MentionFlow AI
Open app
DocsMetrics

Share of Voice

Your share of the mentions inside your tracked set, a zero-sum split across you and the competitors you track. Brands you don't track are not in the split.

Share of Voice (SOV) answers a competitive question. Count every time anyone in your tracked set was mentioned. How many of those mentions were yours?

Note

SOV is a within-set number. It is not your share of the whole market. Its denominator is you plus the competitors you chose to track. Every other brand an answer names sits outside it. On a typical account, answers name far more brands than the account tracks. So a 27.9% SOV means "27.9% of my tracked set". It never means "27.9% of everything the engines say". See What it does not measure.

How it's calculated

For each entity, we count the runs in which that entity appears at least once. Entities are your brand and every tracked competitor. That count is its run-presence. Your SOV is your count divided by the total across the whole set:

share_of_voice = your run-presence count / Σ run-presence counts (you + all competitors)

It is shown as a percentage, 0 to 100%. The denominator is the whole set, so everyone's SOV adds up to 100%. It is zero-sum: for you to gain, a competitor must lose.

Note

Like every other headline metric, SOV blends only the eight search-grounded engines. The two model-knowledge engines (DeepSeek and Mistral) are left out, so a memory-only mention can't shift the split. The dashboard and the API use the same engine scope, so the two can never disagree. Filter to a single engine and you see that engine's real split, knowledge-only or not. See Data honesty.

What it does not measure

Extraction records every brand an answer names, tracked or not. SOV counts only the tracked ones.

That gap is usually large. On real accounts the answers name more untracked brands than tracked ones. Hundreds or thousands of distinct names sit outside any SOV denominator. So:

  • A rising SOV can mean a shrinking market position. Say a competitor you don't track takes attention from a competitor you do. Your slice of the tracked set grows while your slice of the real field falls.
  • Adding a competitor lowers everyone's SOV, yours included, without anything changing in the answers. That is the metric working correctly. The set got bigger. It is not a drop in performance.
  • The par the Overview card bands against ("an even split across N tracked brands") is a within-set reference too. It compares you to the set you picked, which is a fair question to ask.

The untracked field is not hidden. Discovered ranks every brand your answers name that you don't track, with one-click tracking. Names there carry what evidence we have about what they are. Some are review sites or generic concepts rather than competitors. Rows we can't classify say so rather than guessing.

We deliberately do not widen SOV's denominator to "everything the answers name". That set holds communities, review sites and concepts that are not brands competing with you. Including them would understate every account's share just as surely as excluding them overstates it.

The colour bands, and where they come from

SOV has no absolute good level. 20% is dominant in a field of twenty and last place in a field of two. So its band is cut against its own denominator, not a fixed scale. Par is an even split of the set you track:

par = 100 / (you + your tracked competitors)

Below half of par the reading is red: the competitors you picked are drowning you out. Half again above par it is green: you lead them. In between is amber.

On the Overview card the colour paints the number and the card's top hairline. The band is named in words beneath it: "below par", "around par", "above par". The arithmetic behind it ("an even split across N tracked brands is X%") is one hover away. The denominator itself always rides the card face, never a tooltip. So you always know what the share is a share of.

Two consequences of banding against your own set:

  • Track one more competitor and par falls. The same share can then move a band without anything changing in the answers. That is the set getting bigger, which is the honest reading.
  • Below two entities there is no band at all. With nobody to share the voice with, par is 100% and every real number would band red. A scale that can only say "bad" is not a scale, so none is drawn. The card says "no par" in that slot and gives the reason on hover. The number stays uncoloured.

Why it matters

  • It frames your visibility against the field. Rising visibility with a flat SOV means the whole category is getting more coverage, not that you're winning it.
  • It depends on your tracked competitor set. Add or remove competitors and the denominator changes. Keep the set honest to keep SOV meaningful.

How to read it

  • On the Overview, the SOV delta is a point difference (percentage points), not a percent of a percent. That is clearer for a share metric.
  • When the set has no mentions at all in the window, there is no attention to split. SOV reads 0.0% for everyone.

Related