Glossary · 21 terms

AI visibility glossary

The words used when measuring a brand's visibility in AI assistants, defined the way RateMyPrompts uses them. Each term has its own link, so you can point a colleague straight at it.

AI visibility

How often, and how favourably, AI assistants such as ChatGPT, Perplexity, Gemini and Claude mention a brand when answering questions in its category. It is usually reported as a mention rate or share of voice across a prompt set.

Zebora: How the figure is measured and reported against a brand

AI visibility tracker

Software, such as Peec, Profound, Semrush's AI tools or Ahrefs Brand Radar, that runs a prompt set against AI assistants on a schedule and reports how often a brand appears. Most run each prompt once a day per assistant.

Branded contamination

The share of prompts in a set that name your own brand. An assistant asked about a brand tends to mention it, so branded prompts inflate visibility figures without measuring whether buyers discover you.

How much branded prompts inflate the number
Buyer register

Whether a prompt reads the way a buyer describes their situation ("we're outgrowing spreadsheets, what should we move to") or the way a marketer names the category ("best CRM software 2026"). Keyword-shaped prompts tend to return more brand mentions than buyers actually see.

Building a set from buyer language
Coverage

How much of the buying decision a prompt set reaches: the share of the category's topics, and of the six journey stages (discovery, problem, evaluation, comparison, validation and post-purchase), that have prompts in them.

How coverage is scored
Design effect

How much less information clustered observations carry than independent ones. For r repeated runs of the same prompt it is 1 + (r − 1) × ICC, and the effective number of observations is the raw count divided by it.

The design effect, worked through
Generative engine optimisation (GEO)

The practice of making a brand's content and reputation easier for AI assistants to find, trust and cite in their answers. The term comes from a 2023 research paper by Aggarwal and colleagues; it is also called answer engine optimisation (AEO).

Articles on GEO measurement
Intra-cluster correlation (ICC)

How alike repeated answers to the same prompt are, from 0 (independent) to 1 (identical). Whether a brand appears is mostly a property of the prompt, so repeated runs of one prompt are strongly correlated and add little.

Journey stage

Where a buyer is in a purchase when they ask a question: discovering the category, describing a problem, evaluating options, comparing named products, validating a choice, or using what they bought.

Long-tail prompt

A specific prompt carrying a constraint, such as a budget, team size, integration, location or compliance need. RateMyPrompts counts prompts of eight words or more that carry a constraint word.

Margin of error

The range a measured visibility rate would be expected to fall within if the same measurement were repeated, usually at 95% confidence. With 100 independent prompts and a 25% mention rate, it is roughly ±8.5 percentage points.

Why more prompts beat more repeats
Mention rate

The share of AI answers, across a prompt set, that mention a brand. Simple to read, but hard to compare across prompt formats, because some formats name more brands per answer.

Model update

A change to the model behind an AI assistant. Updates can move visibility more than any marketing change, so a version change should reset the baseline rather than be read as a trend.

Near-duplicate prompts

Prompts that ask the same thing in almost the same words. They add runs without adding information. RateMyPrompts treats a pair as near-duplicates at 75% word overlap or 85% character similarity.

Prompt Fit Score

RateMyPrompts' grade for a prompt set, out of 100, weighting seven dimensions of whether the set measures real buyers. It is banded Critical, Weak, Mixed, Strong or Excellent.

The full rubric
Prompt provenance

Where a prompt's wording came from: sales calls, support tickets, site search, Search Console, forums, a keyword tool or an AI generator. Wording taken from buyers themselves is the strongest source.

Prompt set

The fixed list of questions an AI visibility tracker asks assistants on your behalf. Every visibility figure the tracker reports is an average over this list, so the set determines what the number means.

How prompt sets are graded
Share of voice

A brand's mentions as a share of all brand mentions in the answers. More comparable than mention rate across prompt formats, because a ranking prompt that lists ten brands no longer looks worse than a question that names two.

Stable core

The part of a prompt set, typically 70 to 80%, kept word for word between measurement periods so trends stay comparable. The rest is a discovery slice that can change.

Template repetition

Many prompts sharing one opening, such as "best … for …". A set built from one template measures one kind of question many times over.

Unbranded prompt

A prompt that does not name your brand, such as "best CRM for a 12-person agency". Only unbranded prompts measure whether buyers who do not already know you find you.