Share of Voice in AI measures how often your brand appears in AI answers compared to competitors. It is one of the most useful GEO metrics, but interpretation depends entirely on which prompts you use.
Branded vs non-branded prompts
The crucial distinction in SoV measurement: do you ask branded questions (with your brand name in them) or non-branded questions (general consumer questions)?
Branded prompts
Examples: “What does KBC offer as a mortgage?”, “What exactly does Fyrco do?”, “Is Coolblue a reliable webshop?”. With this type of prompt you are often present — the question is explicitly about you. The score here is mostly informative: is the information correct, how are you presented?
Niet-branded prompts
Examples: “Which Belgian bank do you recommend for a starter?”, “Which company does annual fire extinguisher inspections in Flanders?”, “Best project management tool for small teams”. Here your brand is not explicitly mentioned. This is the prompt category where real GEO performance becomes visible.
The gap between branded and non-branded SoV is often large. Brands that score dominantly on branded prompts can almost completely disappear on non-branded questions. That difference is where GEO work has real impact.
Comparative prompts — a third category
Between branded and non-branded sits a third type that is often strategically the most important: comparative prompts. “KBC vs Belfius for first-time buyers”, “Argenta or ING for self-employed”, “Fyrco or Ansul for fire safety”. The user explicitly mentions two or more brands — the mention rate is by definition 100% for each named brand.
This is not a presence measurement. It is a sentiment and positioning measurement: how does AI describe your brand relative to the competitor mentioned in the same breath? Which brand receives the recommendation, which receives caveats?
Comparative prompts are decision moments in the funnel. Whoever wins here often wins the customer. Whoever loses here loses not just visibility but the framing of the entire comparison moment. Treat these prompts as a separate measurement cluster — not together with non-branded, not together with branded.
How do you calculate SoV concretely?
Define a benchmark prompt set per product category or topic. Minimum 20, preferably 50 to 100 prompts per topic for reliable measurements. Run the prompts multiple times across different models, count the number of answers in which your brand appears, and divide by the total.
Note one important filter: not every mention is a valid presence signal. “X is a good choice, alongside other players such as your brand and Y” counts your brand in a raw mention rate, but it is not real presence — you appear as a side-item in a list focused on another brand, without the AI actively bringing you forward. Filter out this type of incidental mention for a clean SoV measurement. The difference between raw and filtered citation rate can be ten to twenty percentage points.
Do the same for competitors. The ratio gives you Share of Voice.
SoV over time
One measurement is a snapshot. Repeat measurement monthly or quarterly with the same prompt set. Patterns of rise or fall say more than absolute numbers. A declining SoV over three months is an actionable signal.
Segmentation of SoV
Raw SoV figures often hide important details. Segment by: intent type (informational vs comparative), product category, audience context, AI platform. The insights are often in the segmentation, not in the total.
Related in the hub
→ How do you build a reliable prompt set? Read 4.3 — Prompt taxonomy, Fragility and Volatility.