GEO Glossary · 76 terms

AI search speaks its own language. Here you understand every word.

Generative Engine Optimization (GEO) is the work of making an AI system such as ChatGPT, Gemini or Google AI Overviews mention a company or cite it as a source. This glossary explains the terms that come up along the way — from model knowledge through grounding and re-ranking to measurement with citation rate and share of AI search.

The groups follow the path a piece of content takes into an AI answer. Every term has its own page with a definition, how it works and what it means for your website.

The guide explains how it all fits together: GEO optimization: what demonstrably works →

Group 1 of 7

How language models work

What an LLM does, where its knowledge comes from and why it turns content into numbers before comparing it.

GEO (Generative Engine Optimization) — term in the GEO Glossary also: Generative Engine Optimization, AI search optimisation, LLM optimisation, AEO, Answer engine optimisation, LLMO, GAIO: GEO is the optimisation of content and brand so that AI systems such as ChatGPT or Google AI Overviews mention a company or cite it as a source.
LLM (Large Language Model) — term in the GEO Glossary also: Large Language Model, Language model, AI model: An LLM is a language model that estimates the probability of the next token for a sequence of tokens. It generates text instead of looking it up in an index.
Token — term in the GEO Glossary also: Tokens, Tokenisation, Tokenizer: A token is the smallest unit a language model works with: a word, part of a word or a single character. All probabilities of an LLM refer to tokens.
Model knowledge — term in the GEO Glossary also: Parametric knowledge, Training knowledge: Model knowledge is everything a language model saw during training, no more and no less. Answers from model knowledge are produced without a web search.
Common Crawl — term in the GEO Glossary also: CCBot, Common Crawl dataset, Harmonic centrality, CC Rank: Common Crawl is a non-profit, freely available archive of the internet with over 300 billion web pages from 19 years, and one of the main sources for training AI models.
Knowledge cut-off — term in the GEO Glossary also: Training cut-off, Knowledge cutoff date, Data cut-off: The knowledge cut-off is the point in time up to which training data has been fed into a language model. Anything after that date the model knows only via a web search.
Embedding — term in the GEO Glossary also: Vector representation, Text embedding, Vector database: An embedding is the translation of a text into a numeric vector that represents its meaning. Texts with similar meaning lie close to each other in vector space.
Cosine similarity — term in the GEO Glossary also: Semantic similarity, Vector similarity, Cosine similarity score: Cosine similarity measures how similar two vectors are: their dot product divided by the product of their lengths. AI systems use it to judge how well a text passage matches a search query.
Chunking — term in the GEO Glossary also: Chunks, Text segmentation, Parsing & extraction, Passage indexing: Chunking is the splitting of a text into small, self-contained sections (chunks) that an AI system stores, compares and uses in answers individually.
AI agents — term in the GEO Glossary also: KI-Agenten, Browser agents, Agentic AI, AI agent, Autonomous agents: AI agents are AI systems that complete tasks independently on a person's behalf: they plan steps, read and operate websites, fill in forms and compare offers.
AI hallucination — term in the GEO Glossary also: Hallucination, LLM hallucination, Confabulation, AI misinformation: An AI hallucination is a statement by an AI system that sounds plausible but is false or rests on no source. It arises because language models produce probable answers rather than verified ones.
Context window — term in the GEO Glossary also: Context length, Context size, Token window: The context window is the amount of text, measured in tokens, that a language model takes into account at once when answering. It includes instructions, the question, retrieved sources and the answer itself.
Knowledge graph — term in the GEO Glossary also: Google Knowledge Graph, Knowledge base graph, Entity graph: A knowledge graph is a knowledge base that stores real-world things as entities with a unique identifier and maps how they relate to one another. Search engines and AI systems use it to check facts.
Retrieval-augmented generation (RAG) — term in the GEO Glossary also: RAG, Retrieval augmented generation, Retrieval-based generation: Retrieval-augmented generation (RAG) is a method in which a language model retrieves matching texts from an external source before answering and bases its answer on those texts.
SynthID (AI watermark) — term in the GEO Glossary also: AI watermark, Google SynthID, SynthID Text, SynthID Detector, KI-Wasserzeichen: SynthID is Google DeepMind's watermarking technology for AI-generated content. It embeds a signal that humans cannot perceive into text, images, audio and video, which Google's verification tools can detect.
Vector database — term in the GEO Glossary also: Vector store, Vector index, Vector DB: A vector database is a store for embeddings, the numerical vectors that represent the meaning of texts. For a given query, it finds the stored vectors that lie closest to it.

Group 2 of 7

Web search and grounding

When an AI system looks things up live on the web — and what has to happen for your page to qualify as a source.

Web search (live retrieval) — term in the GEO Glossary also: Live retrieval, Search grounding, Retrieval-Augmented Generation (RAG): Web search is the live retrieval by an AI system from the web, triggered when the model is not confident in its answer from model knowledge. Only here can current content have an effect.
Grounding — term in the GEO Glossary also: Search grounding, Grounded response, Source anchoring: Grounding is the process of anchoring an AI answer in external, current sources: the system finds matching web pages, extracts the relevant passages and builds the answer on them.
Query Fan-out — term in the GEO Glossary also: Fan-out queries, Query expansion, Fan-out: Query fan-out is the splitting of a user question into several related search queries that an AI system sends to the search index in parallel to collect sources for its answer.
Document Relevance — term in the GEO Glossary also: Page relevance, Document-level relevance: Document relevance is the ‘classic SEO share’ of grounding: a page must rank among the top results for the query or its fan-out queries to be considered a citable source at all.
Grounding Budget — term in the GEO Glossary also: Grounding quota, Grounding chunks: The grounding budget is the quota of pages and passages that an AI system, above all Gemini, draws on to ground an answer. Pages that rank well receive more of it.
Grounded Response Generation — term in the GEO Glossary also: Model context, Final answer, Answer synthesis: Grounded response generation is the final step of an AI answer: the model synthesises model knowledge and model context — user question, personalisation, media and all grounding snippets — into a text with citations.
AI Overviews — term in the GEO Glossary also: AI Overview, Google AI Overviews, AIO, KI-Übersicht, Übersicht mit KI: AI Overviews are AI-generated summaries in Google Search. They answer a question directly above or between the search results and link to the pages the answer is based on.
Google AI Mode — term in the GEO Glossary also: AI Mode, KI-Modus, Google AI Mode search, AI Mode in Google Search: Google AI Mode is Google's conversational AI search. It breaks a question into many searches using query fan-out, combines the results into an answer with links and lets users ask follow-up questions.

Group 3 of 7

From ranking to citation

The four steps in which an AI system selects its sources: initial retrieval, re-ranking, grounding snippets, answer.

Initial Retrieval — term in the GEO Glossary also: Candidate pool, Retrieval, First-stage retrieval: Initial retrieval is the first selection step of an AI system: it collects the pages that rank best for the query and its fan-out queries into a candidate pool.
Re-Ranking — term in the GEO Glossary also: Passage re-ranking, Relevance threshold, Cut-off: Re-ranking is the second selection step of an AI system: individual text passages of the candidate pages are assessed, and only the most relevant ones pass the relevance threshold into the answer.
Grounding Snippets — term in the GEO Glossary also: Grounding chunks, Snippets, Source excerpts: Grounding snippets are the text passages that an AI system extracts from source pages and places as excerpts in the model's context in order to generate the answer from them.
Mention — term in the GEO Glossary also: Brand mention in AI, AI mention, Named mention: A mention is the naming of a brand in an AI answer, regardless of whether a source is linked. It arises from model knowledge, from grounding or from both.
Citation — term in the GEO Glossary also: AI citation, Source link, Source reference, Retrieved vs. cited: A citation is an AI answer's reference to a specific source (URL). It requires that the page was found, a passage was extracted and that passage was used for the answer.
Zero-Click — term in the GEO Glossary also: Zero-click satisfaction, Zero-click search, Answer without a click, Google Zero: Zero-click describes a search that ends without a click on a website: the user receives a direct, summarised answer and is satisfied before visiting any page.
Hybrid retrieval (BM25 and embeddings) — term in the GEO Glossary also: Hybrid search, BM25 plus embeddings, Lexical-semantic search: Hybrid retrieval is the combination of lexical search such as BM25 and semantic search with embeddings. Both result lists are merged into one ranking that takes exact terms and meaning into account at the same time.
Semantic search — term in the GEO Glossary also: Meaning-based search, Vector search, Dense retrieval: Semantic search is a search method that finds results by their meaning rather than by matching words. The query and the texts are compared as embeddings, so sections with no words in common can still match.

Group 4 of 7

Content factors

What makes a text readable, extractable and citable for AI systems.

Bottom Line Up Front (BLUF) — term in the GEO Glossary also: BLUF, Answer first, Inverted pyramid: Bottom Line Up Front is the principle of putting the core statement first: direct answer or summary first, reasoning afterwards. The key insights belong in the first 30% of the page.
Semantic chunking — term in the GEO Glossary also: Chunk-optimised writing, Self-contained paragraphs, Passage optimisation: Semantic chunking is a way of writing in which every paragraph is a self-contained answer to exactly one topic and every sentence remains understandable without context.
Entity echoing — term in the GEO Glossary also: Term echo, Entity reinforcement: Entity echoing is the technique of repeating the term from a heading in the first sentence of the answer: if the H2 asks about site power, the answer begins with the words ‘site power’.
Entity — term in the GEO Glossary also: Entity density, Named entity, Knowledge graph node: An entity is a uniquely identifiable ‘thing’ that an AI system recognises as such: a brand, a product, a person, a place, a technical term. Entity density means how many of them appear early in the text.
Information gain — term in the GEO Glossary also: Uniqueness, Original content, Novelty: Information gain is the amount of new information a sentence or paragraph delivers beyond what is already in the model knowledge and in other sources. High gain makes content citable.
Topical authority — term in the GEO Glossary also: Topic authority, Topic cluster, Content cluster: Topical authority is the recognisable expertise of a domain in a topic area: does the entire content cluster signal that someone here has mastered the topic — instead of having a single page on it?
Query coverage — term in the GEO Glossary also: Search query coverage, Topic coverage: Query coverage is the coverage of a topic area: does a domain rank not only for the main query but also for the related fan-out queries an AI system generates for a question?
E-E-A-T — term in the GEO Glossary also: Experience, Expertise, Authoritativeness, Trustworthiness, E-A-T, Authorship: E-E-A-T stands for Experience, Expertise, Authoritativeness and Trustworthiness — the signals by which search engines and AI systems recognise whether content is first-hand, competent, recognised and trustworthy.
Freshness — term in the GEO Glossary also: Content freshness, Currency, lastModified: Freshness is the recognisable currency of a piece of content. AI systems cite older content significantly less often — ‘as of 2023’ loses to current content, and the date has to be technically visible.
AI-friendly content formats — term in the GEO Glossary also: LLM-readable content, AI-readable content, Structured content: AI-friendly content formats are structures that AI systems extract easily: lists, tables, comparisons as tables instead of prose, a clear heading hierarchy, Q&A headings and a summary before the footer.
llms.txt — term in the GEO Glossary also: llms-full.txt, LLMs.txt standard: llms.txt is a proposed text file in the root directory of a website that offers language models a curated overview of the most important content. A measurable effect on AI citations has not been demonstrated so far.
LLM readability — term in the GEO Glossary also: LLM-Lesbarkeit, LLM Readability Optimization, AI readability, Readability for LLMs: LLM readability is a concept coined in 2024 by Olaf Kopp (Aufgesang): content should be written and structured so that language models can process it easily and cite it as a source.

Group 5 of 7

Technology

The entry ticket: what a crawler has to find so that content is captured at all.

LLM crawlers — term in the GEO Glossary also: AI user agents, AI bots, GPTBot, Google-Extended, OAI-SearchBot: LLM crawlers are the bots of the AI providers — GPTBot, ClaudeBot, ChatGPT-User, PerplexityBot, CCBot and others — that visit web pages for training or for the live retrieval of an answer. They do not render JavaScript.
TTFB (Time to First Byte) — term in the GEO Glossary also: Time to First Byte, Server response time, Server latency: TTFB is the time from the request to the first byte of the server response. For LLM crawlers the guideline is under 500 to 800 milliseconds — otherwise the fetch is aborted, with no retry.
URL Discovery — term in the GEO Glossary also: URL findability, Crawlability, Internal linking: URL Discovery is the question of whether crawlers can find a website's relevant content at all: via HTML links, a current sitemap and without click or login walls in front of important content.
Crawl Budget — term in the GEO Glossary also: Crawl management, Crawl control, Crawl allocation: Crawl budget is the number of fetches a crawler devotes to a website in a given period. Crawl management makes sure this budget goes to the pages that really count.
Index Management — term in the GEO Glossary also: Indexing, Indexability, Noindex and canonical: Index management is the control over which URLs of a website are indexable: noindex only for content that should not appear in AI answers, canonical tags only for unavoidable duplicate content.
Accessibility tree — term in the GEO Glossary also: A11y tree, Accessibility-Baum, Barrierefreiheitsbaum, Browser accessibility tree: The accessibility tree is a simplified version of the DOM generated by the browser, holding the roles, names and states of page elements. Screen readers use it — and so, increasingly, do AI agents.
Agentic commerce — term in the GEO Glossary also: Agentic shopping, Agent-led commerce, Agentischer Handel, Agentic Commerce Optimization, ACO: Agentic commerce is commerce in which AI agents research products, compare offers and complete the purchase on a person's behalf — based on machine-readable product data.
IndexNow — term in the GEO Glossary also: Index Now, IndexNow protocol, IndexNow API, IndexNow ping: IndexNow is an open protocol that lets a website notify search engines instantly of new, changed or deleted URLs. One submission reaches all participating search engines, including Bing — Google is not among them.
Structured data (Schema.org) — term in the GEO Glossary also: Schema.org, Schema markup, JSON-LD, Structured markup: Structured data is machine-readable information in a web page's source code that uses the Schema.org vocabulary to describe what is on the page — such as a company, a product or an article.
WebMCP — term in the GEO Glossary also: Web MCP, WebMCP API, document.modelContext, Agent tools for websites: WebMCP is a proposed web standard that lets websites provide AI agents with structured tools — declaratively through annotated HTML forms or imperatively through JavaScript.

Group 6 of 7

Off-page and brand

Why the AI judges your company through third-party sites as well — and which ones count.

Brand Mentions — term in the GEO Glossary also: Unlinked mentions, Third-party mentions, Brands matter more than websites: Brand mentions are references to a brand on third-party sites — in forums, communities, trade media, comparisons. AI systems weight them even without a link, because the brand image emerges from the sum of all sources.
Consistent brand description — term in the GEO Glossary also: Brand consistency, Uniform brand image, Entity consistency: Consistent brand description means a brand is described with the same wording on every platform — on its own website and by third parties. Only then does an AI system recognise it as one entity.
Grounding page — term in the GEO Glossary also: Grounding pages, Facts page, Fact sheet, Company facts, Brand facts page, Single source of truth: A grounding page is a page on a company’s own website that provides its verified key facts in one place, dated and citable — as a reliable source for AI systems.
Digital PR & Content Distribution — term in the GEO Glossary also: Digital PR, Earned media, Content distribution: Digital PR & Content Distribution is the work of getting a brand written about in recognised publications: trade media, industry blogs, guest articles, expert quotes and placements in comparison articles.
Listicles — term in the GEO Glossary also: Best-of lists, Comparison articles, Third-party rankings: Listicles are ‘best of’ and comparison articles on third-party sites (‘The 10 best providers for …’). AI systems cite them especially often for provider and comparison questions — presence there is a direct lever for mentions.
Community presence (Reddit & forums) — term in the GEO Glossary also: Reddit presence, Forum presence, Community mentions: Community presence means authentic recommendations of a brand in relevant threads on Reddit and in specialist forums. AI systems cite such sources often — self-promotion does not help, nuanced, honest mentions do.
Review platforms — term in the GEO Glossary also: Rating platforms, Review sites, Customer reviews: Review platforms such as Trustpilot, Capterra, Kununu or Stiftung Warentest are rating sources that AI systems cite for trust and reputation questions. Presence there raises the chance of being cited significantly.
YouTube presence — term in the GEO Glossary also: Video presence, YouTube for AI visibility: YouTube presence means a brand's own videos on YouTube that AI systems use as a source. YouTube links are among the top correlation factors for visibility in AI Overviews and ChatGPT — even with low view counts.
Domain Authority — term in the GEO Glossary also: Domain strength, Backlink profile, Site authority: Domain Authority in the GEO context describes whether a domain counts as a leading voice in its field, based on backlinks (including nofollow), presence in trade publications and entries in knowledge sources such as Wikipedia.
Corroboration — term in the GEO Glossary also: Third-party corroboration, Third-party confirmation, Consensus, Source agreement: Corroboration is the confirmation of a statement by several independent sources. AI systems treat matching statements about a brand as a trust signal and are then more likely to repeat them in their answers.
Entity home — term in the GEO Glossary also: Entity home page, Entity canonical, Point of reconciliation: The entity home is the one page that authoritatively describes a brand, company or person. Algorithms reconcile what they find elsewhere about that entity against it — usually it is the homepage or the about page.

Group 7 of 7

Measurement and tracking

How AI visibility can be measured, why a position is useless here and which metrics carry the weight instead.

LLM Visibility Tracking — term in the GEO Glossary also: AI Visibility Tracking, Prompt Tracking, AI visibility measurement: LLM Visibility Tracking is the measurement of AI visibility: prompts are sent to AI systems and the answers are analysed for mentions and citations of the brand, as a probability across many runs rather than a position.
Prompt set — term in the GEO Glossary also: Proxy prompts, Prompt list, Question set, Unbranded prompts: A prompt set is a fixed set of proxy prompts that represents the topics and customer questions of a brand. It is the basis of every measurement of AI visibility, comparable to the keyword set in SEO.
Baseline measurement — term in the GEO Glossary also: Baseline, Initial measurement, Zero measurement: The baseline measurement is the first complete survey of Mention Rate, Citation Rate, Sentiment and Share of AI Search across a prompt set, before any measures, as the reference for everything measured afterwards.
Mention Rate — term in the GEO Glossary also: Brand Mention Rate, Mention frequency, Brand mention share: The Mention Rate is the share of AI answers in a prompt set in which a brand is mentioned by name, measured across many runs, per system, as a probability rather than a position.
Sentiment — term in the GEO Glossary also: Sentiment analysis, Brand Sentiment, Tonality: Sentiment describes how a brand is rated in an AI answer: positive, neutral or critical. A mention is only valuable if the sentiment is right. Mentioned is not recommended.
Citation Rate — term in the GEO Glossary also: AI Citation Rate, Source rate, Citation frequency: The Citation Rate is the share of AI answers in a prompt set in which one of your own pages is cited as a source. It measures whether content is found, extracted and used for the answer.
AI Overview Citation Rate — term in the GEO Glossary also: AIO Citation Rate, AI Overviews visibility, AI Overview source rate: The AI Overview Citation Rate measures how often a page appears as a source in Google AI Overviews, observable via the ‘Generative AI’ report in Search Console.
Share of AI Search — term in the GEO Glossary also: Share of Voice (SOV), AI Share of Voice, Share of AI answers: Share of AI Search is the share a brand holds of the AI answers to a fixed prompt set, compared with competitors and over time. In AI search, it replaces click numbers as the leading metric.
Source analysis — term in the GEO Glossary also: Citation source analysis, Source audit, Cited domains: Source analysis evaluates which third-party sites AI systems cite for the prompts in a set. It shows where a brand needs to be present, because that is where the evidence for the answers comes from.
Log files (LLM crawler requests) — term in the GEO Glossary also: Server logs, LLM crawler requests, Log file analysis: Log files show the actual requests from LLM user agents to the server. They prove that a page was retrieved, not that it ended up in an AI answer. A server log does not yet mean visibility.
Citation Share — term in the GEO Glossary also: Bing Citation Share, Citation share metric, Share of citations, Zitieranteil: Citation Share is a metric in the AI Performance report of Bing Webmaster Tools. It shows what proportion of all citations for a grounding query your website receives in Copilot and Bing's AI answers.

From the term to the number

The terms are explained. The question is how your company scores on them.

The Digital Visibility Audit measures mention rate, citation rate and share of AI search for your company on real buyer questions — the same metrics this glossary explains.

Visibility Snapshot · free

Does AI recommend you when your customers ask?

Put your customers’ question to five AI systems. In about 20 seconds you see whether you are named — and whom AI recommends instead.

ChatGPT
Claude
Gemini
Perplexity
Google AI Mode

Start snapshot

Free · no sign-up