What is Share of Voice in AI Search?

At a glance

• Share of voice is a ratio, so it can fall in a month where your own mentions rose and a rival’s rose faster.
• The competitor set you count against is a choice, and it moves the answer more than any content work will.
• Mentions and citations are different units, and mixing them produces a number nobody can reproduce.
• A share figure means nothing without its prompt set, surfaces and date.

Counting mentions across a competitor set

Start with a prompt list and run it. For each answer, list every brand named, yours included, and count.

Your share is your mentions over all brand mentions in the set. Run it per assistant, because a surface naming six brands per answer and one naming two will not average into anything meaningful.

Settle the counting rules before the first run. Does a brand named twice in one answer count twice. Does a parent company count as its subsidiary. Does a retailer stocking your product count as a mention of you.

No standard exists, so the rule you write down becomes your standard. Apply it to competitors exactly as you apply it to yourself, and record it beside every figure you report.

The denominator decides who wins

Pick a narrow set of three direct rivals and your share looks healthy. Widen it to every brand an assistant might name, marketplaces included, and the same data tells a worse story.

Neither version is dishonest. They answer different questions. The narrow set asks how you do against the companies you lose deals to. The broad set asks how much of the category you hold.

Switching between them is the problem. A team reporting the narrow figure in a good quarter and the broad one in a bad quarter has stopped measuring and started arguing.

Choose the set that matches the decision the number is for, write it into the definition, and change it only at a planned reset. A share figure with no fixed denominator is a story, not a metric.

Define a prompt set you can defend to a finance team

1. Pick prompts by commercial value, not winnability. The questions that precede a purchase, including the ones you currently lose.
2. Fix the competitor list in writing. Name every brand you count, and decide up front how retailers and parent companies are treated.
3. Choose mentions or citations, and say which. Counting both is fine. Reporting them as one number is not.
4. Run the same set, same surfaces, same cadence. Monthly is enough for most categories.
5. Keep the method beside the number. Prompt set version, competitor list, surfaces, dates. A figure you cannot reproduce next quarter will not survive its first challenge.

From Parsnipp’s research: share of voice and share of recommendations can rank brands in different orders. In our test Sony took 181 mentions but 101 recommendations, while Anker / Soundcore converted all 308 of its mentions. Count both. Method: 320 simulated shopper conversations about headphones under $150, on GPT-5.6 and Gemini 3.8 Flash, in two independent batches of 160 (Parsnipp, October 2026). Read the research

Two readings of the same set of answers

How this term is used

Share of voice is borrowed from advertising, where it meant your share of ad spend or impressions in a category. In AI search, it’s your share of brand mentions across a set of answers. Some vendors call a version of it share of answer.

There’s no standard yet for measuring AI visibility, and vendors name the same metric differently. Whatever the label, these metrics track the organic branch of AI search marketing.