Why the share matters and the count does not
Mentions went up forty percent. Good month, or did the whole category get written about because of one news cycle?
A raw count cannot tell you, and it flatters you every time the category gets attention. The share can: your mentions divided by the mentions of a fixed set of brands, in the same window, from the same sources. It is one of the earliest signals available, because being talked about tends to precede being searched for, which precedes being bought.
It is also the signal most often reported dishonestly, because every step of the calculation offers a way to make the number nicer.
Matching a brand name is harder than it looks
This is where most share of voice numbers quietly break, and it is worth being specific because the failure is invisible in the output.
Brand names are words. A pattern without word boundaries will find "cal.com" inside "heysocal.com" and count it as a mention. Short brand names collide with ordinary vocabulary. Two different companies share a name across industries, and one of them is much more famous than the other. Every one of those errors adds mentions rather than removing them, so an unguarded count always drifts upward, and it drifts most for the brands with the most generic names.
The fix is unglamorous: word-boundary matching, an exclusion list, and a check that fails the run when a name-searching source has not gone through the shared matcher. We hold that as an enforced rule rather than a convention, because this is the class of error that produces a plausible number instead of an obvious crash.
Rules that keep the number honest
Fix the set. Adding a brand lowers everyone's share without anyone losing anything. Changing the set restarts the series.
Fix the window and the sources. A share computed over one source this month and three the next measures your collection, not the conversation.
Deduplicate. Syndicated coverage republishes one story twenty times. Counting all twenty rewards whoever got picked up by an aggregator.
Do not confuse volume with sentiment. A recall, a lawsuit and a launch all raise share of voice. The direction of the number says nothing about the direction of the news.
Say which sources are sampled. Where a platform is sampled rather than counted, you can report how many but never which, and a share built on a sample is a share of the sample.
What to do when it moves
A rising share with flat branded search is attention that has not converted into intent yet, which is normal for a few weeks and a problem after a quarter.
A falling share while your own count holds means a competitor is doing something new, and the mentions will tell you what. That is the useful direction of this metric: it points at a cause you can go and read, rather than at a conclusion.
And a share that only moves when you publish is not share of voice, it is your own output measured twice. Separate what others say about you from what you say about yourself before reporting either.
Where this sits in the product
Brand mentions are collected weekly per tracked brand, with the shared matcher between the source and the count. Because the set of tracked brands is yours to define, the share is a share of that set, stated as such, rather than of an industry we cannot enumerate.
The newest of the four shares in the table above is the AI one, and it is the only one where being absent produces no signal at all: nobody tells you that a model recommended somebody else.
Pick one competitor. See what we find.
Your first benchmark is free. One domain, measured against you, no card and no call.