Share of Voice, Mentions, Citations and Endorsement

AI share of voice

Share of voice answers: when buyers ask, how often are you the answer? It is measured on unbranded buyer prompts, the discovery questions where the engine is free to name anyone, and expressed as your slice of all brand recommendations in your category on that audit.

Because it is benchmarked against everyone the engines actually named, share of voice is a category leaderboard, not a vanity score. A rising number means engines pick you more often when nobody asked about you by name.

Mentions vs citations

These two are often conflated and they measure different things:

• Mention: Your brand appears in the answer text. The engine knows you and considered you worth naming.
• Citation: The engine cites a URL as a source for its answer. Citations show which pages, yours and others', the answers are actually built from.

The gap between them is diagnostic. Mentioned but never cited usually means the engines know of you but lean on third-party sources when they need proof. Cited but rarely mentioned in recommendations usually means your content answers side questions without positioning you as a pick. The citation view also reveals which venues, review sites, communities and publications your category's answers depend on: that list is where off-site work pays.

Endorsement: the metric that decides deals

Being named is not the same as being recommended. Endorsement measures the strength of your appearances: where you rank inside answers that name several brands, and how the engine frames you, as a strong recommendation, a neutral listing, or a name with caveats attached.

Two brands with the same share of voice can have very different endorsement, and the one framed as the safe pick wins the deal. Endorsement gets its own deep dive, including the ladder from absent to endorsed and what moves a brand up it: see AI Endorsement: Why Mentions Are Not Recommendations.

Why numbers move between audits

AI answers are not static rankings. Between two audits, several things change under you: models retrain, live retrieval picks different pages, competitors publish, and engines re-weigh their sources. Some week-to-week wobble on individual prompts is expected behavior, not measurement error.

Read accordingly:

• Trend over snapshot: One audit is a data point. Three audits are a direction. Judge fixes on the trend after they ship, not the single run after.
• Category-relative: If your share of voice fell but every rival fell too, the answers changed shape. If you fell while a rival rose, that is a competitive event worth opening.
• Per engine: Engines differ in how they source answers, and one engine moving while others hold usually points at the source that engine leans on. Cited reports each engine separately for this reason.