Glossary
The words, defined.
Every term in this field means something specific, and several of them are used loosely enough to sell things. These are the definitions we work to. Where a term is marketing rather than mechanism, it says so.
AI visibility
How often AI assistants name a brand or product when answering the buying questions its customers actually ask. It is measured by sampling those questions repeatedly and recording what comes back, not by a score a tool assigns on its own.
Generative engine optimisation
The practice of improving how often, and how favourably, AI assistants recommend a brand. In substance it is mostly two things: making a site machine-readable, and getting the brand onto the third-party pages assistants read when they assemble an answer.
Also called: GEOAnswer engine optimisation
A near-synonym for generative engine optimisation. The two terms are used interchangeably in practice, and the distinction between them is a marketing distinction rather than a technical one.
Also called: AEORetrieval
The step where an assistant runs its own searches, pulls a set of pages and reads them before answering. For commercial questions the answer is assembled from those retrieved pages rather than recalled from training.
Parametric knowledge
What a model knows from training alone, with no web access. A brand absent from parametric knowledge is a multi-year problem, which is why it is measured separately from retrieval and judged over quarters rather than weeks.
Mention rate
How often a brand is named anywhere in an answer. It is the weakest of the common metrics, because being named in passing is not a recommendation and does not sell anything.
Recommendation rate
How often a brand is named as an actual pick rather than mentioned in passing. This is the number that corresponds to a buying decision, and the one worth leading a report with.
Also called: Shortlist rateConversion
Recommendation rate divided by mention rate: how often being named turns into being picked. High mentions with low conversion means a brand is known and not trusted, which has different fixes from not being known at all.
Share of voice
A brand’s mentions as a proportion of all brand mentions across a set of answers. Useful for tracking a category over time, and easy to misread: share of voice can rise while recommendation rate falls.
Corroboration
Independent sources agreeing about a brand. Assistants weight what other people’s pages say more heavily than what a brand says about itself, so corroboration is usually the constraint once a site is already readable.
Citation
A page an assistant drew on when composing an answer, usually shown as a linked source. A source cited when a brand loses and never when it wins is a specific place to go and exist.
Also called: SourceControl product
A comparable product deliberately left untouched for the length of an engagement, so that a change in the treated products can be told apart from a change in the engine.
Baseline
A measurement recorded before any change is published, with the questions, repeat counts and scoring frozen. It cannot be reconstructed afterwards, which is why it is taken first.
llms.txt
A plain-text file at the root of a site summarising what the site is and linking its most useful pages, intended for language models. It is a convention rather than a standard, and on its own it changes very little.
agents.md
A markdown file describing a site to automated agents: how it is rendered, where its structured data is, and which paths matter. A convention that costs almost nothing and is not a ranking factor.
Structured data
Machine-readable statements embedded in a page, usually as JSON-LD using schema.org types. Valid markup and useful markup are different things.
Also called: Schema markupMetafield
A custom field on a Shopify product, used to hold a fact that has no built-in home. One value can reach the visible page, the structured data and the product feed at once.
Product feed
A machine-readable export of a store’s catalogue. Several assistants read a feed instead of visiting the product page, so a fact present only in description copy is invisible to them.
Agentic commerce
Purchases initiated or completed by an AI assistant acting for a shopper. It is beginning to be reported as its own channel, which makes the effect measurable rather than inferred.
Hallucinated attribute
A specification an assistant states about a product that the brand never published. Usually a symptom of an absent fact: given nothing authoritative to quote, a plausible value gets filled in.
No commitment
See how AI is selling your products.
Get your free Alethia AI report and see where your brand stands across the AI platforms your customers are already using.
