Keyword Research Without Guesswork: From Query List to Page Map

Keyword research usually produces a spreadsheet with several thousand rows, sorted by search volume, which is then never opened again. The document is not the problem. The problem is that it is the wrong deliverable: a list of phrases cannot be executed, while a list of pages can.

The useful output of this work is a page map. For each page: what question it answers, who is asking, what it must contain, and whether it already exists. Everything before that is input.

Why starting with volume is the wrong order

Search volume is where most processes begin and it deserves less trust than it gets, for three reasons that compound.

• The numbers are estimates. Tool volumes are modelled from clickstream samples and rounded into buckets. Two terms shown as 320 a month may differ by a factor of three in reality.
• They aggregate loosely. Close variants are frequently merged, so the number describes a family of phrasings rather than the one you are looking at.
• Zero is not zero. Terms reported as having no volume routinely produce traffic, and in specialist B2B they are often the entire market. A phrase searched forty times a year by exactly the right person is worth more than one searched four thousand times by nobody who buys.

Four sources of demand that are not a keyword tool

Each of these is free, specific to your business, and better evidence than a modelled estimate.

1. Your own search log. Real customers describing what they want in their own words, unprompted. The single best source of vocabulary you have, and the case for reading it regularly is made in what customers type into site search.
2. Questions sales and support answer repeatedly. Every question asked five times a month is a page that does not exist. These convert unusually well because they arrive from people already deciding.
3. Search Console queries where you rank between positions eight and twenty. This is the highest-return list in the account: pages already considered relevant, one improvement away from a position that gets clicks. Nothing new needs to be created.
4. The questions competitors answer and you do not. Not their keyword list, which is unknowable, but their page titles, which are public and reflect what they decided was worth writing.

Sorting by what the page has to be

Classifying intent into “informational, navigational, transactional” is standard and not very actionable. A more useful question is what kind of page could satisfy the query, because that determines what you build.

The verification is free: search the term and look at what is ranking. If the first page is entirely comparison articles and you plan a product page, one of you is wrong and it is probably not the search engine.

One page per intent, not one page per phrase

The most common structural error in keyword work is a page for every variation. Ten phrases meaning the same thing become ten thin pages that compete with each other, split whatever authority exists, and leave the search engine to pick one arbitrarily.

Group phrases that would be satisfied by the same page. If a single well-made page would leave a person who searched any of them satisfied, that is one page. The grouping does not need software: sort the list alphabetically, read it once, and mark the clusters by hand. For a few hundred terms this takes an hour and is more accurate than automated clustering, because you know the business.

Where two existing pages already target the same intent, the answer is usually to merge them rather than to differentiate them. Merging concentrates the signals and removes an internal competitor; differentiating produces two mediocre pages and a maintenance obligation.

Priority, when everything looks worth doing

A finished cluster list is always longer than the capacity to write it. Three factors decide the order, and volume is not the first.

Commercial proximity. How close is this query to a purchase? A page answering a question people ask a week before buying outperforms one answering a question they ask a year before, even at a tenth of the volume.

Ability to win. Look at who ranks now. If the first page is entirely national publishers and marketplaces, a new page from a small business will not displace them, and the effort belongs somewhere else. This is the most commonly skipped check and the most expensive to skip.

Cost to produce. A page you can write from existing internal knowledge is cheaper than one requiring research, and cheap pages that are merely good beat expensive pages that never get finished.

Multiply rather than add. A high-volume term with no commercial proximity and no chance of ranking scores zero on two of three, which is the correct answer.

Judging whether you can actually rank

“Ability to win” sounds subjective and is not. Four observations, taking about two minutes per term, give a reliable answer without any paid tool.

1. Who is on the first page. If it is national media, marketplaces and government sites, a small business page will not displace them for that query. Choose a narrower variant instead: the same topic with a qualifier that the large sites have no reason to cover.
2. Whether anyone has answered it specifically. If the ranking pages all mention the topic in passing inside broader articles, a page dedicated to exactly that question has a genuine opening. This is the most common winnable case.
3. How old the results are. A first page of articles dated three years ago in a field that has changed is an opportunity. A first page updated this quarter is a warning.
4. Whether you have something they do not. Real data, real photographs, a real case, a local specific. Without one of these, a new page is a slightly different version of what already exists, and there is no reason for anyone to prefer it.

The fourth is the decisive one. If you cannot name the thing your page will contain that the current results do not, the honest conclusion is that the page should not be written, and the effort belongs on a term where you can.

What changed now that assistants answer too

Two adjustments are worth making, and neither is a reason to abandon the method.

The first is that conversational queries are longer and more specific than the phrases tools report, because people type differently when they expect an answer rather than a list. Tools systematically under-report these, which strengthens the argument for the four evidence sources above rather than weakening the method.

The second is that a page can be useful without receiving the click. Being the source an answer draws on has value that does not appear in your traffic report, which is uncomfortable for measurement and does not change what makes a page good. The measurement consequences of that shift are set out in what the zero-click numbers actually measure.

The practical implication for this process is narrow: write pages that answer a specific question completely and early in the text, rather than pages that circle a topic. That was already the right advice.

The page map itself

The deliverable is a table with one row per page, and six columns are enough.

1. The question, phrased as a customer would ask it.
2. The page type from the table above.
3. Exists or new, with the URL if it exists.
4. What it must contain to satisfy the question, in three bullets.
5. What it links to, which is how the map becomes a structure rather than a list.
6. Priority, from the three factors above.

The fifth column is the one that turns keyword research into information architecture, because a page map with no links is just a longer spreadsheet, and the naming discipline that makes those links findable is covered in naming menu items so people find things.

Measuring whether it worked

New pages take months to settle, and the common mistake is judging them on traffic in week three, concluding failure, and abandoning the programme.

Impressions move before clicks. A page accumulating impressions at position twenty-five is working; it is being considered and has not yet earned the click. Watch the position band rather than the visit count for the first quarter.

After that, judge on the same commercial measures as anything else: enquiries and orders by landing page, not sessions. A page that brings four hundred readers and no enquiries is a cost, however well it ranks, and the segmentation discipline behind that judgement is the one described in how to read conversion figures.

The short version

The deliverable is a page map, not a keyword list. Do not start with volume: tool numbers are modelled estimates, they aggregate variants loosely, and zero-volume terms often carry the whole of a specialist market. Take demand from four free sources instead — your site search log, the questions sales answers repeatedly, Search Console queries sitting between positions eight and twenty, and the questions competitors have decided to answer. Classify by what kind of page could satisfy the query and verify by looking at what already ranks. Build one page per intent rather than per phrase, and merge existing pages that compete. Prioritise on commercial proximity, realistic ability to rank and cost to produce, multiplied rather than added. Then judge new pages on impressions and position for a quarter before expecting clicks, and on enquiries rather than sessions after that.