Share of voice is an old marketing concept with a new meaning in the AI era. Traditionally, it measured how much of your market's conversation your brand dominated — in media, in search, on social. In LLMs, it gains a sharper, more revealing definition: of all the times AI answers a question in your industry, in what fraction of those answers does your brand appear? And when it appears, how much space does it occupy relative to competitors?
The Definition That Matters
Start with the right unit. Share of voice in AI isn't measured over the universe of "all possible answers"; it's measured over a defined set of questions relevant to your business — the ones a customer would ask AI before deciding on a purchase in your market.
Within that set, the metric answers two things. First, presence: in how many of those questions you appear. Second, prominence: when you appear, are you the first brand cited, a mention in the middle, or a name dropped at the end. The two together form a more honest picture than either alone.
Why Counting Mentions Isn't Enough
The temptation is to reduce everything to a count: "I was cited 30 times." But that number, alone, misleads. Thirty mentions at the end of long lists, in peripheral questions, are worth less than five mentions as the first recommendation in high-purchase-intent questions.
That's why well-built share of voice weights. Appearing matters, but appearing prominently, in the questions that matter, matters more. A metric that treats all mentions as equal will give you a sense of presence that doesn't match the real impact.
The Denominator Is What Gives It Meaning
The most neglected part of measurement is the denominator: share of what? Your brand can be cited in half the questions and that be great or mediocre, depending on how cited competitors are. Appearing in 50% when the leader appears in 90% is a weak position. Appearing in 50% when no one tops 30% is dominance.
Share of voice only makes sense in relation. It's a comparative metric by nature. Measuring your presence in isolation, without the context of who else occupies that space, is like knowing your test score without knowing the class average.
How to Build the Measurement in Practice
The logic is direct, even if executing it at scale calls for a tool. Define the set of relevant questions — the real ones, in the language customers use. Run them across the models your audience uses. Record, in each answer, which brands appear, in what position, and in what tone. Repeat over time, because a single snapshot is fragile; what counts is the trend.
From that record come the three useful readings: your presence (in how many questions you appear), your prominence (with what emphasis), and your relative share (your slice compared to competitors'). Together, they tell you not just "do I appear?" but "am I winning or losing my industry's conversation?"
Watch Out for Variation
A technical detail that changes interpretation: language models have randomness. The same question, repeated, can give slightly different answers. That means a measurement based on a few runs is noisy — you may be reading luck, not pattern.
Reliable share of voice requires volume and repetition. Many questions, several runs, over weeks. It's the stable average, not one day's result, that reveals your real position. Measure too little and you risk celebrating or despairing for no reason.
From Number to Decision
A metric is only worth what it helps you decide. Share of voice in AI, measured well, points to where to act: the questions where you vanish, the competitors outpacing you, the topics where your presence is weak despite being where you should be strong. It turns the intuition "I don't think AI talks about us much" into a prioritizable diagnosis.
Calculating this consistently, across multiple models and over time, is exactly the problem Genoma solves: turning the scatter of AI answers into a stable, comparable, actionable share of voice — so you stop guessing your position and start tracking it.