AEO metrics: a working glossary

The vocabulary around AI visibility is still forming, which means the same word often means three things in three decks. These are the definitions we use, written so they can be argued with.

Core measures

Mention rate

The share of sampled runs in which a brand appears at all for a given question. Reported per question and per engine, because a company can be well represented in one system and absent from another.

Sampling matters. Answer engines are non-deterministic, so a single run is an anecdote. A mention rate is only meaningful over repeated runs of a fixed question set.

Citation share

Of the sources an answer links to or names, the proportion that are yours. Distinct from mention rate: a model can describe your product accurately while citing a competitor’s comparison page as its source. That is a visible warning that someone else owns the evidence layer in your category.

Answer accuracy

Whether what the engine says about your product, pricing, audience, or capabilities is factually correct. Scored per claim, not per answer, because a response is usually right about some things and wrong about others.

An inaccurate mention is worse than no mention. It travels, it is repeated, and it is harder to correct than an absence.

Competitive displacement

Which companies are recommended in your place, and on what grounds. The useful output is not the list of names—it is the reason attached to each one. “Recommended because it publishes an integration directory” is a task. “Recommended because it is better known” is a strategy problem.

Evidence gap

A question where no available source gives the model enough to answer confidently, including yours. Gaps are the cheapest opportunities in AEO: nobody has claimed the ground, and the first credible answer tends to become the retrieved one.

Supporting terms

Answer engine

Any system that responds to a query with a synthesized answer rather than a list of links—ChatGPT, Claude, Gemini, Perplexity, Copilot, and the AI summaries now built into traditional search.

Entity

The structured representation of a company, product, or person that a system uses to connect facts. Your entity is not your website. It is the resolved understanding assembled from every source that describes you.

Entity ambiguity

The condition where a system cannot confidently determine what a company is, usually caused by vague category language, inconsistent facts across sources, or a name shared with something better known. It is the most common and most fixable cause of absence.

Retrieval corpus

The body of material a system can draw on when constructing an answer: crawled web pages, licensed data, indexes, and live search results. Being outside the corpus is a technical problem. Being inside it and still unused is an evidence problem.

Corroboration

Independent confirmation of a claim by a source you do not control. The mechanism by which a claim moves from “asserted” to “usable”.

Zero-click influence

The case where an answer engine shapes a buying decision using your information without sending you a visit. It is the reason session-based reporting understates AEO, and the reason mention rate has to be measured directly rather than inferred from traffic.

Non-determinism

The property that identical prompts can produce different answers and different sources between runs. It is not noise to be eliminated; it is a characteristic of the medium that any honest measurement has to account for.

Terms we avoid

“AI rank.” There is no ranking to hold. Position language imports an assumption from SEO that does not survive contact with a synthesized answer.

“LLM SEO.” It suggests the same work aimed at a new target. The work differs at the level of what is being optimised—entities and evidence, not pages.

“Guaranteed placement.” Not available from any vendor, in any engine, at any price.

---

If you would find a term here that we have defined badly, or one we have missed, tell us. This page gets updated rather than replaced.