When you read an AI visibility report, you need to know what its terms describe before you use the figures in a decision. These are the terms used across this site when discussing how brands appear in AI assistant answers. Where the industry uses a term inconsistently, the entry says so.
Brand mention
The brand is named in an answer. Nothing more is implied by the term.
A mention can be favorable, neutral, or a warning. An answer naming your company as an option and an answer naming it as the one to avoid both contain a mention. Counting mentions without reading them measures presence rather than reception, which is why a mention count alone is a weak report.
Citation
A source reference in an answer. The assistant points to a page it drew on.
The cited page and the named brand often belong to different organizations. An answer recommending your product may cite a review site, an industry roundup or a forum thread rather than your website. So a brand can be named repeatedly and never cited, and a website can be cited constantly in answers that never name its owner.
Both are worth tracking and they lead to different work. Being named is a positioning question. Being cited is usually a question about which third-party pages the answer relies on, and whether you appear on them.
Recommendation
An option presented as suitable under the conditions of the question.
This is narrower than a mention and harder to count reliably, because it depends on how the answer is phrased. "Companies often consider X, Y and Z" and "X is the best choice for your situation" are different statements, and counting both as recommendations removes that distinction.
Retrieved source
A source obtained during retrieval, while the assistant was assembling an answer. A retrieved source is not automatically a cited one. Some services report what they read separately from what they showed.
Prompt
The question sent to an assistant. In tracking, a prompt is sent word for word.
The prompt's wording defines what is measured. Changing it starts a new series because it asks a different question.
Suggested fanout query
A follow-up search an assistant reports it would run while answering.
Read this precisely. It is the service describing its own intended searches. It is not evidence of what the service privately executed, and no tool can produce that evidence from outside. Suggested fanout queries are useful as research leads about how a question decomposes, and they are not a log.
Where a query is recorded from a specific observed execution, that is an observed query, which is a different and stronger thing.
Sample
The set of runs behind a reported result.
Because assistants return different answers to the same question at different times, a single run is a single observation. The number of runs and the collection period determine how much evidence the result provides. Ask the vendor to explain its sampling method.
Persona
A described buyer whose situation is supplied with the question.
Personas matter because assistants answer conditionally. The same market question asked as a procurement officer at a large company and as a sole trader can produce different sets of names, and neither answer is more correct. Comparing across personas measures which buyer descriptions your brand appears for.
Visibility
A general term for whether and how a brand appears in answers. It has no single agreed definition and no standard unit.
Whenever a product reports a visibility figure, the useful question is what it counts: mentions, citations, share against named competitors, or a composite. Two products can both report visibility and measure different things, so the numbers are not comparable between tools.
AEO, answer engine optimization
Work aimed at how a brand appears in generated answers, with the emphasis on answers, citations and presentation.
GEO, generative engine optimization
A recognized term for broadly the same work as AEO. Usage overlaps heavily and no stable distinction between the two has emerged.
Treat both as supporting labels for the same discipline rather than as separate practices with separate methods. A vendor's choice between them is usually positioning.
AI SEO
An ambiguous term with two common meanings that are worth separating before a conversation continues:
- Using AI tools to perform conventional SEO work.
- Optimizing so that AI assistants represent a brand well.
These require different work. Ask which is meant.
LLM, large language model
The model behind an assistant's answers.
The user-facing product and the model are not the same. ChatGPT is a service that may use several models, apply search, and add its own retrieval and formatting. A statement about a model is not automatically a statement about the product built on it.
LLM SEO and LLMO appear as alternative labels for the same work as AEO and GEO. Define the intended meaning when a reader uses one.
AI Overview and AI Mode
Two different Google search experiences, often conflated.
An AI Overview is a generated summary appearing above conventional search results. AI Mode is a separate conversational experience that returns an answer and its sources rather than a page of organic results.
The practical consequence for measurement: an AI Overview can be observed together with the organic results for the same keyword, whereas AI Mode returns no organic results. They are covered in how AI Mode and AI Overviews differ.
Rank tracking, applied to answers
A recognized search term that also describes tracking appearances in AI answers.
There is often no ordered list to hold a position in. What a report measures is presence, citation and comparison against named competitors across repeated runs, not a fixed rank. A product promising a stable position in every answer is describing something the systems do not provide.
Use these definitions when preparing a visibility report or comparing products. State what each figure counts so the people reading it can interpret it correctly.
See what AI search visibility measures or how citations differ from mentions.