Measurement
AI Share of Voice: How to Measure It (and Why Rankings Don't Work Anymore)
The core metric behind GEO — what it means, how it's calculated, and why a single 'rank' number is the wrong way to think about AI visibility.
5 min read
Why 'rank' is the wrong mental model
Classic search has one page of ranked results, so 'rank #3' is a meaningful, stable idea. AI answers don't have a stable ranked list — a model might name three competitors in one sentence with no explicit order, cite a different set of sources for a slightly reworded version of the same question, or answer differently across ChatGPT, Gemini, Claude, and Perplexity for the identical prompt. Forcing that into a single 'rank' number throws away most of the signal.
What Share of Voice actually measures instead
AI Share of Voice measures, across a defined set of real customer questions and a defined set of engines, what fraction of the answers actually mention your brand at all — and, among the brands mentioned, how prominently. Run the same 10 prompts across 4 engines weekly and count: of the 40 total answers, how many named you, and how many named each competitor? That ratio, trended over time, is a far more honest signal than a single snapshot rank, because it captures both frequency (are you showing up consistently, not just once) and breadth (across which engines, not just your favorite one).
Why the prompt set matters more than the score
A Share of Voice number is only as good as the prompts behind it. The prompts that matter are the ones your actual customers would plausibly type or ask — "best [category] for [use case]", "[competitor] vs [you]", "is [you] good for [audience]" — not generic industry keywords. A high score against irrelevant prompts is worthless; a lower score against the exact 10 questions that precede a purchase decision is the whole game.
Turning the number into action
Share of Voice alone tells you where you stand, not what to do about it. The useful next step is a gap analysis: for the specific prompts where a competitor is cited and you aren't, what is the cited source actually saying, and what would it take to become an equally citable (or better) source for that exact question? That gap-to-action loop — not the raw score — is what actually moves the number over time. Run a free check at /check to see this against your own brand and up to two competitors.