Eleven rows, one buying committee: reading search data in enterprise B2B

Dortmund SEO editorial teamupdated 31 August 2026

Somewhere in your query report there are eleven rows that are not eleven separate events. They are one buying committee at one company, working through an evaluation over three weeks — a technical lead, a procurement officer and someone from compliance, each searching in their own vocabulary. Read as monthly totals, those eleven rows disappear into a number that describes nothing.

This is the central difficulty with search reporting in enterprise software. The report is built to aggregate: it sums impressions across a period and presents an average position. What actually happened was a small number of organisations making a small number of decisions, and the aggregate view is precisely the wrong shape for seeing that.

Four columns, and what they cannot tell you

What Search Console keeps is narrow: the phrasing somebody entered, how many times that put your page in front of them, how often they chose it, and an approximate placing. Four columns, nothing else. In most software companies somebody opens them once a quarter, looks without a specific question in mind, and closes the tab again.

The first row carries the difficulty specific to this market. Four hundred impressions could be four hundred organisations glancing past, or forty organisations looking ten times each while working through a decision. Those are entirely different situations and the column presents them identically. In a business where twenty customers constitute a good year, the second reading is usually the accurate one.

The final row causes most of the misreadings. A mean of nine is compatible with two quite separate realities: consistently ninth, or third for a handful of phrasings and around twentieth for everything else. Each calls for a different response, and the column renders them as the same figure. Anyone who rewrites a page on the strength of that number, without first looking at the spread underneath it, is very likely working on the wrong thing.

Related queries appearing together are one event

The useful first cut through this report is not by page or by device. It is by time and topic together: which queries appeared within the same fortnight, addressing the same subject from different angles. That grouping reconstructs something the aggregate view destroys — the shape of an evaluation.

The practical consequence is uncomfortable. A company can hold an excellent position for its core technical queries and lose evaluations because nothing on the site addresses the licensing question or the data-residency question. Those queries appear in the report with tiny volumes and no clicks — and get ignored, because they look like noise next to the technical rows.

Building this grouping by hand is dull rather than difficult. Sorted by date instead of by total, related phrasings appearing within a fortnight stand out immediately, and a couple of hundred rows can be organised in twenty minutes on the first pass. What matters is that it survives: built carefully once and stored as a view, rather than reconstructed from memory whenever someone raises the subject.

One clarification about the boundary. What groups a cluster is subject and timing, not which page the visitor landed on. A compliance query can perfectly well land on a product page — that combination is in fact the most revealing one, because it shows the question being asked and the answer being absent. Grouping instead by URL prefix measures the shape of your own site rather than the shape of an evaluation.

The three people who search during one decision

The last bullet contradicts every instinct trained on consumer metrics, and it is the single most important adjustment for a company in this position. Sorting the report by volume buries exactly the rows that matter. Sorting by rate, filtered to weak positions, surfaces them — and that view usually contains a list nobody in the company has ever seen.

Position and rate mean nothing in isolation

The fourth row deserves attention in this market specifically. A technical evaluator who reaches a page and finds a "book a demo" form and nothing else will leave — they were not ready to talk to anybody, and asking them to is the fastest way to end the visit. What that visitor wanted was a specification, a limitation, or a straight answer about how something works. The report cannot show this, because it stops at the click.

There is a further consequence for the sales team, and it is the one that tends to justify the whole exercise internally. A cluster appearing this month is an evaluation happening now — months before anybody fills in a contact form. Where the account can be identified at all, that is advance notice of a live opportunity. Where it cannot, the cluster still tells the account team which objections are being examined this quarter, which is more current information than any market survey would provide.

The questions you never appear for at all

Only phrasings your pages have already been shown for can appear here. Where you never surface, there is no entry — and for a company whose site addresses one of three vocabularies, that is exactly where the gap lies. Working only from this report means examining the territory you already occupy and mistaking it for the market.

A watch list should therefore deliberately include the procurement and compliance phrasings you currently have no page for. Those rows stay empty for weeks, and the emptiness is the information: it confirms month after month that the gap is still open. Holding both views in one workspace removes the monthly reconciliation between two tools and makes a comparison over longer periods possible at all.

The division of labour is easy to remember. The report tells you what happened; the watch list tells you whether something you did deliberately had an effect. Asking either source to do the other's job produces conclusions that sound confident and rest on too little. When a company is deciding whether to build a German-language objection section, that matters a great deal — the decision would otherwise rest on precisely the region where no data exists.

Evaluations that never reach your site

Increasingly a search ends before a list appears: the person receives an assembled paragraph, finds what they needed, and stops. In reporting this shows as an impression with no click — outwardly identical to a page that simply lost the comparison. For a growing share of rows, the usual reading is wrong.

In enterprise evaluation this matters more than elsewhere, because the early questions are exactly the kind a paragraph can settle completely: what a technology is, how two approaches differ, whether an integration exists. A vendor can be named regularly in those paragraphs and see no additional visits at all. Establishing whether that is happening requires asking directly — three or four questions an evaluator would genuinely put, once a month, with three notes per run, recorded in a shared overview next to everything else.

One data set, four audiences

Where it goes wrong is usually the same place: one file goes to all four groups. Ten thousand rows in front of a board signals diligence and gets skipped; a condensed summary in the product meeting invites exactly the questions the full table would have settled. Four separate outputs take an hour to configure and never have to be built again.

Frequency deserves the same attention as format. What the product and sales teams read belongs on a monthly rhythm, because it leads to action. What goes to the board belongs on a quarterly one, because it serves a different purpose and more frequent versions mostly create unease without changing anything. A shared rhythm for both produces either unnecessary work or stale figures — in a company with a small marketing function, typically both. Output formats in a shared environment can be configured once per audience and then left alone.

Asking the data a question directly

Most of the hour goes on reaching the right screen rather than thinking about what it says: locating the report, setting the restriction, recalling where last quarter's comparison was saved. Typing the question in ordinary words removes that. Depending on what the question needs, none to three data blocks are pulled, and the twenty most recent exchanges stay available so follow-ups make sense.

• Which clusters appeared last month. Asked once instead of assembled by hand, and repeatable word for word four weeks later.
• Four restrictions cover the realistic cases. Period, query group, page and device handle essentially anything worth asking.
• Twenty exchanges is a sensible ceiling. Landing there repeatedly means the opening question was too wide; two narrow ones get further.
• Sometimes nothing is loaded at all. Some questions resolve from the conversation so far, and not fetching is quicker than fetching what is not needed.

Before any of this, though, one technical question cannot be deferred: does the text actually leave the server? A technical review settles that in a morning. Which procurement and compliance phrasings are realistically reachable is answered by keyword research, and whether the objection pages that should exist actually do belongs to content strategy. Handling all three from one account avoids reaching three conclusions that cannot all be true.

Group your queries by evaluation cluster

Questions from product and sales teams