NewSend finished articles to WordPress, Shopify or any site, as drafts you approve.Publish articles to WordPress and Shopify.
Blog

AI search share of voice: definition, formula and pitfalls

By Citeroot Research

Published 2 min read

Guides

How to calculate share of voice for AI answers, why the denominator matters, and how many answers you need before a difference means anything.

On this page
  1. Formula
  2. Define everything around it
  3. How many answers do you need?
  4. Pitfalls
  5. Reporting template
  6. Sources

Share of voice (SOV) in AI answers is a brand's mentions as a share of all tracked brands' mentions across a defined set of answers.

It's a relative measure. If you and two competitors are named 24, 31 and 22 times in 60 valid answers, your SOV is 24 ÷ (24 + 31 + 22) ≈ 31%.

#Formula

SOV(brand) = mentions(brand) ÷ Σ mentions(all tracked brands)
Mention rate(brand) = mentions(brand) ÷ valid answers

They answer different questions. SOV tells you how you split the conversation with named rivals. Mention rate tells you how often you appear at all. A category where nobody is mentioned much can give you a big SOV and a small mention rate.

#Define everything around it

An SOV figure without its frame is not meaningful. Always state:

  • The prompt set — which questions, how many, which stage.
  • The engines — and the model versions where known.
  • Location and language.
  • The time window and number of runs per prompt.
  • The brand list — SOV changes when you add a competitor.
  • What counted as valid — failed collections are not absences.

#How many answers do you need?

Small samples produce wide uncertainty. If you checked 20 answers and a brand appeared in 5, the plausible range for its true mention rate is roughly 11%–47% (a 95% Wilson interval). A brand at 7 of 20 overlaps almost entirely with it.

Practical rules:

  • Don't call a difference real unless the intervals clearly separate.
  • Prefer more repeated runs of a stable panel over more prompts you won't act on.
  • Use our share-of-voice calculator to see the intervals for your numbers.

#Pitfalls

Counting mentions per answer instead of answers. If a single answer names a brand three times, it is still one answer mentioning that brand. Decide a unit and keep it.

Blending stages. A strong showing in problem-stage questions can hide a weak one in comparisons. Segment before you average.

Position weighting without disclosure. Some tools weight earlier mentions more. That's legitimate, but publish the weighting.

Mixing branded and non-branded prompts. A prompt that names you guarantees you're mentioned. Keep branded prompts out of SOV.

Forgetting citations. A brand may dominate mentions while a competitor dominates the sources. Report both. See AI citation tracking.

#Reporting template

ItemValue
PeriodSep 1 – Sep 28
Prompt set48 non-branded prompts, 4 stages
EnginesChatGPT, Gemini, Claude
Valid / failed runs432 / 11
BrandMentions
Brand A118

Fill in your own numbers; keep the structure.

#Sources

  • Wilson score interval: Wilson, E. B. (1927), "Probable inference, the law of succession, and statistical inference", Journal of the American Statistical Association.
  • Aggarwal et al., GEO: Generative Engine Optimization.

See what AI says about your business today.

Add your website and a few questions your customers ask, then read your first answers. Free plan, or a 3-day free trial of Grow · card to start.