Skip to content
RookAtlas

AI visibility glossary

The terms behind AI visibility, defined the way RookAtlas measures them: what each counts, what it divides by, and what it can't tell you.

27 terms · Last updated 12 November 2026 · Formulas in full in the method

Measuring visibility

AI visibilityAlso: AI search visibility, LLM visibility

How often AI answers name a brand when people ask the questions its buyers ask, measured across engines and over time. In RookAtlas the headline number is the mention rate, pooled over the engines in view, with each engine's own rate beside it.

What AI visibility is and how to move it
Mention rateAlso: Visibility rate

Answers that name the brand, divided by all valid answers in scope. A rate of 40% means four in ten answers to the tracked prompts named the brand. It says nothing about where the brand was placed or whether it was recommended; those have their own measures.

AI mention tracker
Share of voiceAlso: AI share of voice, SOV

Mentions of the brand divided by mentions of every brand in the tracked set: the brand and its competitors. Shares sum to 100% across the set, so adding a competitor changes every brand's share, which is why the competitor set is versioned.

Competitor analysis
Average positionAlso: AI rank, Position in AI answers

The brand's mean rank in the answers that name it. It counts only those answers, so a brand named once, first, outranks one named often in third place: read it beside the mention rate and the number of mentions behind it.

AI rank tracker
Recommendation rate

Answers that explicitly recommend the brand, divided by all the answers analysed for it in scope. Each mention is labelled recommended, listed, offered as an alternative or cautioned against, so being named and being recommended stay apart.

Metrics in the method
Net sentiment scoreAlso: NSS, AI sentiment

Positive mentions minus negative mentions, divided by the brand's mentions labelled with a sentiment, times 100. Each mention is labelled positive, neutral, mixed or negative, so the score runs from −100 to 100 and the labels behind it stay inspectable.

Sentiment and accuracy
AI referral traffic

Visits that reach a site from a link in an AI assistant, as analytics records them. It differs from being mentioned or cited: an answer can name a brand without anyone clicking, and a visit counts here only when the assistant's link passes a referrer your analytics recognises.

GA4 AI traffic segment (free tool)

Citations and sources

CitationAlso: AI citation, Source

A page an AI answer shows as its source, inline or in a source panel. Pages an engine retrieved without showing them are counted apart, as retrieved but not cited, where the engine exposes both.

Citations and sources
Citation share

Citations of one domain divided by all citations in scope, with each domain counted at most once per answer. It shows which sites engines lean on for a topic, whoever they name.

Citations and sources
Owned citation rate

Answers that cite at least one page on a domain the brand owns, divided by all valid answers. Unlike citation share, which compares domains with each other, it tells you how often your own pages are the evidence.

Metrics in the method
Source gap

A site owned by neither the brand nor a competitor that answers cite while naming a competitor and not the brand: in RookAtlas, one cited in at least three such answers in the last 30 days, where the brand appears in under half of the answers citing it. It points to pages worth earning a mention on.

Citation outreach

Engines and answers

Answer engineAlso: AI search engine, Generative engine

An AI system that answers a question in prose, often with sources, instead of returning a list of links: ChatGPT, Gemini, Perplexity, Claude, Grok, Muse and Google's AI Overviews and AI Mode among them.

The AI visibility guide
Google AI OverviewsAlso: AIO, AI Overview

The AI summary Google shows above the usual results for some searches, with links to the pages it drew on. It appears only when Google decides to show one, so a search without an Overview can't count for or against a brand.

Google AI Overviews visibility
AI Overviews trigger rate

Searches that returned an AI Overview, divided by all searches asked. Searches without one leave the AI Overviews visibility denominator rather than counting as misses. In RookAtlas's 12 November 2026 pilot, Google showed an Overview for 96 of 100 Turkish and 95 of 100 English searches.

AI Overviews on 96 of 100 Turkish buyer questions
Google AI Mode

Google's conversational search mode, where a question gets a full AI answer and follow-ups instead of a page of results. It is a separate engine from AI Overviews, with its own answers and sources.

AI Mode vs AI Overviews
Query fan-outAlso: Fan-out queries

The searches an engine runs behind the scenes to answer one prompt. They show what the engine looked for, which is often not the words the person typed. ChatGPT, Claude, Grok and Muse report theirs through RookAtlas's providers.

Query fan-out, explained

Prompts and statistics

PromptAlso: Tracked prompt, Buyer prompt

A question tracked the way a buyer would ask it, such as "best accounting software for freelancers". A branded prompt names the brand; an unbranded one asks about the category, where most competitive visibility is won or lost.

Prompt research
Prompt panelAlso: Prompt set

The fixed set of prompts, each in a market, measured for a brand. Editing, adding or removing a prompt creates a new panel revision, so trends compare like with like instead of mixing two different sets of questions.

Prompt panel builder (free tool)
Market

The country and language a prompt is asked in, such as Turkey in Turkish or the United States in English. The same question can get different answers, and different brands, in each market.

Versions and markets in the method
Confidence intervalAlso: 95% interval, Margin of error

The range a rate is likely to fall in, given that engines answer the same prompt differently from run to run. RookAtlas shows a 95% interval on its rates: mention, recommendation and owned-citation rates, competitors' included. Its method counts a change as real only when the interval of the difference excludes zero for two periods running; until that test ships, the app shows changes without calling them real.

Why AI visibility needs intervals
Prompt clusteringAlso: Effective sample size

Answers to the same prompt resemble each other, so they carry less information than the same number of answers to different prompts. Intervals are therefore computed with prompts as clusters, and adding prompts narrows them more than re-running the same prompts.

Intervals in the method

Access and optimization

Generative engine optimizationAlso: GEO

Work to make a brand's pages and mentions more likely to be used and named in AI answers: crawlable pages, clear facts, sources and statistics, and mentions on the sites engines cite. The term comes from a 2023 research paper on generative engines.

How to improve AI visibility
Answer engine optimizationAlso: AEO

An overlapping name for the same work, older than GEO and common among agencies. Both differ from SEO in what they measure: being named and cited in an answer rather than ranking in a list of links.

AI visibility vs SEO
AI crawlerAlso: AI bot, GPTBot, OAI-SearchBot, ClaudeBot, PerplexityBot

A bot an AI company uses to fetch pages. Makers document separate bots for training, for search and for fetching a page when a user asks, and robots.txt can allow one while blocking another, so a site can stay out of training and still be found.

AI crawlers and robots.txt
llms.txt

A proposed plain-text file at a site's root that lists its key pages for language models, following the format at llmstxt.org. It is cheap to publish, but Google says it doesn't use it for Search, so it is no substitute for crawlable pages.

llms.txt generator and validator (free tool)
Evidence grade

How strong the evidence behind a suggested change is: official (the engine's maker documents it), experiment, correlation, vendor data or weak. RookAtlas puts a grade on every recommendation and keeps weak ones out of scores.

Evidence grades in the method
Prompt gap

A tracked prompt where competitors are named clearly more often than the brand: in RookAtlas, one with at least six answers across the engines in the last 30 days where the brand's 95% interval lies wholly below the most-named competitor's.

Graded actions

Where to go next

The AI visibility guide puts these terms to work, the method gives every formula and limit, and the facts page says what RookAtlas covers today.