Data definitions

Every figure in the product carries a window and a basis. This is what each one counts.

What you'll learn

• The definition of every headline metric
• What a check is, and which checks count in a rate
• Which windows apply where

Check

One prompt asked on one surface at one time, and the result of reading the answer. A check has one of four resolutions:

Citation

A check where your business was named in the answer as a genuine option, by brand, product, service, URL or phone number. Excluded: generic term collisions, mentions that recommend against you, and disputed checks.

Citation rate

Citations divided by checks that count, over the window shown.

One exclusion matters and it is deliberate: a check where Google drew no AI Overview is not in the denominator. There was no answer to be cited in, so counting it would dilute your rate every time Google chose not to render a panel. Those checks are readable on their own as "no Overview", and they still consume coverage, because the fetch really happened.

Excluded surface

A surface where nothing returned in the window could be bound to a source. It has no rate, and it is reported as an exclusion with its staleness rather than as 0 percent. An exclusion is the absence of a measurement. Zero is a measurement.

Movement

The change in a figure between two halves of a window. A surface with no rate cannot have a movement, and a surface whose previous half was excluded has nothing to have moved from. Both return no movement rather than zero.

Share of voice

Your citations as a proportion of all business naming across your tracked prompts, in the window. It moves when your citations move and when the number of businesses being named moves.

Confidence score

How sure the models are that a citation is real and substantive. Averaged across the two models on an agreement; the judge's own on a single-model check. Not a quality score and not a ranking.

Position, sentiment, descriptors

Present only when the answer contained them. Position exists when the answer ranked things. An empty field means the answer had none, not that the value is zero.

Change tiers

Confirmed requires at least eight checks since shipping, across at least two surfaces.

Readiness tiers

Invisible, Emerging, Cited, Preferred. Graded on four levels (Access, Structure, Authority, Citations) and capped by observed engine behaviour: never cited across a real sample caps the grade at Emerging, rarely cited caps it at Cited.

Trace: arrivals and bound visits

An arrival is a visit the tag reported as coming from an AI assistant. A bound visit is one joined to a citation the platform counted. Only the second proves an answer sent somebody. A visit with no AI source recorded is ordinary web traffic and is listed as unattributed.

Coverage

A wallet of answers allocated per billing period, in two pools: monitoring and generative. Metered actions are citation checks, grounded checks, retrieval checks, AI Overviews reads, product checks, content briefs and forecast simulations. Weights are published on the rate card.

Windows

Figures never inherit another figure's window. When comparing two numbers, read both footers first.

Industry Pulse benchmark

A single global benchmark across every workspace the platform measures, identical for every customer. It is not your industry, and the movements listed on it are platform-wide rather than yours.

Try this in Outercite

Ask the assistant any question with a number in the answer, then open the share view from ⌘K and check the definition above against what it says was counted.

Related