AI Share of Voice

To compute AI Share of Voice for a given business, you need three things: a representative set of customer questions, a defined set of AI assistants to query, and a way to detect whether the business is mentioned in each answer.

1. Generate the prompts. Five to twelve natural-language questions a potential customer would ask about the market — covering commercial intent ("best X"), informational intent ("how does X work"), and comparison intent ("X vs Y").
2. Run each prompt against each provider. Wadsworth queries Perplexity (live web search), OpenAI's ChatGPT (parametric / training-data answers), and Anthropic's Claude (parametric). These are the three highest-traffic AI assistants and represent the experience most customers have today.
3. Count mentions. For each (prompt × provider) cell, detect whether the business name appears in the answer. Sum the appearances, divide by total cells, multiply by 100.

The result is a single 0–100 number representing what fraction of relevant AI answers name the business. A score of 0 means the brand is invisible to AI search. A score of 100 means every AI answer about the market names the brand (almost always unrealistic — competitors will also be named).

We'll say this more directly than most vendors: AI Share of Voice is a directional metric, not a guarantee.The major AI assistants (OpenAI, Anthropic, Perplexity) don't publish how they pick what to recommend. Their models also update, which can shift what they say about the same brand without anything you did changing.

That doesn't make the metric useless — it just makes it monitoring, not control. The same critique applies to Google's ranking algorithm (opaque for 25 years) and the SEO industry built around it. AI Share of Voice gives you a directional signal: if it's going up, you're doing the right things; if it's flat, change something. Don't treat any single score as definitive — track the trend.

Direct measurement still beats inference. Traditional SEO guesses at Google's intent from rankings. AI Share of Voice asks the AI directlywhat it knows and what it recommends. That's a stronger position than the alternative, which is guessing.

A boutique coffee roaster in Portland runs an AI Share of Voice check with twelve customer prompts (e.g. “best specialty coffee roasters in Portland,” “where to buy single-origin beans online,” “Portland coffee subscription vs grocery store”) across three AI assistants (Perplexity, ChatGPT, Claude). That's 36 total prompt-by-provider cells.

The check returns:

• Perplexity mentioned the roaster in 8 of 12 answers
• ChatGPT mentioned the roaster in 4 of 12 answers
• Claude mentioned the roaster in 3 of 12 answers

Total mentions: 15 of 36 cells. AI Share of Voice = 42%.

The per-provider breakdown is more useful than the headline. The roaster is well-known on Perplexity (which searches live web — recent press releases and Reddit threads helped). But ChatGPT and Claude (which answer from training data) barely know the brand. The strategic implication: invest in the kind of content AI assistants train on — guest posts on established food blogs, citations on industry directories, structured data on the brand's own site — to lift the parametric scores over the next training cycle.

The term Share of Voiceoriginated in advertising and media planning in the 1980s as a way to measure a brand's presence in paid media relative to competitors — typically calculated as (your ad impressions ÷ total category ad impressions) × 100. It was used to benchmark advertising budgets and predict market-share gains.

AI Share of Voice extends this concept to AI-assistant answers as the new discovery surface. Instead of counting paid impressions, it counts unpaid AI recommendations — closer in spirit to organic search visibility than to paid media. Wadsworth proposed the term in 2026 as the industry began needing a standardized metric for AI-assistant brand presence.