Open glossary

GEO key terms, defined.

Canonical, maintained definitions of the concepts, techniques, metrics, actors and standards of Generative Engine Optimization. Designed for humans and language models. If you find an improvement, write to us.

License CC BY 4.0 — you can cite, copy, and translate freely with attribution to agentsgeo.io.

AEO (Answer Engine Optimization): Near synonym of GEO with emphasis on featured snippets and direct answers in search engines.; Answer Engine Optimization (AEO) precedes GEO in public discourse. Historically aimed at appearing in Google's featured snippets and People Also Ask. Today used as a loose synonym of GEO, though some distinguish: AEO covers any interface that returns answers (including Google), GEO is reserved for generative engines specifically.
Agentic SEO: Practice of optimizing a site to be consumable not by humans but by autonomous AI agents.; Agentic SEO is the frontier where GEO and SEO become a single discipline looking forward: optimizing for agents that browse, read, decide and buy without human intervention. It implies clean APIs, exposed MCP servers, dense structured data, clear paths for actions (not just reading), and technical performance that lets an agent complete a task in few steps.
AI Citability: The degree to which a site's content can be used by a model as a direct answer.; Citability measures how ready a block of content is to be picked up and used by an LLM in its answer. It depends on structure (direct-answer first, FAQ, lists), entity clarity, parsing ease (schema), and attribution (verifiable citations, visible authorship).
AI crawler (AI bot): Automated bot that crawls the web to train or query language models.; Main AI crawlers (2026): GPTBot and ChatGPT-User (OpenAI), Claude-Web and anthropic-ai (Anthropic), PerplexityBot (Perplexity), Google-Extended (Google), Applebot-Extended (Apple), Bytespider (ByteDance / TikTok), Amazonbot (Amazon). Each has different policies regarding robots.txt compliance. Identifying them correctly in logs is the base of technical presence analysis.
AI Overview: AI-generated summary that Google shows above organic results in many searches.; Initially known as Search Generative Experience (SGE) and rebranded AI Overview, it's the synthesized answer Google generates citing a handful of sources. Appears in roughly one of four searches (2026) and significantly reduces organic clicks for informational queries — but opens the direct citation channel for brands that appear as sources.
AI Visibility: The degree to which a brand is present inside the answers of generative engines.; AI Visibility is the conceptual equivalent of SEO Visibility but measured over AI answers. It captures not only whether a brand is cited but how often, in what context, and with what prominence within the generated paragraph. It's the metric that guides GEO work over time.
Brand Mention Monitoring: Practice of measuring frequency, context, and tone with which a brand is cited by language models.; Brand Mention Monitoring for GEO consists of systematically — weekly or biweekly — querying Perplexity, ChatGPT, Claude and Gemini about the brand and its category, logging each response. The time curve is the serious metric: one mention says little; twelve months of rising mentions with qualitative context says a lot.
Citation frequency: Number of times a model cites a brand in answers to a fixed set of test queries.; Primary metric of GEO work. Measured over a defined and stable set of queries (typically 8-20) that reflect how a buyer talks about the category. Frequency is reported per engine (Perplexity, ChatGPT, Claude, Gemini) and compared week over week. Abrupt changes often coincide with model updates or algorithm changes.
Common Crawl: Public and open repository of crawled web pages, the base of many AI training datasets.; Common Crawl maintains a monthly dump of billions of public web pages. It's one of the main inputs for the datasets that train GPT, Claude, Llama and others. If your site isn't in Common Crawl, models without real-time web search probably don't know you. Verifying presence is one of the first steps in a serious GEO audit.
DefinedTerm (schema): Schema.org type to declare the canonical definition of a term within a glossary.; DefinedTerm lets a brand position itself as the canonical source of a concept's definition. When an LLM looks for the authoritative definition of a term, it prefers content marked with DefinedTerm over loose prose. It's one of GEO's most underrated levers for brands that own a new or emerging conceptual category.
Direct-answer first: Editorial pattern that puts the answer to the implicit question in the first 50 words of the page.; Language models cite the block that resolves, not the one that surrounds it. Direct-answer first inverts the traditional editorial structure (introduction, context, development, conclusion) and starts with the conclusion. It significantly raises the probability of a page fragment ending up as a verbatim citation in an AI answer.
Edge Middleware (in GEO): Function running at the network edge to serve different versions of the site depending on the visitor.; Edge Middleware (Vercel, Cloudflare Workers, Deno Deploy) intercepts requests before they reach the server and allows rewriting the response. Applied to GEO, it serves to detect AI crawlers by user-agent and return a condensed version of the content — no visual chrome, semantically dense — while humans get the normal site.
Entity SEO: Practice of optimizing so search engines and LLMs recognize the brand as a unique entity.; Entity SEO treats the brand as a node in a knowledge graph, not a set of keywords. It implies disambiguating (so the model knows it's you, not another brand with a similar name), connecting (sameAs toward Wikipedia, LinkedIn, GitHub) and reinforcing (knowsAbout with the categories you own). It's the base of sustained citability.
FAQPage (schema): Schema.org type to declare question-answer blocks that models can cite directly.; FAQPage structures content as explicit Question/Answer pairs. It's the preferred format for LLMs because it's pre-segmented in the format the model uses to answer. A page with well-structured FAQs often appears verbatim in AI answers, cited with attribution to the source URL.
GEO (Generative Engine Optimization): The discipline of optimizing a brand's content and technical structure so it gets cited by language models.; Generative Engine Optimization (GEO) is the practice of making ChatGPT, Claude, Perplexity, Gemini and other generative engines cite a brand when users ask questions about its industry. It complements classic SEO — sharing fundamentals like authority, schema and content quality — but optimizes for being cited inside the answer, not for appearing in a list of results.
Knowledge Graph: Structure of connected entities that a search engine or model uses to understand real-world relationships.; Google Knowledge Graph was the first at scale (2012). Today all search engines and many LLMs operate on similar representations: nodes (entities) connected by typed relationships. Appearing as a node (not as a loose page) is what allows a brand to be cited by its identity, not just by its literal content. Schema.org feeds knowledge graphs.
LLMO (LLM Optimization): Another synonym of GEO, more common in technical circles.; LLM Optimization (LLMO) describes the same work as GEO from a more technical angle: specifically optimizing for large language models. The difference with GEO is emphasis — LLMO focuses on how the model processes content (tokens, embeddings, semantic structure), GEO also covers editorial and authority dimensions.
llms.txt: Plain markdown file at the root of a site that summarizes the brand for AI crawlers.; Emerging standard proposed by Jeremy Howard (Answer.AI) in 2024. Lives at https://yourdomain.com/llms.txt and describes the site in dense language so AI crawlers don't have to parse HTML, CSS, or wait for JavaScript. Anthropic, Vercel, Mintlify, FastAPI and Drizzle already implement it.
MCP (Model Context Protocol): Open standard from Anthropic for LLMs to connect with external tools and knowledge bases.; Model Context Protocol standardizes how a language model talks to servers that expose tools, resources or documentation. Released by Anthropic in 2024, adoption grew quickly in 2025-2026. For enterprise GEO, exposing an MCP server with canonical brand information is the agentic form of SEO: an agent, not a human, finds the brand and uses it as a source.
RAG (Retrieval-Augmented Generation): Pattern that enriches an LLM's answer by querying an external knowledge base.; Retrieval-Augmented Generation combines a language model with a search system over documents. The model retrieves relevant content (with embeddings or keywords), reads it, and generates the answer based on that. For enterprise GEO it's relevant because many B2B implementations use RAG over curated documentation — and exposing well-structured documentation becomes an agentic sales channel.
Schema.org / JSON-LD: Open vocabulary of types and properties to describe web entities in machine-readable format.; Schema.org is the standard for structured data on the web. JSON-LD is the format recommended by Google and almost all AI crawlers for implementing it. Key types for GEO: Organization (what you are), Service (what you offer), FAQPage (questions and answers), DefinedTerm (category definitions), Article (editorial content), BreadcrumbList (navigation).
Topical Authority: Recognition — by search engines and LLMs — that a site is a trusted reference on a topic.; Topical Authority is built with sustained and deep coverage of a territory (not scattered posts on disconnected topics), external citations from authoritative sites, and editorial consistency. For GEO, it's what makes a model choose to cite your brand instead of a competitor — because it associates you with the full category, not an isolated keyword.
Training cut-off: Date up to which a model has knowledge from its training, independent of web search.; Language models are trained on datasets closed at a specific date. Later events are unknown to the model unless it uses real-time web search. For GEO this means published content takes time to reach the model — 6 to 18 months between cut-offs. That latency is a key strategic variable.
Zero-click search: A search in which the user gets the answer without clicking any result.; Zero-click searches grew from 30% to 65% between 2020 and 2025, driven by featured snippets, AI Overviews, and answers inside generative engines. For classic SEO it's a problem (less traffic). For GEO it's the game's premise: the citation inside the answer is the outcome, not the click.

How to cite this glossary

This glossary is available under the Creative Commons Attribution 4.0 license. You can copy, modify, and translate it freely — we only ask attribution to agentsgeo.io with a link to the term you cite.

Total terms: 24. Last updated: May 1, 2026.