Measuring what AI engines recommend is a new discipline with borrowed vocabulary. These are the terms our audits are written in, and each links to where the method is set out.
The measurement vocabulary, defined by what each observation actually establishes.
Whether and how a brand appears in AI assistant answers, as distinct from where it ranks in a search results page. It is a different surface with different mechanics, and a strong position in one does not transfer to the other.
Where we cover itThat a named engine returned a given company in response to a given prompt. It is the atomic observation everything else is built from, and it is a fact about one engine, one prompt and one moment rather than a general property of the brand.
Where we cover itA result observed across several engines rather than one. It is a far stronger signal than any single engine, because it is much less likely to be an artefact of one model training set or one retrieval index.
Where we cover itA result observed on only one engine. Treated as noise rather than signal unless it repeats, since one engine returning a name once is the weakest evidence available in this method.
Where we cover itThe prompts a real purchaser would actually type, as opposed to brand-name queries. They are the meaningful test: a brand appearing when asked about itself has demonstrated nothing about whether it is recommended.
Where we cover itA composite that weights observations by engine, prompt class and position rather than counting mentions equally. Unweighted mention counts systematically favour whoever is named most in passing rather than whoever is recommended.
Where we cover itThe honest boundaries of the method, which are part of the method.
The period over which a set of observations was collected. It has to travel with any figure, because engines change beneath a measurement and a number without a window is a claim about now that was true then.
Where we cover itThe property that being recommended by an AI assistant cannot be bought directly, unlike an ad slot. It is the central strategic fact about this surface: influence is indirect, slower, and correspondingly harder for a competitor to take from you.
Where we cover itThe specific assistants tested for a given result. Naming them is what makes a visibility claim checkable, and any figure that does not name its engines is not reproducible even in principle.
Where we cover itAdjacent categories that recur in the comparisons.
Software running campaigns, sequences and lifecycle messaging. It is the category most often confused with AI visibility work, because both are described as reaching buyers, and neither affects the other.
Where we cover itThe buyer-defined segments audits are organised by. The category boundary matters more than it looks: a brand can be strongly recommended in one category and absent from an adjacent one that its own marketing treats as the same market.
Where we cover itPairs that get treated as one idea in AI-visibility conversations. Each distinction changes what a measurement is worth.
AI visibility is whether assistants recommend you. Marketing automation is software for running your own outbound. Both get described as reaching buyers and they touch nothing in common: no amount of sequence tooling changes what a model says when asked for a recommendation.
Where this bitesA cross-engine result appeared on several assistants; a single-engine result appeared on one. The second is treated as noise unless it repeats, because one engine naming you once is far more likely to reflect that model training set than any general standing.
Where this bitesEngine surfaced is one raw observation: this engine, this prompt, this moment. A weighted score is a composite that adjusts for engine, prompt class and position. Reporting raw counts as a score systematically favours whoever is named most in passing over whoever is actually recommended.
Where this bitesBuyer questions are WHAT you ask; engines named are WHO you asked. A visibility figure needs both to be checkable, and most public claims in this space state neither, which makes them unreproducible in principle rather than merely unverified.
Where this bitesUnpurchasable describes why this surface is strategically different: influence cannot be bought outright. The capture window describes why any measurement of it is perishable. Together they are the argument for treating this as a position you build and re-measure rather than a slot you buy.
Where this bitesModel behaviour, ranking and coverage change continuously, so any measurement is a snapshot of the engines and prompts tested at the time. Figures here describe what was observed, not what any engine will do next.