How AI Attention Measures Brand Visibility in AI

When you ask ChatGPT "What's the best CRM for startups?" — does it mention your brand? And if so, is it first on the list or buried at the bottom?

AI Attention answers that question with data. We query multiple AI models with your chosen prompts, analyze the responses, and compute a set of scores that tell you exactly where you stand. This post explains how.

Three Metrics

Metric	What It Measures	Range
AI Attention Score (AAS)	How prominently AI mentions your brand	0–100
Visibility Rate (VR)	How often AI mentions your brand at all	0–100%
Share of Voice (SoV)	Your share of all brand mentions in AI responses	0–100%

AI Attention Score — The Core Formula

AAS captures both whether you're mentioned and where you appear. We assign a position weight to each prompt-model pair using exponential decay:

position_weight = 0.75 ^ (rank - 1)

Rank 1 = full credit (1.0). Rank 2 = 0.75. Rank 5 = 0.32. Not mentioned = 0. Then:

AAS = (sum of all position weights) / (total prompt-model pairs) x 100

Quick Example

You're tracking HubSpot with 2 prompts across 2 models:

Prompt	Model	Result	Weight
"Best CRM for startups?"	Model A	Ranked 1st	1.000
"Best CRM for startups?"	Model B	Ranked 3rd	0.563
"CRM tools comparison"	Model A	Not mentioned	0.000
"CRM tools comparison"	Model B	Mentioned, no ranking	1.000

AAS = (1.0 + 0.563 + 0 + 1.0) / 4 x 100 = 64.1

That single zero on one model-prompt pair costs 25 points. AAS rewards consistency across both models and prompts.

How Rank Is Determined

We do not ask AI models to rank brands — that would be circular. Instead, we parse the response structure:

• Numbered lists → rank is the list number
• Comparison tables → rank is row position
• Explicit shortlists ("the top three are...") → rank by stated order
• Unstructured text → no rank assigned; the brand receives full mention credit

Rank detection is entirely rule-based — no LLM involved. This keeps it deterministic and auditable: the same response always produces the same rank.

Visibility Rate and Share of Voice

VR is straightforward: the percentage of prompt-model pairs where your brand appears at all. It measures breadth.

SoV measures your share of the conversation. If AI models mention your brand 15 times and competitors 95 times, your SoV is 13.6%. Share of Voice is count-based, not position-weighted.

Together with AAS, these three metrics give you a complete picture: how often you appear (VR), how prominently (AAS), and how much of the conversation you own (SoV).

What We're Honest About

• Single-sample queries — individual runs can be noisy; trends matter more than single data points
• Prompt dependency — your score depends on which prompts you choose to monitor
• No sentiment weighting — a positive mention and a negative one score the same (for now)
• Competitor scores are directional — useful for relative comparison, not precise equivalents of AAS

Methodology Integrity

When you change your prompts or models, we insert a segment break on the trend chart — old and new configurations are never shown as a continuous line. Every run produces an immutable score snapshot with a full audit trail. We never recalculate old scores retroactively.

Full Methodology

This post covers the essentials. For the complete technical reference — including the exponential decay rationale, entity detection rules, competitor discovery pipeline, and a full changelog — see our Scoring Methodology Reference.

Questions? Contact us at hello@aiattention.ai