Reference

Glossary

Canonical definitions of the field's terms, drawn from the published conventions.

Last reviewed 2026.

Agent traffic: Autonomous traffic from software acting on a user's behalf — navigating and taking actions rather than reading pages for a corpus or index — much of which is not identifiable from user-agent strings.; Defined in BA-C-6
AI visibility: The degree to which an entity is present in, and can be surfaced by, the answers that answer engines generate. It is the phenomenon this series measures — a property estimated from repeated observation, not a quantity read from a single answer — and it is quantified by the metrics of BA-C-2.; Defined in BA-C-7
Answer engine / AI assistant: One canonical pair for the systems that respond to a natural-language question with a synthesized answer rather than only a ranked list of links. "Answer engine" names the answer-generating function under measurement; "AI assistant" names the same class of systems from the user's side — the conversational, task-oriented product a person interacts with. A document uses answer engine when it means the measured system and AI assistant when it means the user-facing surface, and introduces no third term for either.; Defined in BA-C-7
BL-0 Search hygiene: The baseline level at which an entity satisfies the conditions of conventional search indexing: indexation, a valid sitemap, coherent metadata, canonicalization, redirect health, and baseline performance.; Defined in BA-C-1
BL-1 Rendering and crawler access: The level at which an entity's content is retrievable in full by AI crawlers, served as complete HTML without client-side rendering, under a deliberate, class-aware robots policy.; Defined in BA-C-1
BL-2 Structured data and entities: The level at which an entity is machine-legible as a consistent, disambiguated entity via schema.org markup and knowledge-graph links, with every marked-up fact also present in visible text.; Defined in BA-C-1
BL-3 Content form: The level at which content is written as self-contained, excerpt-safe passages supported by statistics, quotations, and citations, in structures answer engines can extract.; Defined in BA-C-1
BL-4 Distribution and sources: The level at which an entity is present in the third-party and community sources that answer engines cite, with independent authority signals.; Defined in BA-C-1
BL-5 Measurement and statistics: The level at which an entity measures its AI visibility as a distribution estimated from repeated sampling with confidence intervals, tracked over time.; Defined in BA-C-1
BL-6 Algorithmic optimization: Concept level. The level at which optimization is an iterated, measured loop evaluated against a visibility signal rather than a fixed checklist of tactics.; Defined in BA-C-1
BL-7 Corpus presence: Concept level. The level at which an entity is present in the openly crawlable web that feeds AI training corpora, distinct from the live index that feeds retrieval.; Defined in BA-C-1
BL-8 Agent-ready: The level at which an entity exposes machine-actionable, queryable endpoints so that agents can invoke it, not merely mention it.; Defined in BA-C-1
Change-point detection: An online statistical method that flags an abrupt, sustained shift in a time series. Used to separate a silent engine update from genuine movement in a brand's visibility.; Defined in BA-C-3
Citation: A source attribution an answer provides — a link or explicit reference to a specific document — as distinct from a mention. The two do not coincide: an entity can be mentioned without being cited (named in prose with no link) and cited without being mentioned (a source link whose visible text never names it). Because they move independently, a reported rate must state which it counts.; Defined in BA-C-7
Discovery Depth (DD): The degree of query constraint at which an entity first enters an engine's recommendation set — for example, the number of qualifying attributes a user must specify before the entity appears. A concept and a reporting requirement, not a probing protocol.; Defined in BA-C-2
Effective sample size: The number of independent observations a correlated sample is worth. Observations collected close together share retrieval state and are positively correlated, so the effective sample size is smaller than the raw count.; Defined in BA-C-3
Estimand: The population quantity a measurement is meant to estimate, specified before any sampling. For AI visibility it is the probability that an entity is mentioned in an answer, defined over a distribution of real user phrasings and indexed by engine, time window, and region.; Defined in BA-C-2
Generative engine optimization (GEO): The practice of improving an entity's AI visibility — the activity, as distinct from AI visibility, the phenomenon it acts on. The term is the field's accepted name, originating in a 2024 study that introduced both it and an associated benchmark; it is adopted here rather than coined. This series specifies how to measure visibility and does not publish GEO techniques.; Defined in BA-C-7
Information need: What a user is trying to find out — a goal, independent of the words used to express it and of any engine. It is the most abstract of the three query levels, and the level at which a measurement's estimand is defined (BA-C-2) and a query-intent class is assigned (BA-C-8). A single need may be served by more than one query.; Defined in BA-C-7
Measurement channel: The deployment surface through which observations are collected — the consumer interface, an official API, or a third-party intermediary. Channels are different deployments of an engine and are disclosed with every estimate; they are not interchangeable.; Defined in BA-C-3
Mention: The observable unit of AI visibility: an instance in which an entity is named, or unambiguously referred to, in the text of an answer. Whether a given span counts as a mention is settled by the detection method a study discloses under BA-C-2 §6. Visibility Probability is, at bottom, the probability that an answer contains at least one mention.; Defined in BA-C-7
Minimum disclosure set: The ten facts any published AI-visibility claim must state for the claim to be evaluable: entity and information need, point estimate with interval and method, sample size, engine with interface and version, measurement channel, time window, region, phrasing representation, refusal handling, and detection method.; Defined in BA-C-4
Parametric (closed-book) visibility: Visibility that arises from what a model internalized in its parameters during training, surfaced without any answer-time lookup; "closed-book" because the answer draws on the model's weights rather than a retrieved document. It shifts only when a model is retrained. Presence in a training corpus does not imply the model retains or reproduces the content; no causal claim is made.; Defined in BA-C-7
Partial pooling: A hierarchical estimation method in which each phrasing's estimate is shrunk toward the query-level mean by an amount set by its sample size, so that low-volume phrasings borrow strength from the query as a whole.; Defined in BA-C-3
Phrasing: A specific surface wording of a query — one lexically distinct way of putting it, and the level at which a single observation is produced. The set of real phrasings for a query is its phrasing distribution (BA-C-3); a measurement samples phrasings from that distribution, and one phrasing is a single draw, not the query itself.; Defined in BA-C-7
Phrasing distribution: The set of real, semantically equivalent ways users pose a given query. A measurement samples from this distribution rather than from a single fixed sentence.; Defined in BA-C-3
Query: A question that operationalizes an information need as something an engine can be asked — the middle of the three query levels, between the underlying need and the exact words. "Query" is reserved for this level; where a document means the abstract goal it says information need, and where it means the exact wording it says phrasing.; Defined in BA-C-7
Query-intent class: A classification of an information need by the kind of goal it expresses, used to establish and disclose which parts of the query space a visibility study covers. This convention defines five: navigational (reaching a specific, already-known entity), informational (learning about a topic with no entity in mind), comparative (weighing a set of entities against one another), constraint-based (a need progressively narrowed by qualifying conditions), and recommendation-seeking (an open request for which entity or entities to choose). Each behaves differently under measurement, so a study assigns a class per information need and does not generalize a result obtained on one class to another.; Defined in BA-C-8
Rank-biased overlap (RBO): A similarity measure for ranked lists that weights agreement at the top more heavily, used to compare recommendation sets across runs, engines, or windows. Unlike plain set overlap, it distinguishes reordering from replacement; a report using it states the persistence parameter.; Defined in BA-C-2
Recommendation set: The set of entities an answer puts forward as candidates in response to a request for options — the list an answer gives to a "which" or "who should I" question. It is the object Discovery Depth and rank-biased overlap (RBO) act on directly (BA-C-2); Share of Voice is defined on the broader set of brand mentions in sampled answers. Its membership and order are unstable across phrasings and reruns (BA-C-3). It names an answer's put-forward candidates specifically, not every entity an answer mentions.; Defined in BA-C-7
Recommendation-seeking query: The query-intent class (BA-C-8) comprising open information needs that ask which entity or entities to choose without supplying a candidate set. Its answer is a recommendation set; it carries the highest commercial stakes and the least stability across phrasings and reruns of the five classes, so single-observation claims about it are the least defensible and its sampling requirements the most demanding.; Defined in BA-C-8
Refusal: An answer in which the engine declines to respond to a query. Recorded as an availability observation, not treated as missing data.; Defined in BA-C-3
Reporting checklist: An itemized, section-by-section list of requirements a full study report must satisfy to claim methodological completeness, in the tradition of the CONSORT and PRISMA reporting guidelines. Adopting it lets a reader assess a study end-to-end and lets an author certify conformance.; Defined in BA-C-5
Retrieval crawler: A crawler that builds a retrieval or answer index an AI system queries at answer time, so that blocking it tends to remove a site from that system's generated answers.; Defined in BA-C-6
Retrieval visibility: Visibility that arises because an engine retrieves content — the entity's own pages, or third-party pages that name it — at answer time and grounds its answer on that content. It depends on crawler access and index inclusion (BA-C-6) and on the content being present and extractable, and it can change as an index updates.; Defined in BA-C-7
Search crawler: A crawler that maintains a traditional web-search index which AI features additionally draw on, so its access decision has both classic-search and AI-answer consequences.; Defined in BA-C-6
Share of Voice (SoV): An entity's share of all brand mentions (the entity plus its competitors) within the same set of sampled answers, computed by mention count or by position-adjusted weight. More robust to engine-wide retrieval shifts than absolute VP, and suited to use as a long-run metric. A valid SoV report also publishes the competitor set and the weighting scheme.; Defined in BA-C-2
The Barkhausen Criterion: The four-condition threshold rule (BA-C-3) under which a visibility change counts: significant against proper intervals, sustained across consecutive windows, not attributable to a flagged engine change, and assessed with multiplicity control.; Defined in BA-C-1
The Barkhausen Ladder: A nine-level maturity model (BL-0 through BL-8) describing an entity's readiness to be found, cited, and acted upon by AI assistants and answer engines.; Defined in BA-C-1
Training crawler: A crawler whose documented function is to collect web content for inclusion in a model-training corpus, controlled where the operator defines a per-token opt-out.; Defined in BA-C-6
User-fetch traffic: A real-time fetch of a specific page issued because a user asked an AI system about it, distinct from bulk crawling for a corpus or an index.; Defined in BA-C-6
Visibility jump: A change in measured visibility large enough and sustained enough to be distinguished from sampling noise.; Defined in BA-C-1
Visibility Probability (VP): The probability that a given entity is mentioned in an answer to a query drawn from an information need's real phrasing distribution, on a specified engine, in a specified time window and region. A single answer is one Bernoulli realization of this probability, not the probability itself. A valid VP report carries a point estimate, a confidence interval, the sample size, the engine and its version or observation date, the time window, and the region.; Defined in BA-C-2
Wilson score interval: A binomial confidence interval that stays valid when the observed proportion is near zero or one, where the normal approximation gives impossible or misleadingly narrow bounds.; Defined in BA-C-3