GEO glossary: the terms of AI visibility, explained

AI visibility describes whether and how often a brand or website appears in answers from AI assistants and AI search systems, either named in the text or linked as a source. It is not a single number but a set of measures that vary by question, system and point in time.

Answer Engine Optimization (AEO) is the practice of shaping content so that answer systems take it over as the direct answer to a question. That includes voice assistants, highlighted search results and, today mostly, AI chats and AI search products that write an answer instead of listing links.

Generative Engine Optimization (GEO) covers the measures that make a website or brand more likely to be named or cited as a source in answers from generative search systems such as ChatGPT, Perplexity, Gemini or Google AI Overviews. The goal is a place in the answer text, not a rank in a result list.

A GEO audit checks whether AI search can read and understand a website and name it for relevant questions. A good audit separates technical prerequisites from the actual measurement of whether the brand appears in real answers, and provides evidence for both.

LLM Optimization (LLMO) is an umbrella term for measures aimed at making large language models know a brand, product or website, describe it correctly and recommend it in answers. The term is used largely as a synonym for GEO and AEO.

A zero-click search is a search after which the user clicks no website, because the answer already appears on the results page or in the AI chat. Examples are weather, unit conversions, featured snippets, AI Overviews and answers from AI assistants without a follow-up click on a source.

AI Mode is a mode of Google Search in which users ask questions conversationally and receive a detailed AI-generated answer with links to web sources. Unlike AI Overviews, AI Mode is a separate view that supports follow-up questions on the same topic.

AI Overviews are AI-generated summaries that Google shows for some searches above or within the regular results. They answer the query in a short text and link the web pages the information comes from. The feature is part of Google Search, not a separate product.

An answer engine is a search system that responds to a question with written answer text instead of a list of links. Examples are ChatGPT with web search, Perplexity, Microsoft Copilot and the AI features of Google Search. Sources appear as references inside or below the answer.

AI referral traffic is visits to a website that originate from a link in an AI answer, for example from ChatGPT, Perplexity, Copilot or Gemini. It is the only visibility metric you can measure directly in your own web analytics, but it captures clicks, not mentions.

AI sentiment describes the tone in which an AI answer talks about a brand: recommending, neutrally listing, or with reservations and criticism. It complements the bare mention, because being named as a negative example counts toward visibility but hurts the purchase decision.

An AI visibility score is a vendor calculated metric that condenses a brand’s visibility in AI answers into a single value. There is no shared standard: each tool sets its own formula, question list and weighting, usually without disclosing them.

Answer variance is the fact that a language model gives different answers to the same question when asked again, with different wording, different sources and different named brands. It is the reason a single query cannot support a stable statement about visibility.

A brand mention occurs when an AI answer explicitly states the name of a vendor, product or brand in its text. It is a different quantity from a source citation, where the engine uses a page from a domain as evidence without necessarily saying the name.

A buyer query is a question with purchase intent, as a potential customer asks it to a search engine or AI assistant, for example about the best provider, a recommendation or a comparison within a category. It is the test question that shows whether an AI recommends a brand.

Citation rate is the share of AI answers to a defined set of questions in which a brand is named or its website is cited as a source. It is a ratio across many runs and is only as reliable as the number of questions and repetitions behind it.

A citation source is a web page that an AI search engine retrieved while answering a question and displayed as evidence, usually as a link or footnote. It shows where the engine got its information, not which vendor it recommends.

A confidence interval gives the range in which the true value of a measured ratio lies with a stated level of certainty. In AI visibility it shows how far a citation rate may be off because of fluctuating answers and a limited sample.

Local AI visibility describes whether a business is named when someone asks an AI for a service in a specific place, for example a dental practice in a city. It depends more on directories, reviews and map services than on the business’s own website.

Mention position is the place at which a brand is first named in an AI answer, counted in the order of all named vendors. It is the AI counterpart of a ranking position in classic search results, but considerably less stable.

Prompt tracking is asking AI assistants the same questions repeatedly over time to observe which brands get named and which sources get cited. It is the measurement method behind most AI visibility tools and produces trends instead of a single snapshot.

Share of voice in AI answers is the portion of all brand mentions within a set of questions that belongs to one brand. Unlike citation rate, it does not compare against the number of answers but against the competitors that appear in the same answers.

A snapshot measurement is a single query to an AI search engine at a specific moment. It reliably documents what that one answer contained, but it cannot tell you how often a brand is named or whether it is missing in general.

Applebot-Extended is an Apple robots.txt token that lets publishers opt out of having content crawled by Applebot used to train Apple’s generative foundation models. It is not a separate crawler that fetches pages.

CCBot is the crawler of the non-profit Common Crawl Foundation, which builds a freely available archive of large parts of the public web. That archive is a widely used raw data source for research and for training language models.

ChatGPT-User is the user agent OpenAI uses to fetch web pages when a person triggers an action in ChatGPT, such as asking it to open a page. It is not an automatic crawler for search or training but a fetch on behalf of a specific user.

Claude-SearchBot is an Anthropic crawler that navigates the web to improve search result quality in Claude. Blocking it in robots.txt prevents Anthropic from indexing the content for search, which according to Anthropic may reduce the site’s visibility in user search results.

ClaudeBot is Anthropic’s web crawler. It collects public web content that, according to Anthropic, may contribute to training Claude models. Disallowing ClaudeBot in robots.txt signals that the site’s future content should be excluded from those training datasets.

Google-Extended is not a separate crawler but a robots.txt token that lets publishers control whether content Google crawls may be used to train future Gemini models and for grounding in Gemini Apps and Vertex AI. According to Google, it does not affect inclusion in Google Search.

Googlebot is the crawler Google uses to fetch web pages and add them to its search index. That index also underpins the AI features in Google Search, such as AI Overviews and AI Mode. Blocking Googlebot removes a site from search and therefore from those features too.

GPTBot is OpenAI’s web crawler that fetches publicly accessible pages to collect content for training future models. Site owners can exclude it through the user agent token GPTBot in robots.txt, separately from OpenAI’s other crawlers.

OAI-SearchBot is OpenAI’s crawler that indexes websites for the search features in ChatGPT. According to OpenAI, sites that disallow OAI-SearchBot in robots.txt will not be shown in ChatGPT search answers; the crawler is not used to train models.

Perplexity-User is the fetcher Perplexity uses to visit a web page when a user asks a question, so it can support the answer and link the page. According to Perplexity, it does not crawl the web and does not collect training data.

PerplexityBot is Perplexity’s crawler that surfaces websites and links them in Perplexity search results. According to Perplexity, it is not used to crawl content for training AI foundation models.

ai.txt is a proposed text file in a website’s root directory meant to state whether and how AI systems may use its content. It is not a recognised standard, the major AI crawlers do not document it as binding, and there is no evidence it affects citations.

FAQ schema refers to the Schema.org FAQPage markup that labels a page’s questions and answers in machine-readable form. Google has removed the related rich result from Search; the type itself still exists and does not need to be deleted.

Grounding means anchoring an AI answer in external sources retrieved at the time of the request, usually through a web search. The model bases its statements on those sources and can show them as references, instead of answering only from its training data.

A hallucination is a statement by an AI model that sounds fluent and convincing but is factually wrong or not backed by any source. Examples are invented product features, wrong prices, studies that do not exist, or quotes attributed to a source that does not contain them.

HTTP 403 Forbidden is a status code a server uses to say it understood a request but refuses to fulfil it. When an AI crawler receives this code, it cannot read the page content, no matter what robots.txt says.

Indexing is the step in which a search engine analyses a crawled page and adds it to its search index. Only indexed pages can appear in classic results, and at Google also in AI Overviews and AI Mode, which draw on the same index.

JavaScript rendering means executing JavaScript to build a page’s visible content in the browser. Crawlers that do not execute JavaScript only see the delivered HTML, which is often an almost empty page.

JSON-LD is a W3C-standardised format for embedding structured data in a web page as a separate script block, apart from the visible HTML. Google recommends it as the preferred form of Schema.org markup.

llms.txt is a proposed file format in which a website places a Markdown overview of its most important content at /llms.txt for language models. It is not a standard, and there is no evidence that AI crawlers use the file.

Query fan-out is a technique used by AI search systems in which a user question is broken into several related subqueries that are searched in parallel. The results of all subqueries feed into one combined answer. Google describes the technique for AI Overviews and AI Mode.

Retrieval-Augmented Generation (RAG) is a method in which a language model retrieves relevant text passages from a document collection or the web before writing an answer and uses them as its basis. This lets the model answer with information that is not in its training data.

robots.txt is a text file in a website’s root directory in which site owners specify which crawlers may fetch which areas. The format is standardized at the IETF as the Robots Exclusion Protocol. Its rules are a request to crawlers, not a technical access control.

Server log analysis means reviewing a website’s server logs to see which crawlers actually requested which pages and what status codes they received. For AI visibility it shows whether GPTBot, ClaudeBot or PerplexityBot visit at all.

Server-side rendering, or SSR, means the server generates and delivers the finished HTML page with all its content, instead of assembling it in the browser with JavaScript. Crawlers receive the content directly, without having to execute JavaScript.

Structured data is machine-readable markup in a page’s source, usually using the Schema.org vocabulary, that describes what the page contains: a product, an organisation, an article. Search engines use it for rich results; as a lever for citations in AI answers it is not supported by evidence.

Training data is the text a language model learned from before release, such as web pages, books and licensed sources. It determines what the model knows about a brand without web search. That knowledge ends at a cutoff date and is only updated with a new model.

A user agent is the identifier a browser, crawler or other program sends with every request to a web server. In robots.txt the term also names the token a group of rules applies to, such as GPTBot.

A WAF or bot protection block happens when a web application firewall, a CDN or a host’s bot protection rejects crawler requests before they reach the website. robots.txt may allow everything; the crawler still only receives an error page.

A citable passage is a short section of text that answers a question completely without relying on the rest of the page. AI search often extracts individual sections; a self-contained section is more likely to be carried over correctly than a scattered thought.

Content freshness describes how recent and up to date a page is and whether that is recognisable to readers and machines. For AI search with live retrieval it can influence which source is chosen for a time-sensitive question.

E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness. Google uses the concept in its guidelines for human quality raters; it is not a single ranking factor and not a measurable score.

An entity is a uniquely identifiable thing, such as a company, person, place or product, that a search engine or language model can distinguish from things with the same name. A brand recognised as an entity is confused less often.

The Knowledge Graph is Google’s database of entities and their relationships, such as which person founded a company or where a business is located. It feeds knowledge panels in Search and helps match queries to the right thing.

A listicle is an article in list form, such as “The ten best CRM tools”. In AI search such rankings are among the most frequently cited sources, which is why being included in third-party lists can be an important route to mentions in AI answers.

Third-party mentions are references to a brand on other people’s pages, such as industry articles, comparison lists, forums or review platforms, whether or not they include a link. They are considered an important influence on whether AI search recommends a brand.

Wikidata is a free, collaboratively maintained knowledge base run by the Wikimedia Foundation, in which things are stored as items with a unique identifier and linked properties. Search engines and many AI systems use it as a source of facts about entities.