Share of Voice is a ratio. Your brand's mentions on top, the whole tracked market's mentions underneath. It answers one question: of all the times these engines named somebody in this space, how often was it you?
Now filter the dashboard to your own brand and watch what happens. In more than one tool on this market, the number jumps to 100%.
It is worth being precise about why that is wrong, because the mistake is subtle, it looks like a filter working correctly, and it is easy to ship.
The aggregation order is the whole story
A visibility filter has to decide when it applies. There are only two moments, and they are not interchangeable.
Apply it before aggregating, and the filter becomes part of the WHERE
clause. You have thrown away every row that is not your brand — including the
rows that were going to form the denominator. What is left computes your
mentions over your mentions. That is 1.0, on every day, for every engine, for
every prompt, forever.
Apply it after aggregating, and the denominator is built over the whole surviving population first. Then, and only then, the report picks out the row you asked to see. Your brand's share of a market that still contains everyone else.
Same filter, same data, same UI control. Two entirely different numbers, and only one of them means anything.
How to tell which one you are looking at
You do not need access to anyone's source code. Three checks, in order of how quickly they settle it:
- Filter to one brand and read the number. If it is exactly 100%, or exactly 1.0, stop. No brand has ever had a perfect month across an entire competitive set.
- Filter to one brand and read the number for a competitor instead. If your chief rival also reads 100% when you select them, the filter is running before the sum on both.
- Add a second brand to the filter and check that the two shares still sum to something below 100%. If they now sum to exactly 100%, the denominator is following your selection around rather than describing the market.
The last one is the most useful, because it catches the softer version of the bug: a denominator that shrinks to the filtered set rather than to a single brand. That reads plausibly — 62% and 38% look like real numbers — and it is still measuring the wrong thing.
A metric that cannot go down is not a metric. It is a label that happens to be rendered as a percentage.
Why this is worth caring about
Because the number is load-bearing. Share of Voice is what gets pasted into the board deck, and it is the one figure a marketing team is likely to check by narrowing down to their own brand — which is precisely the interaction that triggers the bug. The tool is at its least trustworthy exactly when someone is looking most closely.
It also fails silently. There is no error, no empty state, no warning. The chart renders, the axis is fine, the export works. The only symptom is a number that is too good, and a number that is too good rarely gets a second look.
Where we put the line
Be Mentioned selects the reported brand after aggregation, always. The brand
identifier is not permitted into the pre-aggregation WHERE clause at all, and
that rule is checked by a regression test on every reporting surface rather than
being left to reviewer memory. Ratios are never stored; only additive components
are, and the division happens once, at query time, over the summed numerator and
the summed denominator.
That distinction between what a filter narrows and what a denominator counts is the same one that governs which answers count at all — including the ones where an engine refused to name anybody.
If you take one thing from this: go and filter your current tool down to your own brand. It takes ten seconds, and it tells you something about the whole pipeline behind the chart.