AI SHARE OF VOICE
In AI search there is no page two - there is one answer, and a handful of brands inside it. AI Share of Voice measures your slice of that answer versus competitors. Here is what it captures, why it is position-weighted, and how a real benchmark gets run.
- AI Share of Voice = your position-weighted slice of category mentions across AI answers, versus competitors.
- It is reported per provider and per prompt category - you can lead on Perplexity and be invisible on Gemini.
- Position matters: being named first is not the same as being named last, so mentions are weighted by prominence.
- A real benchmark comes from running the prompts, not from a single spot-check - which is what the Validation Routine does.
What is AI Share of Voice?
AI Share of Voice is the share of category answers that name you, weighted by how prominently you appear, measured against your competitors. Where traditional share of voice counted ad impressions or rankings, AI Share of Voice counts presence inside the synthesized answer - the only real estate left when there is no list of links.
It is a competitive metric by design. Being mentioned is good; being mentioned more than the brands you compete with, in the answers your buyers actually trigger, is what translates into demand.
Why is it position-weighted?
Because not all mentions are equal - the first brand named carries more weight than the fourth. A model that opens with your name is making a recommendation; one that lists you last after three competitors is making a footnote. A raw mention count hides that difference, so the metric weights by prominence and first-mention rank.
This connects to the per-answer scoring - presence, sentiment, position, and consensus - described in how we measure.
Why slice it per provider and per category?
Because your visibility is not one number - it is different on every engine and in every prompt category. A brand can dominate Perplexity, which leans on live retrieval and citations, and be absent from a model that leans on older pre-training. Averaging those together hides exactly the gaps you need to fix.
So Share of Voice is always reported sliced - by ChatGPT, Gemini, Perplexity, Claude, and the rest, and by prompt category - so the leak is locatable, not just visible.
How is a real Share of Voice benchmark run?
You cannot benchmark Share of Voice by asking a model once - you run a representative battery of buyer-realistic prompts and score every answer. That is the job of the 1,000-prompt Validation Routine: generate prompts that mirror how real buyers ask, run them across every provider on a cadence, and record where you appear, how you are framed, and who beats you.
We publish category benchmarks from real Validation Routine data, not estimates. Until a given category's data is collected, we do not invent rankings - the methodology is here so you know exactly what a benchmark will and will not claim. The engine behind it is in how we measure.