B2B Keyword Research When the Volume Column Says Zero

Zero volume is not zero demand

Zero search volume is not zero demand. In B2B the volume column is the least reliable number in the spreadsheet, and sorting by it deletes exactly the queries that produce contracts.

Here is the arithmetic that should replace it. The example is invented, so use your own figures: a German manufacturer sells a predictive-maintenance retrofit for packaging lines, deals land somewhere near €40,000, and the keyword tool returns N/A for almost every phrase the sales team hears on calls. On a consumer brief that project is dead before it starts.

Run the multiplication anyway. Assume the query genuinely gets around 20 searches a month. Rank in the top three, take roughly a third of the clicks, and you have six visits a month, seventy-odd a year. If one in twenty visits becomes an enquiry and one in four enquiries closes, that single page produces about one deal a year. One page. One contract. On a query no tool will score.

Every number in that chain is a placeholder, and we say so to clients' faces. Substitute your average deal, your enquiry rate, your close rate, then run it again. That calculation decides your core, not a column labelled Volume.

The volume number is wrong, not just small

"Volume isn't everything" is a platitude. The mechanism is the useful part.

Keyword tools estimate volume from clickstream panels and calibrate against paid-search auction data. Corporate networks, VPN'd office traffic and locked-down browsers barely register in those panels, and that is precisely where B2B research happens. Thin B2B ad auctions give the estimator almost nothing to calibrate against. Tools then round what survives into buckets, so anything genuinely small collapses into "10" or into nothing. On top of that, the same demand splits across singular, plural and word-order variants the tool never merges, so several rows of zero can be one real phrase with real traffic behind it.

Practical consequence: read volume as presence or absence. Someone, somewhere, typed this. Then stop sorting by it.

Keyword difficulty deserves the same demotion. KD is built largely from the backlink profiles of the pages currently ranking. In B2B those SERPs are half-empty and weakly contested, so the score describes a competition that isn't happening. Using it to rank tiny-volume queries is a category error.

What replaces the volume column

You still have to cut a long candidate list down to something the production budget can carry. We score every row on five questions, 0 to 2 each.

• Deal proximity. How many steps from this query to a signed contract?
• Who types it. Someone inside the buying committee, or a student writing a thesis?
• SERP shape. Does the current top ten hold companies we resemble, or only directories, marketplaces and aggregators we cannot displace?
• Can we write it credibly. Is there a case, a measurement, an engineer who will actually talk to us? If not, cut it. A thin page on a 20/mo query is pure loss.
• Shelf life. Written once, still true in a year?

Ten points maximum. This is a decision aid, not a measurement, and two sane people will score the same row a point or two apart. Its value is that it drags the argument into the open instead of hiding a judgement call behind a number the tool made up.

Where the queries come from when the tools are blank

Six sources, listed in the order we trust them.

• Recorded sales calls and lost-deal notes. Highest signal, least mined. Not "talk to sales": pull the last twenty discovery calls and the last ten losses, and listen to the first thirty seconds, where the prospect describes the problem in their own words before anyone teaches them your terminology. No sales team, or no access? Sit in on two calls yourself, or read the last batch of inbound emails.
• The client's inbox and CRM. How did inbound leads phrase it in their very first message?
• Search Console, at the impressions level. Queries with impressions and near-zero clicks are demand that exists and that no tool ever showed you.
• Competitor SERPs. Someone paid to build that page. That is a signal, even when it only reveals what they believe.
• Paid search term reports. Real matched queries, including ones with no reported volume anywhere.
• AI assistants. The phrasings ChatGPT or Perplexity surface for the category. A legitimate proxy in 2026 and the weakest of the six, because it reflects a model's summary rather than a buyer.

Map the buying committee, not a funnel

One need, unplanned line stoppages, fractures by role. The maintenance engineer searches capability and integration: vibration sensor retrofit existing PLC. The plant manager searches outcomes and comparisons. Procurement searches contract shape, compliance and alternatives to a named vendor. The executive sponsor searches risk and proof.

The part nobody writes down is that not every role earns a URL. Our rule: does this role initiate the search, or only validate a shortlist someone else built? Initiators get pages. Validators get a section inside an existing page, because by the time they search, the decision is largely made and a new URL changes nothing.

Then place each query at the moment in a real purchase where it gets typed. Problem not yet named. Category being defined. Shortlist being built. Internal justification being written. Contract being negotiated. Everyone skips the fourth. That is your champion hunting for material to forward to their boss, and those are the highest-leverage, lowest-volume queries in the entire core.

One query, one page, and how cannibalisation really starts

Cannibalisation in B2B rarely comes from duplicate topics. It comes from near-synonyms the team sincerely believed were different: X software, X system, X platform, X tool. From a service page and a blog post both written to sell the same thing. From industry and location variants spun out mechanically because the template made it easy.

The test takes two minutes. Open both SERPs and compare the top ten. If the same URLs dominate both, Google reads them as one intent, so build one page whatever the spreadsheet says. Different URLs and different page types mean two pages.

Do this at design time. Afterwards the repair is consolidate-and-301 or a genuine split of intent, and both cost several times what the decision would have cost upfront. You also surrender the rankings you spent months earning.

When the client's word isn't the market's word

Founders and sales teams use a precise, sometimes proprietary term. The market types something cruder. Build the core on the internal term and you ship a beautiful set of pages nobody finds. Flatten everything to the generic term and you lose the buyers who use the professional word, who are usually the ones holding budget.

We run two vocabulary layers. Market vocabulary in the title, H1 and URL, where discovery happens. Client vocabulary in the body, where credibility is earned. Then one bridging page that defines the proprietary term against the market term, which quietly becomes the page procurement forwards internally.

The political half is harder than the linguistic half. Do not correct the client. Put the SERP for their term next to the SERP for the market term and let them look. It becomes a decision they made.

Four markets, four cores

Translating a validated core is the standard move and it is wrong.

German compresses into one noun what English splits into three, Verpackungsmaschinenhersteller against Hersteller von Verpackungsmaschinen, which breaks tools and clustering silently. The Anglizismus split is unpredictable: some categories the German market searches in English, some in German, and no rule tells you which. Only checking does. Turkish separates the professional term from the everyday one, and agglutinative suffixes fragment the data further. Russian B2B vocabulary is half-borrowed from English, and Wordstat's headline figure counts every query containing the phrase, so it behaves nothing like a Google estimate.

You rebuild per market from local sources and reuse only the underlying need map. That multiplies the work. It is not a translation line item, and any proposal that prices it as one is underquoting you. We ran exactly this for a B2B service company entering DE, AT, CH and NL in six months with one team.

The gate before you commission forty articles

Before anything goes to production, put the core through the gate. Does every page have one query it exists to answer and one next action? Is any query duplicated across two rows? Can we write it credibly, with proof? Does the SERP contain anyone we resemble? Would a real buyer at that stage accept this page as an answer? Then the money question: if every page hits page one, what is the ceiling on leads, and does that ceiling justify the production budget? In our packaging-line example most candidates die here, folded into existing pages as sections or cut outright for lack of proof, and the survivors are a fraction of the original list.

Publish a pilot batch of five or six and read Search Console impressions before committing the rest. Not rankings, not traffic. Impressions tell you whether the queries exist at all.

This pays slowly. Our Cyprus client Morgan & Morgan reached x15 organic traffic and 11 leads a month with no ad spend, over roughly 31 months. Where volume genuinely exists the shape is entirely different: a Vienna private clinic gained +38 new patients a month through local SEO across 12 districts and 200+ reviews.

And the honest disqualifier. If you have consumer-scale volume, run transactional e-commerce, or work in a category so new that nobody searches for it yet, this method is the wrong one. In the last case SEO is not your first channel, and we would tell you that before quoting. If you want a second opinion on a core you already have, we will look at it in a free audit. The procedure above works without us.

Frequently asked questions

See also