Reading a share-of-voice chart

Share of voice across AI answers looks like a familiar chart, and that familiarity is the problem. It resembles a rankings graph closely enough that people read it with rankings habits.

What the line is made of

Each point is a proportion: of the brand mentions across every prompt you track, how many were yours. It is relative by construction, which has an immediate consequence — your line can fall while your mentions rise, if a competitor's rose faster.

Always read the line alongside the raw count. One tells you how you are doing, the other tells you how you are doing compared to everyone else, and confusing them produces some very confident wrong conclusions.

Small movements are noise

The same prompt asked twice does not reliably produce the same brand list. That variance is inherent, and it aggregates into a line that wobbles even when nothing about your brand or your competitors has changed.

A single week's movement is almost never a finding. A direction sustained across a month, across several prompts, usually is.

The three misreadings

The first is treating a dip as a failure. If a new entrant appears and takes a slice, everyone's share falls, and nobody did anything wrong.

The second is reading the total as fixed. Answers can name one brand or six, so the pool itself grows and shrinks, and the denominator moving is not the same as your position moving.

The third is averaging across prompts that have nothing to do with each other. A brand can be dominant in the questions that matter and invisible in the ones that don't, and a single blended line hides exactly the distinction you needed.

Segment before you conclude

The chart is most useful split by intent. Awareness questions, comparison questions and purchase questions behave differently, and a brand strong in one is frequently weak in another.

Splitting them turns a chart that says "we went down a bit" into one that says where you are losing ground, which is the only version you can act on.