Vocabulary research with a commercial filter
Keyword research is the work of finding the words a market uses for a problem, a product and a purchase, then deciding which of them justify a page. The keyword phrases themselves are the visible output; the valuable part is the map of who says what, and why.
It gets confused with two other activities. The first is exporting a tool's suggestions and sorting by volume, which produces a list rather than research. The second is the counting exercises — keyword density, exact-phrase repetition — that stopped describing how search works years ago. Google's own advertising documentation treats semantically equivalent queries as one pool of demand: "By default, all keyword match types are eligible to match to close variants. There's no way to opt out." Close variants for exact match include synonyms, paraphrases, reordered words and queries with the same search intent. A page written around a single exact string is solving a problem that no longer exists.
Four groups, one product, four names
The same equipment or software carries a different name in every room it is discussed in, and only some of those names get typed into a search box.
| Group | How they name it | What that tells you |
|---|---|---|
| Buyers and operators | By the problem or the outcome | Highest commercial value, lowest visibility in keyword tools |
| Engineers | By mechanism, specification and standard | Precise, low volume, strong intent once the category is settled |
| Procurement | By category, contract vehicle and supplier status | A shortlist exists; comparison and credential pages win here |
| Vendors and analysts | By market category shorthand | Useful for competitive research, often searched by nobody outside the industry |
Collect the real language where it appears unprompted: recorded sales calls, support tickets, the question sections of recent requests for proposal, your own site search log, and the paid search terms report. Read competitor specification sheets and the relevant standards for formal terms. Then check each phrase against a tool — not to discover it, but to see whether anyone outside your building uses it.
Where a search volume number comes from
Search volume in Google Keyword Planner is a modelled monthly average, commonly reported in rounded bands rather than exact counts and averaged across a period rather than observed live. Third-party tools blend Planner figures with clickstream panels and their own models, which is why three tools return three numbers for one phrase. Treat any volume figure as an order of magnitude with wide error bars, never as a count of buyers.
Two documented facts show how thin the data gets. Google Ads reports no Quality Score at all — a dash — when, in its own words, "there aren't enough searches that exactly match your keywords to determine a keyword's Quality Score." That is routine in low-volume business accounts and is not a fault. Search Console, separately, withholds low-volume and anonymised queries, so even your own data is incomplete at the tail.
The consequence for a B2B keyword plan: for specialised phrases, absence of volume data is not evidence of absence of demand. A tool reporting a dash is reporting the limits of its sample.
Long tail phrases that carry commercial weight
A long tail keyword is one with low individual search volume; length is a symptom of that, not the definition. Two very different things hide under the label. The first is a specific phrase used by few people because few people have the problem — and those few buy. The second is a long string nobody searches at all, which appeared in your export because a tool can generate combinations.
The qualifiers that indicate money in business markets are recognisable once you look for them:
- a standard, certification or regulation named in the query
- a part number, model number or material specification
- an integration pair, one named system with another
- replacement, retrofit, end-of-life and migration language
- an industry qualifier: for water utilities, for aerospace, for hospital laboratories
- comparison and alternative phrasing, which usually means a shortlist already exists
The opposite signals are equally clear: definition queries, salary and career phrasing, coursework language, and the words free or template attached to something you sell. Those phrases report healthy volume and produce nothing.
Difficulty scores describe your competitors, not the search engine
Keyword difficulty is a vendor calculation, not a Google metric. Each tool computes it differently, mostly from link-based signals attached to the pages currently ranking, which is why one phrase scores 34 in one tool and 51 in another. Used inside a single tool, as a way of ordering a long list, it is a reasonable proxy for how well resourced the incumbents are. Used as a law of nature, it misleads.
What a score cannot know is decisive in B2B. It cannot tell whether the ranking pages answer the query or merely mention the words. It cannot see that your company holds twenty years of specialised authority the incumbents lack. It cannot tell that the top five results are distributor catalogue pages, a forum thread and a brochure PDF — a results page a genuinely useful page would displace.
Use the score to sort, then open the results page and read it. A page full of near misses is the strongest opportunity signal available, and no tool reports it.
Group phrases by the job the searcher is doing
Grouping by volume produces a plan that mirrors a spreadsheet. Grouping by the searcher's job produces one that mirrors a purchase. Gartner's framing helps: buyers work through six jobs — problem identification, solution exploration, requirements building, supplier selection, validation, and consensus creation — revisiting them rather than passing through in order.
- Problem identification and solution exploration. Symptom and failure-mode phrasing, served by pages that name the problem in the buyer's words.
- Requirements building. Specification, sizing and compatibility phrasing, served by technical detail and data tables.
- Supplier selection and validation. Comparison, alternative and supplier-status phrasing, served by honest comparison pages and evidence.
- Consensus creation. Business case, total cost and risk phrasing, served by material a champion can forward to a committee.
One page can serve many phrases and should. Because search engines match meaning rather than strings, a well-built page routinely earns positions for dozens of variants nobody listed. The test for splitting is simple: would one page satisfy both searchers? If yes, build one. Two pages made for near-identical phrases divide their references and internal links, then compete with each other.
Mistakes that appear in almost every keyword deck
- Research done entirely inside a tool. Tools report what many people already search for, which in a specialised market is the vocabulary of outsiders. Start with recorded calls and tickets; use the tool to validate.
- One phrase per page. It produces dozens of thin near-duplicates that compete with each other and read as filler. Build one substantial page per job the searcher is doing.
- Deleting every phrase with no reported volume. In business markets the highest-intent phrasing often sits below a tool's reporting floor. Judge it on whether a qualified person would type it.
- Treating difficulty as a verdict. A score is a proxy for competitor strength. Open the results page before accepting it.
- Never revisiting the list. Product names change, analysts rename categories, and acquisitions rewrite vocabulary overnight, so a two-year-old plan quietly targets words nobody uses.
- Chasing a competitor's entire keyword footprint. Much of what a rival ranks for is recruitment, support and legacy product traffic that earns them nothing, and copying it buys you the same dead weight.
What to measure, and when the research is the wrong task
Measure keyword work by group rather than by phrase: impressions and clicks per job-based cluster, entrances to the page built for it, form fills, and qualified opportunities. Two numbers deserve to be ignored. Average position across a portfolio hides everything, since one page rising from 40 to 12 and another falling from 3 to 8 net out to nothing. Total keywords ranking is a vanity count inflated by irrelevant variants.
There are times to skip the exercise. If you have created a category nobody has named, there is no vocabulary to find — the words have to be invented and taught, which is demand creation. If your market is a hundred named accounts reached through relationships and tenders, a keyword plan is a modest side project rather than a strategy. And if the site cannot publish new pages this quarter, stop researching: a plan with nowhere to land is a document, not a programme.
Frequently Asked Questions
What is a good keyword difficulty score to target?
There is no universal number, because difficulty is a vendor calculation rather than a Google metric and every tool scales it differently. A 40 in one tool is not a 40 in another, so the figure only means something when comparing phrases inside one tool.
A more reliable method: shortlist by difficulty to save time, then read the actual results page for each candidate. If the top results are specific, current and well resourced, the phrase is hard whatever the score says.
How do we find keyword phrases our competitors are not targeting?
Start where competitors are not looking, which is inside your own company. Support tickets, recorded sales calls, requests for proposal and your site search log contain phrasing no keyword tool suggests, because the volume is too low to register. Then search those phrases and read what ranks.
The reliably neglected categories are failure modes and symptoms, part and model numbers, standards references, integration pairs, end-of-life and replacement queries, and vertical-specific phrasing. Competitors optimise head terms because that is what a volume-sorted export shows them.
Should we target keyword phrases with no search volume data?
Often yes, if a qualified buyer would plausibly type them. Volume figures are modelled estimates with a reporting floor, so a phrase showing zero or a dash may simply sit below that floor rather than being unsearched. Google Ads has a documented equivalent: it reports no Quality Score when there are too few exactly matching searches.
The sensible approach is to serve those phrases inside a page built for a broader job rather than to build a dedicated page for each, so genuine but invisible demand gets covered without creating a thin page.
How often should keyword research be redone?
A full review once a year suits most business markets, with a lighter quarterly check on the phrases you actively target. Vocabulary in specialised industries moves slowly, and rebuilding the plan more often produces churn rather than insight.
Three events should trigger a review regardless of the calendar: a product or brand renaming, because your own vocabulary just changed; an acquisition on either side, because a competitor's category name may become yours; and a standards or regulatory change, since a new standard number becomes a searched phrase almost immediately.