Share of Voice
Share of Voice (SoV) measures what share of the total visibility on a topic your brand has – compared to the competition. In the GEO context this means: how often do AI assistants like ChatGPT, Claude or Perplexity name or recommend your brand, measured against all mentions on a given question? A high SoV means that you dominate the conversation.
Why Share of Voice matters
Visibility is not a yes-or-no state but a share of a pie. Even if your brand appears regularly in an AI query, that says little as long as you don't know how often the competition is named. Share of Voice puts your presence in relation to the whole topic field. This makes progress measurable: if your share rises, you push competitors out of the AI assistants' answers. In classic advertising, SoV was long the key figure for advertising pressure. In AI visibility, the principle carries over: whoever is recommended more often gains attention, trust and, in the end, customers – without users having to search long themselves.
How it works
To measure it, you first define a set of typical user questions – such as "Which accounting software for small businesses?". You send these prompts repeatedly to several AI assistants and count how often your brand is named and how often each competitor. Your Share of Voice is your share of all brand mentions combined. Because AI answers fluctuate from run to run, you need many runs for robust figures. It's sensible to distinguish between a mere mention and an active recommendation: being recommended in first place weighs more heavily than a mention in a subordinate clause. Some models therefore additionally weight the position and tone of the mention.
Common mistakes
The biggest mistake is deriving SoV from a single query. AI answers are not deterministic – the same question delivers different brands. Without enough repetitions you measure chance, not a pattern. A second mistake: counting only your own mention and ignoring the competition. Then you have a mention rate but no share. Third, delineating the topic field cleanly is often forgotten: if you take questions that are too broad or unsuitable, this waters down the result. Comparing across different AI assistants is also tricky, because each model uses different sources. Measure per assistant separately and only compare afterwards, instead of throwing everything into one pot.
Relation to AI recommendations
Share of Voice is the bridge between abstract AI visibility and concrete business impact. When a user asks an AI for a recommendation, they usually get only a handful of names – not ten pages of search results. Whoever doesn't appear in this narrow window practically doesn't exist for the purchase decision. A high SoV means that you land in exactly these recommendation lists above average. That's why SoV is the central steering metric for Generative Engine Optimization: you don't optimize for a click but for the AI to name your brand as the answer. If your share rises over time, your GEO measures are working.
Example
Imagine a provider of project management software. The team puts ten typical questions like "Which tool for agile teams?" fifty times each to ChatGPT, Claude and Perplexity. In total there are 900 brand mentions. The provider's own product appears 180 times, the strongest competitor 270 times, the rest spreads across smaller providers. The Share of Voice is thus 20 percent, the market leader's 30. After three months of targeted GEO work – better comparison articles, clearer facts, more citations – the provider's own share rises to 28 percent. The gap to the market leader shrinks measurably.
Common questions
What is the difference between Share of Voice and mention rate?
The mention rate counts how often your brand is named in absolute terms. Share of Voice puts this figure in relation to all brand mentions in the topic field – so you see your share of the overall conversation instead of just your raw value.
How many queries do I need for a robust Share of Voice?
Because AI answers fluctuate, individual queries aren't enough. As a rule of thumb, you should ask each question several dozen times per assistant. Only over many repetitions does chance average out and a stable share becomes visible.