AI Share of Voice Calculator

The clean formula is simple: your mentions / total vendor mentions * 100. If your brand appears 2 times and competitors appear 18 times, your AI share of voice is 10%. That number is uncomfortable, but it is also clear enough for a team to act on.

The discipline is in the denominator. You should include competitors named in the same answers and same intent class. Do not combine law-firm CRM prompts with ecommerce accounting prompts, and do not compare a branded support prompt with an unbranded buying prompt.

A direct recommendation is the strongest mention. A shortlist mention is still useful because it puts a brand in the buyer's consideration set. A passing mention in a paragraph can count if it clearly positions the brand in the category, but it should be labeled differently from a recommendation.

Do not count every occurrence of a word. If the answer names the same competitor three times while discussing one recommendation, count the vendor once for that answer unless your methodology says otherwise. Consistency is more important than squeezing the metric upward.

A simple average across platforms is the safest starting point. ChatGPT, Gemini, Perplexity, and Google AI features each have different behaviors, citation surfaces, and user expectations. Weighting too early can hide a serious platform-specific weakness.

There are valid exceptions. A local service business may care more about Google AI results because buyers search near me. A technical B2B SaaS company may care more about ChatGPT and Perplexity if prospects use them for shortlist research. Document the weight so the number stays honest.

Do not treat those thresholds as universal law. A category with ten legitimate vendors will have lower shares than a local niche with three strong players. The useful question is whether your share is reasonable for your market position and whether it is improving after fixes.

The more important comparison is often against one or two named competitors. If a rival appears in 14 answers and you appear in 2, the absolute market benchmark matters less than the immediate competitive gap.

A share-of-voice number without answer evidence is hard to trust. The report should preserve enough prompt and answer detail that another person can reproduce the count. That is especially important for agencies using reports as a sales motion.

Another mistake is acting as if share of voice explains everything. A brand can have decent mentions but poor citations, or a low mention count caused by one wrong third-party directory. The SOV number should send you into source analysis, not replace it.

Start by identifying which competitors appear and why. Are they cited through their own comparison pages? Are they present in directories, review sites, industry lists, Reddit threads, or analyst pages where you are missing? The answer determines the fix.

Then build the smallest useful backlog. That might be one brand-versus-competitor page, one best-X-for-Y page, one FAQ update with schema, and three third-party profile corrections. The win is not publishing more content. The win is changing the evidence available to the answer.