AI Share of Voice (AI SoV): The Only Metric That Matters for AI Visibility

If you can only track one number to measure your brand's visibility inside ChatGPT, Gemini, Claude and Perplexity, it should be AI Share of Voice (AI SoV). It is the cleanest, most defensible, and most actionable metric for the new generative-search era — and it maps directly to commercial outcomes in a way that older metrics like keyword rank or impression share never could.

The definition

AI Share of Voice is the percentage of buyer-intent AI answers in your category that mention your brand vs. your tracked competitors. Formally:

AI SoV (%) = (your brand mentions ÷ total tracked-brand mentions across all prompts) × 100

If you and four competitors are tracked across 100 prompts, and your brand is mentioned 60 times while the four competitors are mentioned 40 times combined, your AI SoV is 60 / (60 + 40) = 60%. Simple, intuitive, easy to explain to a non-technical leadership team.

Why AI SoV beats older metrics

• Traffic metrics miss the answer-only buyer — many AI answers send zero clicks but still shape the decision.
• Keyword rank tracking is meaningless inside a chat interface — there is no ranked list to rank in.
• Impression-based metrics conflate "I was seen" with "I was recommended" — a meaningful distinction for buyers.
• Sentiment scores are noisy and hard to reproduce — mention count is binary and auditable.

How to compute AI SoV correctly

The metric is only as good as the prompt set behind it. A 50-prompt AI SoV is noisy; a 500-prompt AI SoV is statistically meaningful. Aim for at least 150 buyer-intent prompts per category, spread evenly across top-of-funnel, middle-of-funnel, and bottom-of-funnel queries. Run them across at least 4 major models (GPT-5, Gemini 2.5, Claude Sonnet, Perplexity) per cycle to control for model-specific biases.

Use whole-word, case-insensitive matching across the full response. Count every distinct mention, not just whether the brand appears once. Strip out mentions inside refusals ("I can't recommend specific products") and inside the prompt echo. Aggregate across all prompts and all models for a single rolled-up score, then optionally break out per-model SoV as a diagnostic.

Reading the number

• 0–10%: invisible — you are losing every category-level deal you don't already have a relationship in.
• 10–25%: peripheral — you show up on bottom-of-funnel queries but not on "best X" prompts.
• 25–50%: contender — you are in the consideration set; the next push gets you to default.
• 50%+: category leader — you are the brand the model names first; protect this position aggressively.

How to grow AI SoV

The highest-leverage moves, ranked roughly by impact-per-week-of-effort:

• Ship comparison pages (you vs. competitor) for your top 3 competitors — LLMs lift these almost verbatim.
• Complete your Wikidata entity and clean up schema.org Organization markup with full sameAs links.
• Drive 10+ fresh G2 reviews per month — recency outweighs total count.
• Earn 1 high-signal Reddit thread per month in your category (be genuinely helpful, not promotional).
• Land 1 mention per quarter in a tier-1 publication LLMs already cite (TechCrunch, The Verge, Wirecutter).
• Maintain a weekly cadence on LinkedIn — Gemini grounds heavily on LinkedIn company pages.

Reporting AI SoV to leadership

Present AI SoV as a single headline number, trended weekly, with your top 3 competitors as overlay lines. Annotate the chart with what shipped each week (new comparison page, Reddit AMA, G2 push) so you can attribute lift to action. Frame the metric in commercial language: "We grew from 18% to 34% AI Share of Voice on our top 100 buyer prompts, putting us second only to [competitor]. At current category volume, that translates to roughly X additional AI-mediated buyer touches per month."

Common mistakes

• Tracking too few prompts (under 100) — the score is too noisy to act on.
• Tracking only one model — you miss model-specific weaknesses you could fix.
• Treating AI SoV as a vanity metric — it should drive content prioritization, not just dashboards.
• Ignoring per-prompt breakdowns — the average hides which specific queries are killing you.

AI Share of Voice is becoming the GA4 of the generative era — the one metric every marketing leader will eventually ask about. The teams already tracking it weekly will spend the next two years compounding their advantage while everyone else is still trying to figure out how to measure ChatGPT visibility at all.