AI Share of Voice

AI share of voice (AI SoV) adapts the classic marketing metric to answer engines. Instead of asking “what share of results-page slots do we own,” it asks “what share of answer appearances do we own” — mentions, citations, and recommendations inside AI-generated responses, measured across a defined prompt set against named competitors.

Why the prompt set is the whole game

SoV is a ratio. Change the denominator and the number moves. A prompt set of 10 brand-adjacent questions gives one answer; 50 prompts across the buyer journey gives another, more honest one. The set should include:

• High-intent prompts (“best X for [use case]”, “X vs Y”)
• Problem prompts (“how to fix [problem X solves]”)
• Comparison prompts where competitors appear by name

Run the same set weekly. Same prompts, same engines, dated snapshots. The trend is the deliverable — a single run is a screenshot, not a measurement.

What to count

Keep mentions and citations separate, per the Ahrefs Brand Radar convention: a mention is the answer naming you; a citation is the answer linking a source. You can be mentioned without being cited (brand recall) and cited without a click (attribution). Count both, plus which page got cited — that is your retrieval surface.

Tooling

Our ai_sov tool runs a keyword/prompt set against live SERP data, detects AI-era modules (AI Overview, featured snippet, PAA), and computes brand presence against a competitor list from those modules plus organic — zero LLM API cost, one SERP call per prompt.

Related concepts

• Query Fan-Out — the prompt families your SoV set should cover
• GEO Score — site-side readiness, the complement to answer-side measurement

FAQ

What is AI share of voice?

AI share of voice is the percentage of AI-generated answer slots — mentions, citations, recommendations — where your brand appears, relative to tracked competitors, across a fixed set of prompts. It measures presence inside synthesized answers rather than rankings on a results page.

How do you measure AI share of voice without paid tools?

Define a fixed prompt set, run it against AI search engines on a schedule, and count brand appearances per prompt against competitors. The method matters more than the tool: a stable prompt set, fixed model versions, and dated snapshots make trends meaningful.

Why does AI share of voice fluctuate so much?

AI answers vary by model version, retrieval freshness, location, and phrasing. Single checks are noise. Weekly snapshots of the same prompt set are signal — expect variance between runs and look at direction over weeks, not day-to-day movement.