A glossary of AI visibility

This glossary defines the main terms used in AI visibility and in the measurement of AI answers, in plain English. Where a term has a specific meaning in Citewake's published method, the entry says so. The method is described on the methodology page, and its first edition is due in early 2027.

A

AEO (answer engine optimisation). Improving the chances that an answer engine, such as an AI assistant or a search feature that writes answers, uses and cites your content. See GEO, AEO and SEO: what each means and how they differ.

AI assistant. A chat product that answers questions in natural language. The four most used by buyers are ChatGPT, Claude, Gemini and Perplexity.

AI visibility. How often and how favourably AI assistants name, recommend or cite a brand when people ask questions in its category. See what is AI visibility.

Alias. Another name for the same brand: a short form, an abbreviation or a former name. Aliases let a measurement count a brand once.

Anchor question. A benchmark question that is never edited, so the same wording can be asked over years. See anchor questions.

Archive. In Citewake's method, the stored record of every answer, so that any question can be viewed across editions.

B

Benchmark. A fixed, versioned set of scored questions for a category. See how to build a benchmark set of questions.

Brand. In measurement, a canonical entity with a name, aliases, products, a parent company and domains.

C

Category. A market defined by the job a buyer wants done, with stated boundaries.

Citation. A source an answer attributes something to, usually shown as a link. A citation is different from a mention. See mentioned, recommended or cited.

Cited-page presence. In Citewake's method, the share of usable cited third-party pages that list a brand positively, based on a saved copy of each page.

Cluster bootstrap. A way of estimating uncertainty by resampling whole questions, keeping each question's answers together. See margins of error.

Competitive share. A brand's share of all positive recommendations in a benchmark. It is relative, so it can fall when rivals gain even if the brand's own rate holds.

D

Decision question. A question where a good answer names products. See decision and informational questions.

E

Edition. One calendar month of measurement, in which each scored question is asked of each assistant once.

Entity resolution. Deciding whether different names refer to the same brand. See entity resolution in AI answers.

G

GEO (generative engine optimisation). Improving how generative AI systems describe and recommend you. Used much like AEO.

Hallucination. An answer that states something untrue as if it were fact. Checking an answer against its cited sources is the usual defence.

Grounding. When an assistant bases an answer on pages it has retrieved, rather than only on what it learned in training.

H

Head question. A broad buying question, where established leaders tend to win.

I

Informational question. A question where a good answer may name no product.

Intent. What the asker wants to do: recommend, compare, implement, evaluate price or learn.

L

Live search. When an assistant searches the web while answering, so the answer reflects current pages. See training data or live search.

M

Margin of error. A range around a figure showing how far it could move by chance.

Mention. A brand's name appearing in an answer, in any role.

Method version. A numbered change to how measurement is done. Figures are compared within a version.

N

Neutral mention. A brand named without being recommended, as history, context or an example.

Negative mention. A brand presented as one to avoid, inferior or unsuitable.

O

Observation. One question, one assistant, one collection: the full response and everything it returned.

Own-site citation rate. The share of valid answers that cite a page on a brand's own domains.

P

Panel. In Citewake's method, answers collected from customers' own checks, used for demand and discovery and kept apart from benchmark figures.

Precision and recall. Two checks on reading accuracy. Precision asks how many of the brands found were right. Recall asks how many of the real brands were found.

Prompt. The text sent to an assistant. In measurement, a scored question goes in exactly as written.

R

Recommendation. A brand presented as a choice to make for the question asked.

Recommendation rate. The share of valid answers that positively recommend a brand.

Reported retrieval presence. A page the provider's interface reports as surfaced or consulted while answering. Where an assistant reports nothing, the measure is unavailable.

Role. How a brand is presented in an answer: recommended, neutral mention or negative.

Role accuracy. How often the role assigned to a brand matches what the answer says, checked against hand-labelled examples.

S

Share of voice. A brand's portion of the mentions or recommendations in a set of answers. See share of voice in AI answers.

T

Tail question. A specific-need question, where specialists and newer products can win.

Tier. A band of benchmark questions: head (broad), mid (a use, budget or buyer type) or tail (a specific need).

Top-pick rate. The share of valid answers where a brand is the expressed first choice.

V

Valid answer. An observation that completed and returned an answer. Valid answers form the base for rates.

Validation set. A separate private set of questions used to check a benchmark, never for published figures.

Where should you go from here?

For a first look, the Citewake checker shows how assistants describe your own site. For the reasoning behind the terms, the methodology gives the full description.