AI search visibility glossary

The measurement of how AI assistants such as ChatGPT, Google AI Overviews, Perplexity, Gemini and Claude answer the questions a brand’s buyers ask: whether the brand is named, how often compared with competitors, in what position, how it is described, and which pages the answers came from.

How often, how early and how favourably AI assistants such as ChatGPT, Gemini, Perplexity, Claude and Google AI Overviews name a brand when people ask them questions in its market. It is measured across many questions and many runs, because a single answer changes from one run to the next.

The practice of making a brand and its content more likely to be named, cited and described accurately in answers written by generative AI. The term comes from a 2023 research paper of the same name. It builds on SEO rather than replacing it.

Optimising content so that answer engines, meaning AI assistants and search features that reply with an answer instead of a list of links, use it and name the brand behind it. In practice AEO and GEO describe the same work, and most teams use the terms interchangeably.

The AI-written summaries Google shows above its regular results for some searches, with links to the pages they drew on. Google says a page needs nothing special to appear in them: it has to be indexed and eligible to show in Search with a snippet.

The share of AI answers, across a fixed set of questions and engines, that name a brand at all. If a brand appears in 184 of 1,785 answers, its visibility is 10.3%. On its own it says nothing about order or tone; those are position and sentiment.

A brand's share of all brand mentions in a set of AI answers. Where visibility asks whether you were named, share of voice asks how much of the conversation you hold against competitors, so it falls when rivals are named more often even if your own mentions hold steady.

Where a brand appears, on average, in the order an AI answer names brands, counted only over the answers that name it. A position of 2.6 means the brand is typically the second or third brand named. Being named first matters because readers often stop at the first suggestion.

How favourably the AI answers that name a brand describe it, usually scored from 0 to 100. It is measured only over answers that name the brand, so a brand can have high sentiment and low visibility: rarely mentioned, but praised when it is.

A web page an AI answer links to or names as a source. A citation is different from a mention: an answer can cite your page while recommending a competitor, or name your brand without citing anything you published. Tracking both shows whether your content is being read and whether you are being chosen.

A question tracked over time, worded the way a buyer would ask it, such as “best WMS for a 3PL”. Keeping the wording fixed matters: a reworded prompt is a different question, so its results should start a new line on the chart rather than continue the old one.

The full list of questions a brand tracks, usually grouped by topic or buying stage. A good set covers the questions that would change a purchase, not every question that mentions the category, and stays stable enough that its numbers can be compared month to month.

When an AI search system answers one question by running several related searches behind the scenes and combining what it finds. Google describes AI Overviews and AI Mode working this way. It is why covering a whole topic tends to beat writing one page per keyword.

Fetching live web pages or documents and giving them to a language model so its answer is based on current sources rather than training data alone. Grounded answers are the ones a brand can influence quickly, because changing the pages that get fetched changes what the model reads.

A page that AI answers rely on for a question, such as a review site, directory or community thread, where your brand is missing or described badly. Closing a source gap means getting listed, corrected or discussed on that page, not changing your own site.

A fact about your brand that you want AI answers to state correctly, written in plain words, such as a price, a plan limit or an integration. Checking answers against a claim shows whether each engine states it correctly, states something different, contradicts it or leaves it out.

A claim that AI answers use to recommend a competitor that your own pages never state plainly, such as “bills each client natively”. If the fact is true of your product but missing from your site, the engines cannot repeat it. Closing the gap means writing it where they read.

The range around a measured number within which the true value probably lies, given how much the sample varies. AI answers change from run to run, so a visibility of 10.3% might carry a 95% margin of ±3.0 points. A change smaller than its margin should not be read as a move.

A label for a change too small to tell apart from normal run-to-run variation, because it sits inside its margin of error. Treating within-noise changes as real is the most common way teams misread AI visibility data, celebrating or panicking over random movement.

The change in a brand’s numbers after a fix, minus the change competitors saw over the same questions and days. Subtracting the field removes movement that hit everyone, such as a model update, so what remains is the part the fix can claim.

A plain-text Markdown file at a site’s root that summarises what the site is and links to its most useful pages, proposed in 2024 as a guide for language models. It is a convention, not a standard, and Google says AI Overviews need no AI text files, so treat it as a low-cost extra, not a lever.

A bot that fetches web pages for an AI company. Vendors run separate bots for separate jobs, such as building a search index, fetching a page a user asked about, or collecting training data, and each can be allowed or blocked on its own in robots.txt.