How to Measure Share of Voice in AI Search

Summary

Share of voice in AI search is the percentage of brand mentions that belong to you when an AI engine answers the questions your buyers ask. To measure it, define a fixed prompt set, run it across each engine, count every brand named in the answers, then divide your mentions by the total. This guide gives you the formula, a repeatable step-by-step method, weighting for commercial intent, and the mistakes that quietly inflate or deflate the number.

What Is Share of Voice in AI Search?

Share of voice (SOV) in AI search is the proportion of brand mentions you own when an AI engine, ChatGPT, Gemini, Perplexity, Claude or Copilot, answers a question relevant to your category. If a buyer asks for the best project management tool and the answer names five products, and you are one of them, your raw share of that single answer is one in five. Aggregate that across hundreds of prompts and several engines and you get one comparative number: how loud your brand is in the AI conversation versus everyone else competing for the same recommendation.

Where a citation rate only asks “did I appear?”, share of voice asks “how much of the available attention did I capture, and how much went to rivals?” It is inherently competitive, which is exactly why brand managers and CMOs gravitate to it. It turns a fuzzy worry: are we showing up in ChatGPT?: into a scoreboard you can defend in a board meeting.

Why Share of Voice Beats a Single Citation Rate

A citation rate of 40% can read as a win or a disaster depending on context. If your two main competitors each sit at 15%, you dominate the category. If the category leader sits at 85%, you are an afterthought in the very answers that close deals. Share of voice supplies that context by normalizing your presence against the whole field rather than against zero.

It also surfaces the dynamic that matters most in AI answers: substitution. When an engine recommends a finite list, usually three to six brands, every competitor mention is attention that did not go to you. Traditional search rankings are positional but not zero-sum in the same way; a generated answer often is. Tracking share of voice keeps you honest about that trade-off instead of celebrating a citation rate in a vacuum.

The Share of Voice Formula

At its simplest, share of voice is a ratio you can calculate three ways depending on how much precision you need:

• Raw share of voice = (your brand mentions) ÷ (total brand mentions across all tracked prompts) × 100.
• Per-engine share of voice = the same calculation run separately for ChatGPT, Gemini, Perplexity, Claude and Copilot, because your standing is rarely the same on each.
• Weighted share of voice = each prompt’s contribution multiplied by its commercial importance before you sum, so high-intent questions count for more than idle curiosity.

The denominator is the part most teams get wrong. “Total brand mentions” means every brand named in the answers, not just you and your favourite rival. If you only count yourself and one competitor, you will overstate your position badly: the long tail of alternatives the engine also recommends is real attention you are losing.

How to Measure It Step by Step

You can produce a defensible share-of-voice number manually before you automate it. The method is what matters; the tooling just makes it repeatable.

• Build a fixed prompt set. Collect 30 to 100 real buyer questions across the journey: problem-aware (“how do I solve X”), comparison (“best tool for X”, “X alternatives”) and brand-specific. Freeze the list so results stay comparable.
• Choose your engines. Decide which AI engines your audience actually uses and measure each one separately; never average them into a single blurred figure.
• Run every prompt on each engine. Use a clean session with no personalization so results reflect the model, not your own history.
• Extract every brand named. For each answer, list all brands mentioned or cited: yours and all competitors, including ones you did not expect.
• Tally and divide. Count your mentions and the total mentions, then apply the formula per engine and overall.
• Repeat on a schedule. Re-run weekly or monthly. A single snapshot is noise; AI answers shift with model updates and fresh content, so the trend is the signal.

Weighted Share of Voice: Not All Prompts Are Equal

Owning the answer to a purchase-ready question like “best CRM for a 50-person sales team” is worth far more than appearing in a vague “what is a CRM” explainer. If you treat every prompt as equal, a flood of low-intent visibility can mask the fact that you are absent from the questions that convert. Assign each prompt a weight based on commercial intent, a simple 1 to 3 scale works, and multiply each prompt’s contribution by its weight before summing. The weighted figure usually tells a sharper, less flattering, and far more useful story than the raw one.

Mistakes That Distort Your Numbers

• Tiny prompt sets. Ten prompts give you a number with no statistical meaning. Aim for at least 30, ideally more, spread across intents.
• Counting only known competitors. If your denominator excludes the brands you have not heard of, your share of voice will look artificially high.
• Measuring once. AI answers are non-deterministic and change over time; a one-off reading cannot show direction.
• Blurring engines together. Strong in Perplexity and invisible in ChatGPT is a completely different problem from being mediocre everywhere: the average hides both.
• Treating a mention as a recommendation. Being listed with a caveat is not the same as being named the top choice; pair share of voice with sentiment to read it correctly.
• Ignoring brand-name ambiguity. If your brand name is a common word, filter for genuine references so you do not over-count.

From Measurement to Action

A share-of-voice number is only useful if it changes what you do next. Read it by cluster: a low share on comparison prompts points to missing or weak comparison content and third-party reviews, while a low share on problem-aware prompts means the foundational, citable explainers are not there yet. Watch where competitors out-cite you and reverse-engineer why, usually it is clearer structure, stronger sourcing, or more authoritative mentions on the sites the engines trust. Then re-measure. Share of voice in AI search is not a vanity metric you check once; it is the scoreboard you use to decide where the next piece of content, the next review push, or the next PR effort should go.