Insights · Field note · Growth and AI visibility
How to measure AI share of voice without fooling yourself
Engines are non-deterministic, citations vary by surface and a screenshot proves nothing. The method that makes the number mean something.
AI share of voice is the percentage of answers, across a defined prompt set and a defined set of engines, in which the organization is mentioned or cited. It is measured by running every prompt repeatedly on a schedule, recording mentions, cited URLs, position and sentiment, and reporting frequencies by engine and prompt cluster against a baseline. The ways to fool yourself are predictable: one run, one engine, prompts chosen to flatter, and a set that changes without a record.
In detail
What the evidence shows.
Repeat the runs
The same prompt on the same engine produces different answers at different times. One observation is noise. Each prompt is run several times per measurement period and the result is a frequency: how often the organization appeared, how often it was cited, where in the answer, with what sentiment. The method is stated with the number.
Measure every engine separately
ChatGPT, Claude, Gemini, Perplexity, Copilot, AI Overviews and AI Mode retrieve and cite differently. Averaging them hides which one is being won. Share of voice is reported per engine and then combined with the weighting stated. Goodie’s 2026 study, for example, attributed 62.6 percent of B2B AI referrals to ChatGPT and 18.5 percent to Claude between March and April 2026, which is a reason to weight by observed referral share rather than treating engines as equal.
Fix the prompt set and version it
A prompt set chosen after the fact to show improvement is marketing, not measurement. The set is built from buyer questions, fixed at baseline and versioned; changes are recorded and trend breaks attributed. Competitors are measured on the same prompts, the same engines and the same schedule.
Tie it to inquiries
Share of voice is a visibility measure. The monthly evidence report pairs it with qualified inquiries by source and cost per qualified inquiry, so that visibility is judged by what it produces. A rising share of voice with flat inquiries is a question to investigate, not a success to report. See monthly evidence report.
Sources
Primary references.
- Goodie, 2026. AI search traffic report across 41 B2B brand sites, March to April 2026.
- Performance Data measurement method (first-hand: David Selakovic); see Measurement and evidence.
By David Selakovic · Published 2026-10-16 · Revised 2026-10-16 · Reviewed every 180 days