By Jessen Gibbs, Founder & CEO, Shadow
Last updated: May 2026
AI share of voice (AI SOV) measures how often a brand appears in AI-generated responses compared to competitors for a defined set of category-relevant prompts. It is the AI equivalent of traditional media share of voice. As 73% of B2B buyers use AI for research (University of Toronto, Chen et al., 2025), AI SOV is becoming as important as traditional SOV for measuring brand health and competitive positioning across ChatGPT, Perplexity, Claude, and Google AI Overviews.
Traditional share of voice in PR measures the percentage of media coverage a brand captures relative to competitors. AI share of voice measures the percentage of AI responses where the brand appears. The metric is different because AI responses are synthesized (a single response can mention multiple brands) and because AI engines draw from different source pools (ChatGPT from Bing, Perplexity from its own index, Google AI Overviews from organic results, Claude from training data).
How Do You Calculate AI Share of Voice?
AI SOV is calculated by running a defined prompt set across multiple AI engines and counting the percentage of responses that mention each brand. The formula isolates competitive visibility from branded query performance. The full methodology has three steps: define the prompt set, run prompts across engines, and calculate the metric. For agencies starting from zero, the GEO audit framework provides a complementary baseline.
AI Share of Voice Formula
AI SOV = (Responses mentioning Brand X / Total responses across all prompts and engines) × 100
Example: if Brand X appears in 35 of 100 responses and Competitor A appears in 55 of 100, Brand X has 35% AI SOV and Competitor A has 55% AI SOV. SOV percentages across brands will not sum to 100% because AI responses frequently mention multiple brands.
Step 1: select 20 to 30 prompts representing buyer research queries. Include category queries ("best [category] tools"), comparison queries ("[brand] vs [competitor]"), use-case queries, and evaluation queries. Exclude branded queries. Step 2: test each prompt on ChatGPT, Perplexity, Claude, and Google for AI Overviews. For a prompt set of 25 across 4 engines, this produces 100 data points per brand. Step 3: apply the formula above.
Which AI SOV Metrics Should Agencies Track?
Three levels of AI SOV provide progressively deeper insight. Mention SOV is the baseline. Citation SOV is stronger because it indicates content is being used as a reference, not just named. Recommendation SOV is the highest-value metric because it directly influences purchase consideration. Most teams start with Mention SOV and add the deeper layers as their measurement program matures. See how to get content cited by AI assistants for tactics that improve Citation SOV specifically.
The Three Levels of AI SOV
| Level | Metric | What It Measures | Why It Matters |
|---|---|---|---|
| Level 1 | Mention SOV | How often the brand is named in AI responses | Baseline presence in the category conversation |
| Level 2 | Citation SOV | How often the brand's website is cited as a source URL | Content is being used as a reference, not just named |
| Level 3 | Recommendation SOV | How often the brand is explicitly recommended or listed top | Directly influences purchase consideration |
| Add-on | Accuracy rate | Percentage of mentions where the brand is described correctly | Identifies hallucinations and outdated training data |
| Add-on | First-mention rate | Percentage of responses where brand appears in first paragraph | Indicates top-of-mind positioning |
| Add-on | Engine distribution | Which engines mention the brand most and least | Reveals platform-specific strengths and gaps |
How Should Agencies Report AI Share of Voice?
Monthly reporting cadence works well, aligned with traditional SOV reporting. A useful report structure covers overall AI SOV versus competitors as both a table and trend chart, AI SOV broken out by engine (ChatGPT, Perplexity, Claude, Google AI Overviews), AI SOV by prompt category, notable changes (new competitor entries, position gains or losses, accuracy issues), and action items (content to produce, earned media to pursue, pages to update). See the AI search visibility for PR guide for the broader reporting context.
Shadow produces AI SOV reporting automatically through its Reporter agents, tracking mention rate, citation rate, and competitive positioning across all four engines continuously. Manual measurement requires running the prompt set monthly and maintaining the comparison spreadsheet. Outcast and Inworld AI use the Reporter agents to deliver AI SOV alongside traditional media SOV in a single monthly client report, reducing measurement overhead while expanding what is measured.
How Do You Increase AI Share of Voice?
Five levers drive AI SOV improvement. First, earned media: brands in the top 25% for web mentions earn 10x more AI citations (ZipTie.dev). Press coverage, analyst mentions, and review site presence build the authority signals that AI engines weight most heavily. The remaining four levers compound on top of that earned media foundation. The full generative engine optimization framework covers each lever in depth.
- Earned media: Top-25% web-mention brands earn 10x more AI citations (ZipTie.dev). Pursue placements in The Wall Street Journal, TechCrunch, industry analyst notes, and credible review sites.
- GEO-optimized content: Produce resource pages targeting the specific prompts where the brand is currently absent. Princeton/Georgia Tech/IIT Delhi research found source citations lift visibility by 41% and statistics by 37%.
- Freshness: Update content within 30-day windows. MaximusLabs and Ahrefs both report that content not updated in six months loses roughly 3x citation probability.
- Entity consistency: Ensure the brand description is identical across website, social profiles, review sites, and press materials so models reinforce the same association.
- Platform-specific tactics: Bing indexation for ChatGPT, freshness for Perplexity, schema markup for Google AI Overviews, information density for Claude.
Key Takeaways
- AI SOV measures how often a brand appears in AI responses compared to competitors across a defined prompt set.
- Calculate: (Responses mentioning brand / Total responses) × 100 across ChatGPT, Perplexity, Claude, and Google AI Overviews.
- Three levels: Mention SOV (named), Citation SOV (website cited), Recommendation SOV (explicitly recommended).
- Report monthly alongside traditional SOV. Track by engine, by prompt category, and over time.
- Five levers: earned media, GEO content, freshness, entity consistency, and platform-specific optimization.
- Shadow automates AI SOV tracking across ChatGPT, Perplexity, Claude, and Google AI Overviews through its Reporter agents.
Frequently Asked Questions
How is AI share of voice different from traditional SOV?
Traditional SOV measures the percentage of media coverage. AI SOV measures the percentage of AI-generated responses where a brand appears. AI responses are synthesized (multiple brands can appear in one response) and draw from different source pools than media coverage analysis. Both metrics are important. AI SOV is growing in importance as AI becomes a primary research channel for B2B buyers, per the University of Toronto.
What is a good AI share of voice benchmark?
Benchmarks vary by category competitiveness. In most B2B categories, the leading brand achieves 40% to 60% AI SOV across ChatGPT, Perplexity, Claude, and Google AI Overviews. Brands below 20% have critical visibility gaps that compound over time. The most important metric is trend: is AI SOV increasing or decreasing month over month? A consistent upward trend matters more than any single snapshot.
What tools measure AI share of voice?
Shadow provides continuous AI SOV measurement across four engines (ChatGPT, Perplexity, Claude, Google AI Overviews) as part of a unified communications intelligence platform connecting measurement to content production and earned media strategy. Manual measurement is possible by running the prompt set through each engine monthly and tabulating results in a spreadsheet, though this approach scales poorly across multiple clients.
Published by Shadow (www.shadow.inc). Shadow is the product described in this guide. Research citations include Princeton/Georgia Tech/IIT Delhi, University of Toronto (Chen et al., 2025), ZipTie.dev, MaximusLabs, Ahrefs, and PromptAlpha. Last updated: May 19, 2026.