# The AI Search and GEO Glossary: 30 Terms Defined **Author:** John Morabito (Founder, /winston) **Published:** June 14, 2026 **Reading time:** 12 minutes **Canonical:** https://www.winstondigitalmarketing.com/playbooks/ai-search-glossary/ Thirty terms that define how AI search works and how brands win inside it, each written as a self-contained definition you can quote. From Generative Engine Optimization and citation share to llms.txt, RAG, and AI Overviews, this is the vocabulary of the answer-engine era in plain English. Terms with a dedicated playbook link to the deeper treatment. ## Generative Engine Optimization (GEO) Generative Engine Optimization is the practice of making content more likely to be retrieved, quoted, and cited by AI answer engines such as ChatGPT, Perplexity, and Google AI Overviews. It optimizes for inclusion in a generated answer rather than for a ranked position on a results page. Winston treats it as a full discipline in [our GEO service](https://www.winstondigitalmarketing.com/services/generative-engine-optimization/). ## AI Overviews AI Overviews are Google's generated answer summaries that appear above the traditional blue links for many queries. They synthesize information from multiple web sources, link to those sources, and often resolve the question on the results page so the user never clicks through. Earning a citation inside one is the goal, and the [citation playbook](https://www.winstondigitalmarketing.com/playbooks/how-to-get-cited-by-chatgpt-in-2026/) covers the signals that get you there. ## Google AI Mode Google AI Mode is a conversational search experience that replaces the ranked results page with a generated, multi-turn answer built on Gemini. Users ask follow-up questions in one thread, and the system grounds each response in live web sources it cites inline. The signals that win citations here are detailed in [how to get cited by Gemini and AI Mode](https://www.winstondigitalmarketing.com/playbooks/how-to-rank-in-google-gemini/). ## ChatGPT search ChatGPT search is OpenAI's feature that lets ChatGPT retrieve live web results, synthesize them into an answer, and cite the sources it used. It turns the chatbot into an answer engine, so being citable by ChatGPT becomes a distinct visibility goal from ranking in Google, with its own signals and its own measurement. ## Perplexity Perplexity is an AI answer engine that responds to queries with a synthesized, conversational answer and numbered inline citations to the web sources it used. It is built around retrieval and attribution, which makes source citability the core visibility lever rather than ranked position. The engine-specific approach is in [how to rank in Perplexity](https://www.winstondigitalmarketing.com/playbooks/how-to-rank-in-perplexity/). ## Citation share Citation share is the percentage of AI-engine citations your domain wins across a fixed set of prompts, measured per engine over a set period. It replaces keyword ranking as the primary GEO metric because answer engines cite sources rather than listing ranked pages. The full instrumentation and weekly cadence is in [citation share, the metric that replaced rankings](https://www.winstondigitalmarketing.com/playbooks/citation-share-replaces-rankings/). ## Share of voice (AI) Share of voice in AI search is the proportion of AI answers, for a defined prompt set, that mention or cite your brand relative to competitors. It measures how much of the generated-answer conversation you own across engines, combining both mentions and citations into one competitive view of where you stand against the brands you compete with. ## Liftable chunk A liftable chunk is a self-contained passage, usually a section under one heading, that answers a single question completely in a few sentences. Because it needs no surrounding context, an AI engine can extract it whole and attribute it, which makes chunk structure central to GEO. The sentence-level craft of building them is in [how to write content AI cites](https://www.winstondigitalmarketing.com/playbooks/how-to-write-content-ai-cites/). ## Answer-first content Answer-first content states the direct answer in the opening sentence of a section, before any background or caveats. This structure suits both featured snippets and AI engines, which prefer to extract an unhedged core answer they can quote without reading the whole page. It is the writing habit that turns an ordinary section into a liftable chunk. ## Entity An entity is a uniquely identifiable thing (a person, brand, place, or concept) that a search system can recognize and connect to facts, rather than a string of keywords. Search and AI engines reason about entities and their relationships to decide what a page is really about, which is why establishing your brand as a clear entity matters. ## Entity SEO Entity SEO is the practice of establishing your brand, people, and key topics as recognized entities that engines can identify and trust. It uses consistent naming, connected schema with stable identifiers, and corroborating references so a brand becomes a known thing rather than a keyword. The connected-graph pattern lives in [schema markup for AI engines](https://www.winstondigitalmarketing.com/playbooks/schema-markup-for-ai-engines-2026/). ## Knowledge graph A knowledge graph is a structured database of entities and the relationships between them, used by search and AI systems to ground answers in verified facts. Google's Knowledge Graph powers knowledge panels, and being represented in it as a clear entity strengthens how engines understand a brand and its place among related people, products, and topics. ## Schema markup / structured data Schema markup, or structured data, is standardized code (usually JSON-LD from the schema.org vocabulary) that labels page content so machines can read it precisely. It tells engines what an entity is, its attributes, and how it connects to others, which supports rich results and entity recognition. The 2026 minimum is spelled out in [schema markup for AI engines](https://www.winstondigitalmarketing.com/playbooks/schema-markup-for-ai-engines-2026/). ## FAQPage schema FAQPage schema is a structured-data type that marks up a list of questions and their answers on a page. It maps cleanly onto the chunked, one-question-per-answer format AI engines prefer, making each answer easy to parse, attribute, and quote. It is one of the highest-leverage schema types for content built to be cited. ## llms.txt An llms.txt file is a proposed plain-text file at a site's root that lists its most important pages, often as links to clean markdown versions, to help AI systems find and parse the canonical content. It is an emerging convention, not an official standard, and engine support varies. We cover how we ship it alongside markdown in the [agent-ready website plan](https://www.winstondigitalmarketing.com/playbooks/agent-ready-website-4-week-plan/). ## Markdown twin A markdown twin is a clean markdown version of an HTML page, served at a parallel URL, that strips navigation, scripts, and styling down to the core content. It gives AI crawlers and retrieval systems an unambiguous, low-noise copy of the page to parse and quote. How we built ours is in [how we built this site](https://www.winstondigitalmarketing.com/playbooks/how-we-built-this-site-in-6-hours/). ## RAG (retrieval-augmented generation) Retrieval-augmented generation is an AI architecture that retrieves relevant documents from an external source at query time and feeds them to a language model so its answer is grounded in that retrieved content. Most AI answer engines use RAG to cite live sources rather than rely only on training data, which is exactly why retrievable, well-structured content earns citations. ## Grounding Grounding is the practice of tying a language model's output to verifiable external sources, such as retrieved web pages, so the answer reflects real evidence rather than the model's parametric memory alone. Grounded answers can cite their sources, which is why retrievable, citable content is the raw material an answer engine reaches for. ## Hallucination A hallucination is a confident but false or fabricated statement produced by a language model, such as an invented fact, citation, or statistic. Grounding answers in retrieved sources reduces hallucination, which is one reason answer engines lean on retrieval and inline citations rather than answering from memory alone. ## Prompt set A prompt set is a fixed, representative list of queries you run repeatedly across AI engines to measure visibility over time. It is the denominator for citation share and AI share of voice, so building a stable, intent-mapped prompt set is the first step in any GEO measurement program. The method is in [GEO prompt research](https://www.winstondigitalmarketing.com/playbooks/geo-prompt-research/). ## Mention vs citation A mention is when an AI answer names your brand in its text; a citation is when the answer links to your page as a source. A mention builds awareness and can occur without a link, while a citation drives referral traffic and credits your content as the evidence behind the answer. A serious program tracks both, because they reflect different things an engine does with your brand. ## E-E-A-T E-E-A-T stands for Experience, Expertise, Authoritativeness, and Trustworthiness, the framework in Google's Search Quality Rater Guidelines for judging content quality. It is not a direct ranking factor but a model of the signals real evaluators reward, and it shapes which sources engines treat as credible. We audit against it in [the E-E-A-T audit](https://www.winstondigitalmarketing.com/playbooks/eat-audit-google-rater-guidelines/). ## YMYL YMYL stands for Your Money or Your Life, Google's label for topics that can affect a person's health, finances, safety, or wellbeing. YMYL content is held to a higher quality and trust bar by raters and systems, so credibility signals matter more for these pages in both classic search and AI answers. ## Model Context Protocol (MCP) The Model Context Protocol is an open standard, introduced by Anthropic, that lets AI applications connect to external tools and data sources through a consistent interface. An MCP server exposes capabilities that any compatible AI client can call, standardizing how models reach outside their context. The marketing-team use cases are in [MCP servers for marketing teams](https://www.winstondigitalmarketing.com/playbooks/mcp-servers-for-marketing-teams/). ## Claude Skill A Claude Skill is Anthropic's packaging format for a reusable capability: a folder with a SKILL.md instructions file plus optional scripts and references. Claude loads a Skill automatically when a task matches its description, and because it is just files it is portable and version-controlled across Claude products. How it compares to a Custom GPT is in [Custom GPTs vs Claude Skills](https://www.winstondigitalmarketing.com/playbooks/custom-gpts-vs-claude-skills/). ## Custom GPT A Custom GPT is OpenAI's packaging format for a task-specific assistant inside ChatGPT, wrapping a system prompt, optional knowledge files, and optional Actions behind a named assistant. It lives inside ChatGPT and is distributed through the GPT Store, unlike a portable Claude Skill. The decision framework for choosing between them is in [Custom GPTs vs Claude Skills](https://www.winstondigitalmarketing.com/playbooks/custom-gpts-vs-claude-skills/). ## Agentic workflow An agentic workflow is a process in which an AI model plans and executes multi-step tasks with some autonomy, calling tools, reading results, and deciding the next action rather than producing a single response. In marketing it powers research, content production, and reporting pipelines that run with a human on the judgment calls rather than the keystrokes. ## Content decay / freshness Content decay is the gradual loss of rankings, traffic, or citations as a page ages and its information goes stale. Freshness is a signal engines weigh, especially for fast-moving topics, so keeping pages updated and honestly re-dated helps protect both search position and AI citations. The standing system for it is in [the content-refresh system for AI search](https://www.winstondigitalmarketing.com/playbooks/content-refresh-system-ai-search/). ## Doorway pages Doorway pages are low-value pages created mainly to rank for specific queries and funnel visitors to a single destination, often near-duplicates differing only by location or keyword. Google treats them as spam, so scaled local or topic pages must carry genuinely distinct, useful content. The way to build local pages at scale without crossing that line is in [50 local landing pages in a week](https://www.winstondigitalmarketing.com/playbooks/local-landing-pages-50-in-a-week/). ## Map pack The map pack, or local pack, is the block of map-based local business results Google shows for queries with local intent, typically three listings with a map. It is driven by proximity, relevance, and prominence, and remains the highest-converting placement for near-me and booking searches even as AI answers grow around it. ## Branded vs non-branded query A branded query includes a specific brand or product name; a non-branded query describes a need or category without naming a brand. Non-branded queries are where AI engines decide which brands to surface, so winning citations on them is how you reach buyers who do not yet know you. They are the contested ground in any GEO program. ## Frequently asked questions ### What is GEO? GEO stands for Generative Engine Optimization. It is the practice of making content more likely to be retrieved, quoted, and cited by AI answer engines such as ChatGPT, Perplexity, and Google AI Overviews. Where SEO optimizes for a ranked position on a results page, GEO optimizes for inclusion in a generated answer through citable chunks, entity clarity, connected schema, and authority the engines already trust. ### What is citation share? Citation share is the percentage of AI-engine citations your domain wins across a fixed set of prompts, measured per engine over a set period such as a week. It is the GEO equivalent of share of voice and the metric that replaces keyword ranking, because answer engines cite sources inside an answer rather than listing ranked pages. You measure it by running a stable prompt set across each engine and counting how often your domain is the cited source. ### What is the difference between a mention and a citation in AI search? A mention is when an AI answer names your brand in its text without necessarily linking to you. A citation is when the answer links to your page as a source of the claim. Mentions build awareness and can happen even when your site is not the source, while citations credit your content as the evidence and can drive referral traffic. A strong GEO program tracks both, because they reflect different parts of how an engine uses your brand.