Skip to content

Glossary

Share of Voice (SoV)

Share of Voice in LLMs is the percentage of answers in your category where a brand is mentioned, measured across ChatGPT, Gemini, Perplexity and more.
  • The share of LLM answers in your topic where a specific brand is named.

  • Competitor-relative: if you and three rivals split mentions 40/30/20/10, your SoV is 40%.

Definition

Share of Voice (SoV) in the LLM context is the percentage of answers in a given category where a brand is mentioned, compared to its competitive set. It is the classic paid-media and SEO metric adapted to generative answers: instead of impressions on a SERP, the denominator is the set of LLM responses to your prompt pack.

How it's computed

For each prompt we collect answers from every model in your coverage. A brand is counted once per answer regardless of how many times it is mentioned. SoV = brand mentions ÷ total answers with at least one brand mention × 100%. Getllmspy computes SoV per model, per topic, per week so you can see where you are gaining or losing ground.

How to read it

SoV is always competitor-relative. A 35% SoV is strong in a market with ten equals and weak in a duopoly. Watch the trend, not the absolute value: a 5-point weekly drop is a signal, even at 60%.

Industry benchmarks

In highly contested spaces (SaaS, fintech, e-commerce), SoV above ~25% is usually strong. In smaller B2B, industrial, or regional categories, 40–50% can be realistic after 3–4 months of focused GEO work.

Important: SoV is computed only within your chosen competitor basket. Adding a new rival mechanically lowers everyone’s SoV — do not confuse that with a true visibility loss.

Share of Voice (SoV) vs LLM-Score

LLM-Score evaluates quality of mention (correctness, sentiment). SoV evaluates quantity of mention vs competitors. Use both: high SoV with low LLM-Score means LLMs talk about you a lot but get it wrong.

When to use

  • Competitive benchmarking each quarter.
  • Measuring the impact of a PR or backlink campaign.
  • Picking which models to prioritize (go where your SoV is worst).