Core AI Visibility Metrics

Introduction to AI Visibility Metrics

In traditional SEO, we rely on rankings, impressions, and clicks. However, AI surfaces—such as Google Search Generative Experience (SGE), Perplexity, and OpenAI Search—operate on a different logic. Success is no longer about being 'Result Number One' on a list; it is about being the primary reference in a generative response. To measure this effectively, practitioners must pivot from traditional SERP tracking to AI-specific metrics that account for synthesis, attribution, and user trust. This lesson identifies the core metrics that define success in the age of generative engines.

1. Share of Model Voice (SoMV)

Share of Model Voice is the AI-era equivalent of Share of Search. It measures how frequently your brand or product is included in generative responses for a specific set of target prompts compared to your competitors.

How to Calculate SoMV

To calculate SoMV, you need a defined set of 'Core Prompts' (e.g., "What is the best CRM for small legal firms?").

1. Select 50-100 high-intent prompts.
2. Run these prompts through target models (GPT-4o, Claude 3.5, Gemini).
3. Count how many times your brand is mentioned as a recommendation.
4. Divide by the total number of brand mentions across the set.

Example: In a test of 100 prompts for 'sustainable footwear,' Brand A was mentioned in 30 responses, and the total mentions for all brands was 150. Brand A has a 20% SoMV.

2. Citation Depth and Accuracy

Being mentioned is the first step; being cited correctly is the second. Citation Depth tracks how many times an AI engine links back to your specific domain as the source of truth for a claim.

• Primary Citation: Your link is the first one shown in the source list.
• Inline Citation: Your link is embedded directly within the text of the generated answer.
• Citation Accuracy: Does the AI attribute the correct data to your brand? If an AI claims your software costs £50/month when it actually costs £80/month, the visibility is high but the accuracy is low, leading to poor user experience and potential conversion issues.

3. Brand Sentiment and Attribution Tone

Generative AI does not just provide links; it provides context. If an AI mentions your service but adds a caveat like "though users often report high latency," your visibility is working against you.

The Sentiment Matrix

Practitioners should categorise mentions into three buckets:

• Foundational/Factual: "Brand X is a provider of [Service]."
• Recommendatory: "Brand X is highly recommended for users seeking [Feature]."
• Critical: "While Brand X offers [Feature], it lacks [Competitor Feature]."

Monitoring the shift from 'Factual' to 'Recommendatory' is a primary KPI for AI Visibility Practitioners.

4. Reference Share (The 'Link-to-Text' Ratio)

Reference Share measures the density of your brand's presence in the 'Sources' or 'References' panel vs. the actual text output. High Reference Share with low presence in the main text summary suggests you are being used as a 'footnoted source' rather than a 'primary authority.'

Conversely, being in the text but missing from the citations often indicates the AI is using your data as 'general knowledge' without giving credit. This helps you identify where your schema markup or 'About' pages might be failing to trigger proper attribution.

5. Information Gain Score

AI models prioritize 'Information Gain'—the introduction of new, unique facts that aren't present in other top-ranking documents. Measuring your 'Information Gain Score' involves comparing your content against the average LLM training data or top-cited competitors.

• Metric: Number of unique data points or unique perspectives cited by the AI from your content that do not appear in competitor citations.

6. Prompt-to-Purchase Path (P3)

This is a qualitative-quantitative hybrid metric. It tracks the number of follow-up prompts a user must enter before the AI offers a direct link to your conversion page.

• Low P3 (1-2 prompts): The AI identifies you as the definitive solution immediately.
• High P3 (4+ prompts): The user has to narrow the field significantly before your brand appears, indicating a lack of topical authority.

Worked Example: Calculating AI Visibility for a Fintech App

Let’s look at 'FinSafe,' a fictional personal finance app. We want to measure its visibility for the topic 'best high-interest savings accounts UK.'

Step 1: Define the Prompts We use 20 prompts, ranging from broad ("What are the best UK savings accounts?") to specific ("Which UK app has the best automated saving features?").

Step 2: The Audit (Google Gemini)

• FinSafe appeared in 12 out of 20 responses.
• In 8 of those, it was the first-listed recommendation.
• In 4 of those, it was cited but not recommended (listed under 'other options').

Step 3: Calculating SoMV FinSafe SoMV = (12 mentions / 20 prompts) = 60% presence. Recommendation Rate = (8 recommended / 12 total mentions) = 66%.

Step 4: Citation Check In 10 responses, the AI cited the official FinSafe blog. In 2 responses, it cited a third-party review site. This suggests FinSafe’s 1st-party content is strong but could be improved to capture more direct citations.

Putting it into Practice

To move these metrics from spreadsheets to strategy, follow this weekly checklist:

1. Prompt Testing: Use a tool or manual browser (incognito/signed out) to run your top 10 commercial prompts through Perplexity and Gemini.
2. Accuracy Audit: Verify that the prices, features, and claims attributed to you by the AI are factually correct. If not, update your site's Schema.org markup immediately.
3. Competitor Delta: Identify which competitor is being cited when you are not. Visit their page and look for 'Information Gain'—what facts are they providing that you aren't?
4. Sentiment Mapping: Look for 'but' or 'however' in the AI's description of your brand. These phrases indicate the AI has ingested negative reviews or critical data points that need addressing in your content strategy.