Glossary · 19 terms
AI visibility glossary
The words used when measuring how a brand appears in ChatGPT, Claude, Perplexity and Google AI Overviews — defined the way Zebora uses them. Every term has its own page, so you can send a colleague straight to one.
Where a definition depends on how something is measured, we say so. A visibility figure is a property of the method that produced it, which is why these entries spend as much time on denominators as on vocabulary.
19 terms
Concepts
The vocabulary of AI search, defined without the sales pitch.
- AI Overviews and AI ModeAI Overviews are the AI-composed answers Google places above its ordinary results. AI Mode is Google's separate conversational surface. They draw on different sources often enough that being cited in one is a poor predictor of being cited in the other, so they are worth tracking as two distinct surfaces.
- AI visibilityAI visibility is how often a brand is mentioned in answers from AI assistants such as ChatGPT, Claude, Perplexity and Google AI Overviews, measured across a defined set of prompts over a defined period. The figure only means something once you know what it was divided by, and most tools do not say.
- Answer engineAn answer engine responds to a question with a composed answer rather than a list of links. ChatGPT, Claude, Perplexity, Google AI Overviews and AI Mode are all answer engines, and they differ enough in how they retrieve and cite that a brand's standing has to be measured on each one separately.
- Answer engine optimisationAEOAnswer engine optimisation is the practice of getting a brand named and cited in AI-generated answers. It is the same discipline as generative engine optimisation; the two terms emerged in parallel and are used interchangeably across the industry, including by the tools that sell measurement for it.
- CitationA citation is a source an AI answer attributes — a link or named domain the engine says it drew on. Citations reveal which content an engine trusts in your category, which makes them the most directly actionable signal in AI search: unlike a mention, you can go and change what sits on the cited page.
- EntityAn entity is a specific thing an AI system can identify and reason about — a company, a product, a person, a place — as distinct from the words used to name it. Being understood as an entity, rather than as a string that happens to appear, is what lets an assistant describe you consistently.
- Generative engine optimisationGEOGenerative engine optimisation is the practice of making a brand easier for AI assistants to find, trust and name in their answers. It covers the content on your own site, the structured data describing you, and the third-party sources engines lean on. The term comes from a 2023 research paper by Aggarwal and colleagues.
- MentionA mention is a brand's name appearing in the text of an AI answer. It is the unit Zebora counts: whether the assistant said your name when a buyer asked a question in your category. A mention is not the same as a citation, and the two diverge sharply between engines.
- Query fan-outQuery fan-out is what happens when an assistant takes one question and expands it into several internal searches before composing an answer. The buyer asks one thing; the engine runs many. Which pages get retrieved, and which of those get cited, is settled across those hidden sub-queries rather than the visible one.
- RecommendationA recommendation is an AI answer going beyond naming your brand to endorsing it as the right fit for the person asking. Being present and being chosen are different outcomes, and the gap between them is where most brands lose: mentioned in the list, but never the one the assistant suggests.
Method
How the measurement is designed — sampling, weighting, and how we tell a real change from a wobble.
- Confidence rangeA confidence range is the band a reported score would be expected to fall within if the same measurement were repeated. Zebora publishes one on every figure, at every level of the roll-up, because a visibility score is an estimate from a finite sample of variable answers rather than a count of something fixed.
- Effective sample sizeEffective sample size is the honest count of how much information a weighted sample actually carries. Because prompts and assistants carry unequal importance weights, a batch of several hundred answers can carry the statistical strength of considerably fewer, and every confidence range has to be built from the smaller number.
- Importance weightAn importance weight expresses how much a tag, category or assistant should matter to a brand's overall score. Weights redistribute influence between parts of the measurement; they never add to it. A heavily weighted tag moves the headline more, but no amount of weighting can push a score past what the answers support.
- Monthly batchA monthly batch is the reporting container for a brand's measurement: the weekly pool runs that collect into one period, and the unit compared month to month. Weekly values exist as a series inside the batch, but the batch is what a trend gets built from.
- Noise thresholdThe noise threshold is the smallest change between two periods that Zebora will describe as a rise or a fall. Below it, a move cannot be distinguished from what you get by measuring the same thing twice, so it is reported as stable. The threshold is worked out for each comparison rather than fixed.
- Prompt taxonomyA prompt taxonomy is the structured set of questions a brand is measured against, grouped into tags and categories that reflect how its buyers actually ask. It is the denominator of every visibility figure Zebora reports, which makes it the single design decision that determines what the number means.
- Prompt-mean ruleThe prompt-mean rule is Zebora's decision that every prompt contributes exactly one unit of weight to a score, however many times it was sampled. A question measured four times counts once, as the average of its four answers, rather than four times over. It is what keeps repeat sampling from distorting the result.
- Weekly visibility poolThe weekly visibility pool is Zebora's sampling process: the full prompt taxonomy runs every week against every tracked assistant, and those weekly runs collect into one monthly reporting batch. We measure weekly and report monthly, because a single monthly snapshot cannot tell a real move from ordinary variation.
- Within measurement noiseA change is within measurement noise when it is too small to distinguish from the ordinary variation of measuring the same thing twice. Zebora reports those moves as stable rather than as a rise or a fall, because calling them a trend would be reporting a wobble as a story.
Don't let AI decide your brand's future without you.
See exactly where you stand vs. competitors—and what to do about it.