B2B Keyword Research: Volume Is the Wrong Denominator

B2B keyword research breaks for one specific reason: the number every tool puts in front of you is counting the wrong population. Search volume tells you how many people typed a phrase. It never tells you how many of them could sign your contract. In a market of 400 qualified accounts, those two figures sit so far apart that sorting a keyword list by volume is close to sorting it at random.

The fix has two parts. Rank the list by how much of your buyable population a term reaches. Then write for the group that makes the decision instead of the individual who ran the search. The second part contradicts what almost every guide on this topic recommends, and there is good data explaining why.

Search volume counts searches rather than buyers

Start with what the tool vendors say about their own numbers. Ahrefs founder Tim Soulo published a piece titled There's No Such Thing as "Accurate" Search Volume in March 2024, and it is blunter than anything a customer would write. Google Keyword Planner reports rounded annual averages sorted into volume buckets, with similar-meaning keywords grouped together and their volumes summed. Ahrefs blends multiple sources to split those clusters apart, but Soulo states plainly that none of their metrics are accurate. They are built to be directionally accurate: good enough to tell you one term has more demand than another, not good enough to tell you the real number.

Directional accuracy is fine at consumer scale. The gap between 40,000 and 60,000 searches does not change what you do. At B2B scale it stops being fine, because the gap between 90 and 140 monthly searches is exactly the decision in front of you, and it sits inside the error bar.

The obvious escape hatch is to assume the small numbers are wrong in your favour. Ahrefs tested that too. Joshua Hardwick compared low-volume keywords the Ahrefs blog ranked in the top five against real Search Console impressions and found the zero-volume optimism mostly unfounded: most drove the same or fewer impressions than estimated, fewer than 1% cleared 100 impressions, and the average came in at 11.3 impressions against an estimate of roughly 10. Where a page did pull real traffic, it was because the term was a long-tail variant of a popular topic rather than a hidden pocket of demand.

So the small numbers are genuinely small, and they are roughly right. Precision is not the problem worth solving here. Relevance is.

The number to rank your list by is the buyable population

Write the denominator down before you open a keyword tool. How many companies match your profile? How many people inside each one touch the decision? For a firm selling to B2B companies at roughly one to ten million in revenue with a founder still running sales, that might be 400 companies with three relevant people each. The entire universe of humans who can ever buy from you is 1,200 people. No keyword can send you more than a slice of 1,200.

Once that ceiling is on paper, a head term stops looking like its volume. Run two examples through it.

A category head term does 1,900 searches a month. Who is searching it? Marketing students, agency staff doing client research, enterprise teams ten times your size, and a long tail of people outside your country. If 3% of those searchers are in profile, the term reaches 57 buyable people a month.

A narrow operator term does 90 searches a month, phrased the way only someone personally running their own pipeline would phrase it. If 60% of those searchers are in profile, it reaches 54 buyable people a month.

Fifty-seven against fifty-four. The terms are 21x apart on volume and within touching distance on the number that matters, and the second one you can rank for in weeks rather than quarters. The figure to sort by is in-profile searches per month: volume multiplied by the share of searchers who could plausibly buy.

You will never measure that share precisely. You can bracket it, and bracketing it beats ignoring it. Even a crude estimate written next to each term reorders the list, which is more than raw volume does. This is a different exercise from scoring keywords on deal size, and it produces different answers. Two terms can carry identical contract value and still differ tenfold in how much of your account list they reach. Both belong in a full B2B SEO strategy, and the population figure is the one people skip.

The searcher is a committee, and it is larger than you think

The other reason volume misleads in B2B is that one purchase generates searches from several different people with different jobs.

Gartner surveyed 632 B2B buyers between August and September 2024 and found buying groups ranging from five to 16 people across as many as four functions. Seventy-four percent of those teams showed what Gartner calls unhealthy conflict during the decision: members with competing objectives, disagreement on the right move, or being overruled from outside the group.

An earlier Gartner survey of 1,120 technology buyers put a sharper edge on it. Sixty-seven percent of people involved in technology-buying decisions are not in IT, which is to say the person researching a technical purchase is usually not the person who owns the function it belongs to.

For keyword research, the consequence is that a single phrase carries several different jobs depending on who typed it. "Lead routing software" from an operations manager who already knows the category is a shortlist query. The same phrase from a founder who just watched two leads go cold is a "do I even need this" query. The tool reports one number and one intent. There are at least two.

Why a page per persona is the wrong response

The natural conclusion is to build a keyword set per role and a page for each. That is where nearly every guide ranking for this topic lands, and it is where the Gartner data points the other way.

In the same 2024 survey, tailoring content to the buying group improved consensus by 20%. Content tailored to individual-level relevance had a 59% negative effect on it. Gartner's reading is confirmation bias: material written to suit one member reinforces the view that member already holds, which makes the group less likely to converge on a shared direction. And convergence is the part that pays. Groups that reached consensus were 2.5 times more likely to report a high-quality deal.

So persona-mapped content architecture optimizes the wrong step. It wins the individual read and taxes the group decision. That tracks with how B2B content actually travels: the operations manager who found your page forwards the URL to the founder, and the founder reads the same document. If it was written to flatter the manager's priorities, it reads as noise to the person holding the budget.

The working rule is one page per decision rather than one page per persona. A page targeting a category term should answer the practitioner's mechanics question and the economic buyer's justification question inside the same document, because both are going to open the same URL. The place to enforce that is the brief, not the draft. Ours carries the forward test as a standing requirement, and the structure is in our SEO content brief template.

Where the query list actually comes from

If volume cannot rank the list and keyword tools cannot see the committee, the seed language has to come from records of real buying. In order of signal quality:

Closed-won deals. The exact words in the first inbound message, and in the discovery call notes before anyone taught the buyer your vocabulary. This is the highest-value keyword source most companies own and never read.

Your own Search Console query data. Real queries from real in-market people, already filtered to those who reached you. It is the only free source of ground truth about your buyable population, and it needs nobody's permission to access.

Lost deals. Objection language converts directly into commercial-intent terms, and almost nobody mines it. A prospect who said your setup looked too heavy for their team just handed you a comparison query.

Support and onboarding tickets. The questions people ask after they buy are the questions the next buyer will search before they do.

Every guide on this SERP recommends talking to sales. Few say what to do when sales will not hand over call recordings, which is the ordinary case at most companies. Search Console and your own CRM close-won records are available without a negotiation. Start there, and use the won-deal language to interpret what the query data shows.

Competitors come in as a cross-check rather than a source. Once you have a list grounded in your own deals, competitor keyword research tells you which of those terms someone else already answers well, which is a build-order question rather than a targeting one.

Which B2B keywords still send a click

A correctly ranked list still has to survive the results page, and that has moved.

Rand Fishkin's analysis of Similarweb clickstream data found that in the first four months of 2026, 68.01% of US Google searches ended without a click, up from 60.45% in 2024. Ahrefs, working from aggregated Search Console data across 422,421 websites, reports that when an AI Overview is present, clicks to the top organic result fall by roughly 58%, a sharp jump from the 34.5% drop they measured earlier. They also found AI Overviews on 57.9% of question queries, with 99.9% of them sitting on informational intent.

Read together, that is a sorting instruction. Definitional and how-to B2B terms are the most exposed, because they are exactly the queries an AI answer handles well. Comparison terms, alternatives terms, pricing terms and category-plus-buyer terms are far less likely to trigger an AI answer, and they sit closer to a purchase regardless.

That is not an argument for abandoning informational coverage. Being the source an AI answer is built from still moves the decision even when it sends no visit. It is an argument for not counting those terms as traffic. Keep two lists: terms you expect to send visits, and terms you expect to be cited for. Reporting them as one number is how a program that is working looks broken. The SEO and AEO checklist covers the citation side in more detail.

The sequence

Write down the buyable population. Rank terms by how much of it each one reaches. Build one page per decision rather than one per persona. Take the seed language from deals you already closed. Volume goes last, as a tiebreaker between two terms that are otherwise equal.

Done in that order, the output is a short list of terms that reach people who can actually buy, which is a very different document from the 200-row spreadsheet a keyword tool hands you. It is also only worth building if the rest of the system catches what it produces. A term reaching 54 in-profile buyers a month earns its place when the page books a conversation and the follow-up runs when it does not. That is the job of a search and AI-search visibility engine rather than a content calendar, and if you want to know where yours is leaking before you write anything, the Growth System Score grades it in 48 hours without a call.