For sixteen months, a UK employee wellbeing platform's search traffic had one defining trait: brand dominance. About 96% of attributed clicks came from people searching the company's name. The strategic goal of the whole engagement was to grow the other 4% - visitors who did not already know the company existed.
Then, in the August 2026 re-audit, a Search Console split appeared to show brand share at 47%.
From 96% to 47% in one quarter. That number says the strategy worked spectacularly - non-brand discovery roughly matching brand traffic, the dependence broken, the engagement vindicated. It was hours away from being the headline of a client report.
It was an artifact. The honest figure was 93.6%, barely moved. This case study is about the mechanism that manufactures the flattering number, why it nearly shipped, and the one-sentence rule that stops it.
brand share · the artifact
brand share · the honest figure
§ The half of your data that has no queries
The mechanism lives in a property of Search Console that everyone technically knows and almost every analysis forgets: a large share of clicks carry no query at all. Google anonymizes rare and privacy-sensitive queries, reporting the click but withholding what was searched. On this site, in this 90-day window, that share was enormous: of 1,653 total clicks, 814 - 49% - had no query attached.
Now watch what happens to a brand-share calculation. You filter for clicks on queries containing the brand name and divide by total clicks. The numerator can only count clicks whose queries are visible. The denominator counts everything, including the 814 query-less clicks. Half the denominator is unclassifiable, all of it lands implicitly in "non-brand," and the brand share collapses - not because behavior changed, but because you divided a visible-only numerator by an everything denominator.
Compute the share on the attributed basis instead - brand clicks over clicks that have a query at all - and the answer is 93.6%. Against the prior period's 96% on the same basis: mild improvement, honest trend, story intact but modest.
§ Why the wrong number almost won
The 47% figure had everything going for it except truth. It was dramatic. It validated the work. It was the number the client would have been happiest to see and the consultant proudest to present. Every incentive in the room pointed at shipping it.
That is exactly the property that should trigger the check. The self-serving surprise is the most dangerous number in a report, because nobody in the delivery chain has a motive to challenge it. A figure that flatters the strategy, produced by a method change nobody noticed, will glide through review in a way a damning number never would - the damning number gets triple-checked by instinct.
The tell, in retrospect, was the size of the jump. Real search behavior does not move 49 points in a quarter without a cause you can name. When a metric leaps and you cannot point at the mechanism, the first suspect is not the audience. It is the calculation.
§ The rule: state the basis, every split, every time
The fix costs one sentence per number. Every share, every split, every percentage in a report names its basis: "of attributed clicks" or "of property totals" - never a naked percentage. The two bases differ by whatever your anonymization share is, and at 49% anonymized, any share statistic can swing by tens of points depending on which basis the calculator silently chose.
Two figures from the same window, both technically derived from real data: 47% and 93.6%. The entire difference is the denominator. A reader given either number without the basis has been told nothing, and may have been told worse than nothing.
The comparison discipline follows from the same rule: period-over-period claims are only valid on matching bases. This site's 96% history was attributed-basis; comparing a property-basis 47% against it manufactures a 49-point improvement out of a definition change. Like-for-like or not at all.
§ How to audit your own brand split
- First, measure your anonymization share: total clicks minus the sum of clicks across all visible queries, divided by total clicks. If it is material - and on smaller sites it routinely runs a third to a half - every share calculation you make is exposed to this artifact.
- Compute brand share on the attributed basis only: brand-query clicks over query-attributed clicks. Say so in the label.
- Treat any share metric that jumps dramatically at the same time as a tooling, filter, or method change as guilty until proven innocent. Re-derive it on the old basis before believing it.
- When a number is about to headline a report, ask who benefits from it being true. If the answer is "everyone in this engagement," check it the way you would check a number you hated.
- Keep the anonymized share itself in the report. "49% of clicks carry no query" is a finding about the limits of the data, and clients make better decisions knowing where the floor is.
Numbers dated as of the August 2026 audit, from Google Search Console over a 90-day window, with the prior sixteen-month period as the comparison basis; the site is anonymized as a matter of client confidentiality.
The confession that makes this piece worth writing: the wrong number was not caught by a process. It was caught because a 49-point improvement felt too good, and the feeling prompted one more query. The process now exists - basis stated on every split - precisely so that next time, it does not have to depend on the feeling.