Share of Voice

Share of voice (SoV) is a long-standing advertising idea ported into AI search. The definition: across a fixed category prompt set, what share of brand mentions in AI answers go to you versus competitors? Where citation rate measures your absolute presence, SoV measures relative occupancy. It is the cleanest way to quantify category positioning under generative search.

The measurement protocol is category-level. Define category prompts ('best GEO platforms in Korea', 'AI search tools for marketing teams'). Define a comparison set of five to ten brands. Run each prompt 50 to 100 times. Count which brands appear in each answer. Compute per-brand share. The unit of analysis is the category, not the single page, which makes SoV better suited to mental-real-estate questions than to URL-level optimisation.

Marketers gravitate to SoV because it speaks the language executives already use — GRPs, SoV, mental availability. It quantifies how much of a category your brand owns inside AI answers. Villion measures SoV against pre-defined category sets, tracks trends over time and surfaces movement against named competitors so campaign impact is easier to defend.

Compared to citation rate the angles are complementary. Citation rate is your absolute presence, SoV is your share. Most teams put both on the same KPI board because divergence is informative — flat citation rate with falling SoV usually means competitors are investing in GEO faster than you are. Across surfaces, the same category prompt often surfaces a different brand pool on ChatGPT, Perplexity and AI Overviews, so split SoV by surface or you will average away the insight.

Two myths to retire. First, treating SoV as an absolute number. Results swing heavily with prompt design, so the right use is a quarterly trend against a fixed prompt set, not a single headline figure. Second, equating high SoV with revenue lift — SoV is an awareness-side metric, and the correlation with sales depends on industry, buying cycle and how heavy the consideration phase is.

Sensible next steps: standardise five to six category prompt sets of 20 to 30 prompts each, agree the competitor pool with stakeholders, re-measure quarterly against the same set, and report SoV, citation rate and definition accuracy on a single board. SoV tells the strongest story executives want to hear — how much mental real estate the brand actually holds inside the AI answers their customers see every day.