyword strategy: the arguments that come up in the planning meeting

Should we chase high volume keywords or long tail ones?

Long tail first, high volume as the destination. Specific queries convert better, rank faster and build the topical depth that makes the broad term reachable later. Put the broad term in the plan as a hub page you expect to win in year two.

How many keywords should one strategy contain?

One primary query per page, plus the handful of close phrasings that belong on the same page. For the plan itself, a working list of twenty to forty entries is enough for a small team for a year, and anything larger becomes an archive rather than a plan.

How often should the strategy be rebuilt?

Review quarterly, rebuild annually. The quarterly pass is the four bucket sort, which takes an afternoon in Search Console. The annual pass is the one where you check whether the clusters still match what the business sells.

What do we do when a competitor outranks us on a query we hold?

Read their page against yours and name the specific thing it does better: a worked example, a price table, a calculator, a more current date, a clearer answer in the first paragraph. Fix that one thing on your page. Rebuilding from scratch is rarely the shorter route.

Should paid and organic keyword lists be the same list?

Keep one list with a column for channel. Paid data tells you which queries convert before you have ever ranked for them, which is the fastest intelligence available for an organic plan, and splitting the lists loses that link.

Is keyword difficulty a reliable way to choose?

Treat it as a sorting aid rather than a verdict. The difficulty score is each vendor's model of the link profiles ranking today. The useful check is the one you do by hand: open the top five results and decide whether a page like yours belongs among them.

Buyers who ask a question instead of typing a keyword

Every part of the method above starts with a query somebody typed into a search box. A growing share of the same buyers now type the question into ChatGPT, Gemini, Perplexity or Claude instead, in a full sentence, and get back three or four named options rather than ten links. Those conversations leave no impression in Search Console and no row in any keyword tool, so a strategy built only from keyword data has a blind spot where that demand sits.

The practical question for a planner is simple: when somebody asks an assistant the exact question your hub page answers, which companies get named? A tool like AI Knows Us exists to answer that one, running the buyer questions through the assistants and reporting what came back, along with the pages each answer drew on. Reading those answers tends to change the cluster you build next, because the assistant usually names the source it found easiest to quote rather than the site with the most links.