How Lantern scores Brand AI Health

Brand AI Health is a single 0–100 number that captures how a brand is doing on AI-assisted commerce surfaces. It is a composite — not a single measurement — and the composition matters. This piece documents what the number contains, how it is computed, why we chose this particular weighting, and how it fails in revealing ways.

The three dimensions

Website Signals is the dimension your Catalog and Content agents own. Brand Reputation is the dimension the Offsite agent owns. Content Strategy is the dimension the Query agent owns most directly, with Content as a secondary owner.

The formula

formula
Brand AI Health = round(
    0.40 * Visibility_score
  + 0.35 * Favorability_score
  + 0.25 * Citation_Score_driver
)

where each component is scaled to 0–100 before weighting:

  Visibility_score        = aggregate Visibility (%) * 100  // clamped 0..100
  Favorability_score      = 0.7 * sentiment_score + 0.3 * descriptor_cloud_score
  Citation_Score_driver   = count-weighted median Citation Score of Own URLs

Confidence interval:
  CI_95 = base_score ± 1.96 * sqrt(variance_n / n)

Surfaced confidence tiers:
  single-event  n <= 2      // visible, never auto-actioned
  low           n in [3, 7]
  medium        n in [8, 24]  // default surface threshold
  high          n >= 25

Why visibility is measured answer by answer

In the overall score above, visibility carries 40 percent. The visibility measure itself is computed answer by answer: Every answer counts. An answer that leaves the brand out scores zero. Leading the answer is worth 100, and every position further down is worth 80 percent of the one above it, so second place is worth 80, fifth is worth about 40, and a brand buried deep in a long list earns close to nothing. A mention with no clear position earns modest credit. The published visibility score is the average across every answer we track, so a brand reaches 100 only when AI features it first, every time. Warmth is deliberately kept separate; it is its own measure, favorability, so a brand can see at a glance whether its problem is being unseen or being underestimated.

Confidence intervals and tiering