Share of Voice, usually shortened to SOV, is the percentage of AI-generated responses that mention or cite your brand compared to competitors across a set of tracked prompts. It is the closest thing AI search has to a scoreboard: not whether you were mentioned once, but how often you show up relative to the field.
A marketing buyer cares about this number because it converts a vague worry, "are we losing ground in AI search," into something trackable over a quarter. Budget conversations need a metric that moves, and SOV moves in response to real work: a rewritten comparison page, a new pricing table, a round of PR that gets picked up and cited. Without it, a team optimising for AI visibility has no way to tell a board whether the effort is paying off or whether a rival is quietly eating the category.
In practice, SOV is computed against a defined prompt set, not against the open internet. A team picks a list of prompts that represent how real buyers ask questions in their category, things like "best project management tool for a 20-person agency" or "alternatives to X for enterprise reporting." Each prompt is run against one or more AI engines, and every response is checked for whether it names the brand and, separately, whether it names each competitor. Share of voice is then the count of prompts where the brand appears, divided by the total prompts tracked, expressed as a percentage. Some tools weight this further by position in the answer or by whether the mention includes a link, but the core arithmetic is a simple count over a chosen denominator.
Here is the calculation with real numbers. Suppose a team tracks 120 prompts relevant to its category across ChatGPT and Gemini. The brand is named in 42 of those 120 responses. That gives a share of voice of 42 divided by 120, or 35%. If the leading competitor is named in 66 of the same 120 prompts, their SOV is 55%, and the gap between the two numbers, 20 points, is the thing a growth team actually needs to close. Track that monthly and the trend line tells you whether last quarter's content push moved the needle or whether the competitor pulled further ahead.
People confuse this with traditional advertising share of voice, and the confusion is understandable because the name and the intent are identical: what fraction of the conversation is yours versus the competition's. But traditional media SOV is measured against a known, fixed universe, total ad spend in a category, total impressions on a channel, something with an agreed denominator that every buyer in the market would compute the same way. AI share of voice has no such fixed universe. The denominator is a prompt set you choose yourself, and two teams tracking the "same" market can build different prompt lists and get meaningfully different numbers. That makes SOV only as good as the prompts behind it. A prompt set skewed toward queries where you already rank well will flatter the number; one skewed toward a competitor's stronghold will punish it. Any report quoting an SOV figure is worth a follow-up question: which prompts, how many, and who chose them. Treat it as a lens tuned to a specific view of the market rather than an absolute, universally agreed truth.