You measure share of voice across multiple LLMs by running the same prompt set on each platform, counting how often your brand appears relative to all brands mentioned for those prompts, and expressing your presence as a proportion, both per platform and aggregated across them. Because each LLM behaves differently, the per-platform breakdown matters as much as the combined figure, and the trend over time is more reliable than any single measurement.
In short
- Share of voice is your brand's presence as a proportion of all brand mentions on your tracked prompts.
- Measure it per platform first, since LLMs differ, then aggregate for an overall figure.
- Use a consistent prompt set and competitor set so the proportions are comparable over time.
- Read the trend across runs, since AI answers vary between runs.
What is share of voice in AI search?
Share of voice in AI search is the proportion of brand presence you hold relative to competitors across a set of tracked prompts. If you and your rivals are the brands that could appear for a group of questions, your share of voice is how much of that total presence belongs to you.
It is a competitive metric, which distinguishes it from a visibility score. A visibility score tells you how often you appear in absolute terms, while share of voice tells you how much of the conversation you own relative to others, so a brand can have a steady visibility score while its share of voice falls because competitors are appearing more. This is why share of voice is the natural metric for competitive benchmarking, as covered in what is share of voice in AI search. Measuring it well across multiple platforms is the focus here.
How do you calculate share of voice?
You calculate share of voice by counting your brand's appearances across your tracked prompts and dividing by the total appearances of all tracked brands for those same prompts. The result is your share, usually expressed as a percentage of the whole.
The method has a few steps. First, define the brand set: your brand plus the competitors you are measuring against, since share of voice is always relative to a defined group. Second, run your prompt set and record, for each prompt, which of those brands appear. Third, total your appearances and divide by the total appearances across all brands in the set, giving your share. The choice of competitor set matters, because adding or removing competitors changes the denominator and therefore your share, so the set should be consistent and representative. The same logic underlies competitive benchmarking generally, as covered in how to benchmark against competitors in AI search.
How do you handle multiple LLMs?
You handle multiple LLMs by measuring share of voice separately on each platform and then aggregating, because platforms differ enough that a single blended number can hide important variation. The per-platform view is often where the insight is.
The reasoning is that each LLM draws on different sources and behaves differently, so your share of voice can vary substantially from one to another. You might hold a strong share on one platform and a weak one on another, which a combined figure would average away, hiding both the strength to defend and the weakness to fix. So the sound approach is to compute share of voice per platform first, then aggregate across platforms for an overall figure, ideally weighted by how important each platform is to your audience. This per-platform discipline reflects the same point made throughout measurement, that surfaces behave differently and a one-platform or fully blended read can mislead, as covered in how to measure AI visibility. Tracking which platforms favor you and which do not is itself a useful finding.
What pitfalls should you watch for?
The pitfalls in measuring share of voice come mainly from inconsistent definitions and from reading noisy data too literally. Keeping the inputs stable and reading the trend addresses most of them.
A few are worth watching. Changing the competitor set between measurements makes share of voice incomparable over time, since the denominator shifts, so the brand set should stay consistent unless you deliberately revise it. Using an unrepresentative prompt set distorts the picture, since share of voice is only as meaningful as the prompts it is computed over, which ties back to building a sound prompt set, covered in how to build a prompt set worth tracking. Reading a single run as definitive ignores the non-determinism of AI answers, so share of voice should be tracked as a trend across runs rather than judged from one snapshot. And blending platforms without looking at them individually hides variation. Avoiding these keeps share of voice an accurate and comparable measure rather than a moving target.