Presence

Presence is how often assistants put you in the answer. The console splits it into the handful of numbers that behave differently, because the one chart everyone wants is the one that hides the problem.

The number at the top

The Visibility Score is the share of grounded probes in which a model recommended you. One rate, not two: what this product counts as a mention is already a recommendation rather than a passing reference, because being named in a list of eleven options is not what sends you a buyer.

Grounded, training, or both

Every figure reads in three modes. Grounded means search was available, so the answer reflects what a model can find today. Training means it had no search, so the answer reflects what it already believes about you. All mixes them, which is fine for counting and misleading as a rate. The Visibility Score is the grounded one.

Read the band, not the point

Every rate here is an estimate from a set of probes, so it carries a confidence band. A move inside the band is noise. A move outside it is news. The overview shows the band rather than hiding it behind a single tidy percentage.

Where the numbers come from

Every figure traces back to a stored run: the prompt, the model, the full answer text, the searches it ran, the sources it pulled and which of those it quoted. Nothing in the charts is a model of a model. Open any point on any chart and you reach the answer that produced it.

Concepts rather than clicks: how to measure share of voice in AI answers.