How to measure share of voice in AI

Begin by defining a representative prompt set. Include discovery questions, comparison prompts, category recommendations, and objection-handling queries your buyers actually ask. Ten to fifty prompts is enough to start if they map to real purchase journeys.

Run those prompts on a fixed schedule across the engines you care about. Record whether your brand is mentioned, cited with a link, misattributed, or ignored. Note which competitors appear and whether the answer is accurate.

Convert raw runs into a simple score: mentions divided by opportunities, optionally weighted by engine importance or commercial intent. Track citation rate separately from vague brand mentions so you can see whether models treat you as a source or just a name.

Use the trend, not a single snapshot. Model updates, content refreshes, and competitor campaigns all move the numbers. A weekly or monthly cadence reveals whether AEO fixes are working before you over-invest in tactics that do not move citations.

Pair visibility data with on-site signals: schema coverage, answer-first openings, author credibility, and crawl health. When share of voice drops, audits show whether the cause is technical discovery, thin content, or competitor narrative strength.

The goal is not one perfect score. It is a consistent operating rhythm: measure, diagnose, fix, and re-measure so leadership can see progress in generative discovery the same way they once watched organic rankings.