Knowing you show up in ChatGPT is fine. Knowing you show up and your three competitors don’t is what decides whether you win the category. That’s AI share of voice: not how visible you are, but how much of the mention pie is yours. No theatre here — what it measures, the formula, and how to calculate it without the number lying to you.
What AI share of voice is (and how it differs from measuring your visibility)
AI share of voice is the percentage of your category’s mentions that your brand captures in the answers generative engines write — ChatGPT, Perplexity, Gemini, Google AI Overviews, Claude. You define a set of questions that represent your market, count every brand that appears, and look at what fraction of those mentions is yours.
The difference from "measuring your visibility" is one word: relative. Measuring visibility looks at your brand alone — do I appear? in what position? in what tone? Share of voice measures you against the rest of the table. And that changes decisions: you can improve your visibility this month and still lose share because two rivals grew faster. The absolute number congratulates you; the relative one tells you the truth.
The formula, no mystery
There’s no hidden algebra. AI share of voice is calculated like this:
A concrete example: you define 50 discovery questions for your category. Your brand shows up in 15 of those answers and competitors add up to 35 appearances between them. Your share is around 30%. If tomorrow you publish and climb to 20 appearances but the whole set rises to 55, your share barely moves: you grew in absolute terms, not relative ones. That’s why the formula matters more than the feeling.
The one non-negotiable is counting everyone the same way. There are three ways to count and none is wrong: mentions (does the name appear?), citations (does it link your domain as a source?), or position-weighted (does it lead or fill space?). What breaks the measurement is mixing them. Tools on the market don’t agree on what they count, so if you compare two numbers from two different dashboards you’re almost certainly comparing apples to oranges. Pick one criterion and apply it to the whole category.
How to measure it without fooling yourself
Share of voice is easy to calculate and easy to inflate by accident. Three rules keep the number honest.
Lock the prompt set
Define a closed list of real buyer questions and don’t touch it between measurements. The 2026 guides recommend at least 50 prompts per engine, split between informational intent ("what is X") and commercial intent ("best providers of X"). If you change the questions every week, you’re not measuring your share — you’re measuring your question list.
Repeat to tame the variance
The same prompt doesn’t always return the same answer: models have baseline variability. A single pass is a blurry photo. Run each prompt in clean, isolated sessions — no history to contaminate it — and repeat it 3 to 5 times, then average. That’s the difference between a data point and an anecdote.
Count the same thing in every engine
Use the same counting criterion and the same set of competitor brands across ChatGPT, Perplexity, and Gemini. If you count citations in one engine and mentions in another, the aggregate means nothing. Method consistency is what turns three loose numbers into a comparable series.
Why your share changes by engine
Your share of voice isn’t one number, it’s several. Each engine reads a different internet and splits the share differently. ChatGPT leans on Wikipedia-type sources and structured publishers; Perplexity pulls heavily from forums, media, and technical docs. The same brand can dominate one and vanish in the other for the same question.
| Metric | Measuring your visibility | Share of voice |
|---|---|---|
| Question it answers | Am I in? | What part is mine vs rivals? |
| Nature | Absolute | Relative (share %) |
| Needs competitors | No | Yes, fixed set |
| Warns you about | Whether you exist in the answer | Losing ground even while growing |
That’s why measuring in a single aggregate is the classic mistake: it hands you a reassuring average that hides a 5% share in the engine where your buyer actually is. Always break it down by engine before drawing conclusions.
What to do with the number
Share of voice isn’t a trophy for the dashboard: it’s a leading indicator. 2026 studies put around 73% of B2B buyers using AI during their research, so your mention share today is a clue to your market share tomorrow. Three direct uses:
- Prioritize the engine where you’re worst, not the one you already own. Climbing from 5% to 15% where your buyer is beats polishing the engine where you already lead.
- Spot who you’re ceding share to. The same count tells you which competitor rises when you fall. That’s competitive intelligence, not vanity.
- Turn it into a time series. A single figure means nothing; the week-over-week trend tells you whether what you’re doing works.