Core concepts

A dozen words carry the entire product. Learn them once and every screen reads the same way.

Workspace

A team account. It owns the brand you track, the members who can see it, and the subscription that pays for it. Every URL in the app is scoped to one: /home/{workspace}/....

Brand (project)

The website being tracked, plus its market. The brand supplies the domain used to detect you inside answers, and its location determines how queries are localized.

Prompt

A question you believe your buyers ask an assistant. Prompts are the unit of measurement and the unit of cost: each one runs against three engines every day.

Every prompt has an intent — visibility or perception — and a subject — your brand or one of your products. Crossing them gives the four groups the product is organized around. Visibility prompts may not contain your brand name; see Prompts.

Entity

Anything tracked inside answers: your brand, a competitor, or a product. Each has a name and, for brands, a domain. Both are matched against answer text during extraction, which is why an accurate name matters as much as an accurate domain.

Engine

The AI surface being measured — ChatGPT, Google AI Overviews, Google AI Mode, Perplexity. Engines are captured from the real consumer interface, so what is measured is what a user sees, ads and shopping units included.

A second class of engine — provider APIs called with search enabled — runs as a small calibration panel and is never mixed into the same metric. See AI engines.

Answer

One capture of one prompt on one engine, on one day: the text as rendered, its citations, whether it was grounded, the fan-out queries the engine ran, and the surface units it displayed. Every number in the product drills back to answers you can read.

Mention

An entity detected inside an answer, with its position (order of first appearance among tracked entities), how often it appears, and a snippet for context. Mentions are extracted from answer text, never self-reported by a model. See Mentions and sentiment.

Citation, source and domain

A citation is a source the engine displayed. It is either quoted — cited inline against a claim — or listed in a sources panel, and those are very different achievements.

Citations resolve to sources (specific URLs) which roll up to domains. This is how you learn which third-party pages decide what the assistants say about you.

Placement

A shopping or ad slot rendered alongside an answer. Measured as share of slots, because a carousel of ten is ten chances to be seen. See Ads & Shopping.

Run and confidence interval

Each prompt is asked more than once per engine per day. The spread across those runs produces a 95% confidence interval on every rate, and a noise floor below which a change is not marked significant.

This is the habit that separates a measurement from a number: check the floor before believing a delta.

How they fit together

Workspace → Brand → Prompts ──┐
                              ├─→ Answers (per engine, per day) ─┬→ Mentions → metrics
   Entities (brands,     ─────┘                                  ├→ Citations → Sources → Domains
   competitors, products)                                        └→ Placements

Read it as one sentence: your prompts and entities produce answers, and answers produce the mentions, citations and placements every dashboard aggregates.

Two metrics you will use daily

• Visibility — the share of tracked answers that mention you. Are you in the room?
• Share of Voice — of the answers that mention any tracked entity, the share that mention yours. When you are in the room, how much of it do you own?

Both are defined precisely in Metrics.