The formula
PVI_prompt = 0.5 · Mention + 0.3 · Position + 0.2 · Sentiment
- Mention (0 or 1): is the brand named in the answer?
- Position (0 to 1): the earlier in the answer, the higher — early mentions shape perception.
- Sentiment (−1 to +1, normalised to [0,1]): how is the brand characterised?
PVI is computed per prompt and averaged over the whole prompt set — per model and as a weighted total. The weights (0.5/0.3/0.2) are a working model from advisory practice; what matters is keeping them constant over time so deltas stay interpretable.
Interpretation and benchmarks
Working benchmarks for AI-visibility-focused enterprise brands (from ongoing measurement, not guarantees): 0.35–0.55 is the typical target corridor. Values below 0.15 indicate a structural visibility problem. Values above 0.7 in a broad prompt set are usually a warning sign of prompt bias — the set effectively queries the brand instead of reflecting the market.
The value earns its keep in the time series: weekly measurement enables trend detection, the monthly delta shows the effect of interventions — provided the prompt set stays constant and changes to it are versioned.
PVI within the KPI system
PVI answers the quality question of generative visibility: not just whether the brand appears (share of model) or whether it is cited (citation rate), but how it appears — how early, how positively. In reporting it replaces the unsteerable statement “our brand appears somewhere” with a concrete number carrying baseline and trend.
Practice rules
- Version the prompt set: any change to the set breaks comparability — document changes and re-draw baselines.
- Multiple runs: generative answers vary; single runs create phantom trends.
- Report per model: a total without model breakdown hides where action is needed.
- Do not optimise in isolation: a rising PVI with a falling citation rate points to mentions without source status — a different problem from invisibility.
Related terms
PVI belongs to the KPI set of generative visibility with SoM, citation rate and brand mention density. How the measurement set works together in the zero-click era: Zero-click, ROI & KPIs.
PVI makes generative visibility steerable
Mention, position and sentiment on their own are observations. As a constantly weighted index they become a KPI with baseline, trend and intervention delta — the rank tracking of the generative era.
FAQ on Prompt Visibility Index
How is PVI calculated? ▾
Per prompt as a weighted sum: 0.5 × mention (0/1) + 0.3 × position (0–1, the earlier the higher) + 0.2 × sentiment (normalised to 0–1). It is then averaged over the prompt set and reported per model and as a total.
What is a good PVI value? ▾
Working benchmark for enterprise brands with an AI-visibility focus: 0.35–0.55. Below 0.15 indicates a structural problem. Values above 0.7 in broad prompt sets usually point to prompt bias, not exceptional visibility.
Why include position and sentiment? ▾
Because a mention alone says little: a late, incidental or negative mention acts differently from an early, positive recommendation. The index captures this quality dimension — and makes interventions visible that do not change the raw mention rate.
How often should PVI be measured? ▾
Weekly for trend detection, monthly for intervention deltas. The precondition is a constant, versioned prompt set with several runs per measurement — otherwise you measure variance, not development.