Brand share in AI answers (also known as Share of Model or AI Share of Voice) is the percentage of relevant buyer prompts in which an answer engine mentions, recommends, or cites your brand relative to all competing brands across a statistically valid sample of model responses. Accurate measurement requires separating raw narrative mentions from clickable source citations and applying position-based weighting, because being recommended first in a three-vendor shortlist drives dramatically higher buyer consideration than appearing as a seventh footnote.
Core Formulas: How to Calculate Brand Share in AI Answers Properly
Prompt Mention Rate (PMR): The percentage of total sampled prompt runs where your brand appears at least once in the generated response.
Competitive Share of Mentions (SoM): Your brand's mention count divided by the sum of all tracked brand mentions (yours plus competitors) across the prompt cohort.
Weighted Recommendation Share (WRS): Assigns higher weight to primary recommendations (e.g., 1.0 for 1st position, 0.6 for 2nd/3rd, 0.25 for passing mentions) and positive sentiment.
Owned vs Earned Citation Share: Tracks how often your own domain URLs are cited versus third-party articles that mention your brand.
Why a Single 'Visibility Score' Can Mislead Executive Teams
As boardrooms ask marketing leaders to report on AI search performance, many teams fall into a common trap: reporting a single, unweighted 'AI mention percentage' based on a handful of ad-hoc prompts. A raw mention rate can look healthy while masking severe commercial vulnerability.
Consider a scenario where your brand appears in 45% of AI responses for your category, but in two-thirds of those responses, the model positions your product as 'a legacy option that lacks modern automation compared to Competitor X', while citing Competitor X's comparison blog as the source URL. A binary mention counter scores that response as a win. A rigorous brand share methodology separates presence, prominence, citation authority, and recommendation polarity.
The 4 Mathematical Formulas That Define Brand Share in AI Answers
To make brand share in AI answers auditable and comparable month over month, standardise the exact denominators your team uses across every model (ChatGPT Search, Perplexity, Google AI Overviews, Claude, and Microsoft Copilot).
Metric Name & Mathematical Formula | What It Measures & Strategic Interpretation |
1. Prompt Mention Rate (PMR) | Measures: Top-of-funnel recall frequency. |
2. Competitive Share of Mentions (SoM) | Measures: Relative competitive density. |
3. Position & Sentiment Weighted Share (WRS) | Measures: True commercial recommendation power. |
4. Grounded Citation Share (GCS) | Measures: Direct source authority and referral click potential. |
Designing an Unbiased Prompt Cohort for Share of Model Tracking
Your brand share metrics are only as reliable as the prompt set you feed into your AI visibility tools. If half of your tracked prompts contain your brand name, your overall share score will be artificially inflated. Exclude pure navigational brand queries from your Share of Model denominator and segment non-branded prompts into three commercial tiers:
Tier 1: Unbranded Category Discovery (40% of cohort): Broad buyer prompts such as 'What are the best enterprise GEO platforms for B2B SaaS?' or 'Top tools to track brand visibility in ChatGPT and Perplexity'.
Tier 2: Problem & Constraint Qualification (35% of cohort): Prompts with specific technical or regional requirements, such as 'Which AI search analytics tool tracks server-side crawler logs and supports UK data residency?'.
Tier 3: Comparative & Evaluation Prompts (25% of cohort): Prompts evaluating category leaders and alternatives, where buyers narrow a shortlist to two or three vendors before booking demos.
Why Multi-Run Sampling Is Essential for Statistical Confidence
Unlike deterministic database queries, large language models sample tokens probabilistically and trigger dynamic web searches that vary slightly from hour to hour. In empirical testing across NeuraCite prompt panels, roughly 35% to 45% of commercial prompts exhibit rank order variance across consecutive runs on the same day.
To prevent your marketing team from overreacting to single-run randomness, execute at least 3 to 5 independent runs per prompt per engine during each measurement cycle. Track both the mean Weighted Recommendation Share and the Inclusion Stability Score (the percentage of runs for a given prompt where your brand consistently appears). Prompts where your brand appears in 1 or 2 out of 5 runs represent your highest-leverage 'threshold opportunities': small improvements in page structure or third-party citations typically lock those prompts into 5-out-of-5 stability.
Diagnostic Matrix: Reading the Gap Between Mention Share and Citation Share
When you plot Prompt Mention Rate against Grounded Citation Share in NeuraCite, every content cluster falls into one of four strategic quadrants, each dictating a specific action plan:
Quadrant Pattern (Mention Share vs Owned Citation Share) | What Is Happening & Next Action |
Quadrant 1: High Mention Share + High Owned Citation Share | Diagnosis: Category Authority. Models both recommend your brand and cite your domain as the primary source of truth. |
Quadrant 2: High Mention Share + Low Owned Citation Share | Diagnosis: Borrowed Narrative. Models know your brand via review sites or competitor comparisons, but rarely cite your own site. |
Quadrant 3: Low Mention Share + High Owned Citation Share | Diagnosis: Uncredited Data Supplier. Models cite your educational guides or statistics, but omit your product from vendor recommendations. |
Quadrant 4: Low Mention Share + Low Owned Citation Share | Diagnosis: Category Invisibility. |
By reporting Brand Share in AI Answers across these four dimensions rather than relying on a single vanity percentage, marketing and SEO teams gain a precise steering mechanism that connects content engineering directly to market share.