Keyword research has a reputation problem. Done badly it produces a spreadsheet nobody opens, full of terms chosen because a tool said they had volume. Done well it is the single most useful document in a growth programme, because it decides where every subsequent hour gets spent.
The difference is entirely in what you optimise for.
What keyword research is actually for
It is not a list of words. It is a map of the questions your buyers ask between first realising they have a problem and choosing a supplier — ranked by how close each question sits to a purchase.
That framing changes the output. A search with 200 monthly queries and obvious commercial intent beats one with 20,000 informational queries for almost every B2B business, because you are buying pipeline, not pageviews.
Why this matters more in 2026
Two shifts have raised the cost of getting it wrong.
First, informational queries increasingly resolve inside a Google AI summary or an AI assistant without a click. Building a content programme on top of broad informational terms now buys less traffic than the same plan would have bought three years ago.
Second, the bar for ranking at all has risen. Publishing a mediocre page against a competitive term is no longer merely inefficient — it does nothing. Choosing the right target has become more important than producing more.
The seven-step process
1. Start with what you already have
Google Search Console is the most underused research tool available and it costs nothing. Filter for queries where you rank between positions five and twenty. Those pages already have some authority and relevance; moving them is usually cheaper than creating anything new.
2. Build the seed list from the business, not the tool
Write down what you sell, in the words your customers use rather than your internal terminology. Ask sales what prospects actually say on calls. This list is short and it is the foundation — tools expand seeds, they do not generate them.
3. Expand
Feed the seeds into a keyword tool for related terms, questions and variations. Include the "People also ask" boxes and related searches from the live results, which reflect what Google currently associates with the topic.
4. Classify by intent
Sort every term into one of four buckets:
- Transactional — ready to buy ("seo agency london", "hire a fractional cmo")
- Commercial investigation — comparing ("best seo agency", "seo pricing")
- Informational — learning ("what is technical seo")
- Navigational — looking for a specific brand
The first two buckets justify service and comparison pages. The third justifies guides, and should be judged against whether the query still earns a click at all.
5. Score difficulty honestly
Look at who ranks in the top five. If they are national publishers with substantial authority and your site is new, that term is not a realistic target this year regardless of what a difficulty score says. Difficulty metrics are useful for sorting, not for deciding.
6. Cluster
Group terms that should be answered by the same page. Search engines resolve many phrasings to the same intent, so one thorough page usually beats five thin ones — and five thin ones compete with each other, which is worse than doing nothing.
A cluster is defined by the answer, not by the wording. If two queries would be satisfied by the same page, they belong together.
7. Prioritise by value, not volume
For each cluster estimate: realistic achievable traffic, likely conversion rate given the intent, your average deal value, and the effort to compete. Rank by expected return. The result is almost never the same order as the volume column, and that is the point.
Understanding intent properly
The most common expensive mistake is matching the right term to the wrong page type. If the top ten results for a query are all comparison articles, Google has told you what that query means — publishing a product page against it will not work, however good the page is.
Read the current results before writing anything. They are the clearest available statement of what searchers want.
Competitor analysis
Competitor research answers two questions: which terms send them traffic that should be yours, and which questions they have not answered well.
The second is usually the better opportunity. Displacing a strong incumbent on their best term is expensive. Owning a cluster nobody has covered properly is not.
Tracking performance
Track clusters, not individual keywords. A single term's position bounces for reasons that have nothing to do with your work; a cluster's aggregate visibility is a real signal.
The metrics worth a monthly look: total clicks from the cluster, average position across it, conversions attributed to its landing page, and — for AI visibility — whether you are cited when the cluster's questions are asked of an assistant.
Then act on it: the pages sitting at positions five to fifteen are where the next month's work should go. See our content strategy guide.
Frequently asked questions
How many keywords should we target?
Are low-volume keywords worth targeting?
Do I need paid tools?
How often should keyword research be redone?
See which searches your competitors own
The audit maps the queries taking traffic to competitors instead of you, and what that gap is worth each month.