Reference · Glossary Updated Jul 19, 2026

The GEO glossary.

Definitions for the vocabulary of Generative Engine Optimization — from GEO and the GEF Score to agentic web readiness protocols.

Terms

49 defined

Core concepts

Generative Engine OptimizationGEO: The discipline of structuring a page or brand so generative engines — ChatGPT, Claude, Gemini, Perplexity — can access, understand, trust, and cite it when composing answers. Where SEO optimized pages to rank as links, GEO optimizes pages to be cited inside synthesized answers.
Generative Engine FriendlyGEF: The measurable degree to which a page or brand can be accessed, understood, trusted, and cited by generative engines when composing their answers. GEF is the property that GEO produces.
GEF Score: A 0–100 quantification of GEF across four dimensions — Accessibility, Extractability, Authority, and Citability — built from the 16 pre-registered on-page signals (OSF: 10.17605/OSF.IO/XCG7J). It is computed from the page itself, never by querying an engine, so it is a predictor of citation rate, not a record of it.
Answer Engine OptimizationAEO: The answer-focused slice of GEO: optimizing for direct-answer surfaces such as Google AI Overviews and chat assistants, where a single synthesized answer replaces a ranked list of links.
AI Visibility (citation rate): The realized outcome: the proportion of generative-engine answers that cite or mention your domain across a set of queries. It is measured by querying engines, and is the outcome the GEF Score is built to predict — never an input to it.
Generative engine: An AI system that composes a synthesized natural-language answer from retrieved sources rather than returning a ranked list of links — for example ChatGPT, Claude, Gemini, or Perplexity.
Answer engine: A retrieval surface that returns one synthesized answer, often with citations, in place of ten blue links. Overlaps with generative engine and includes Google AI Overviews and chat assistants.
Share of Citation vs citation rate: Share of Citation expresses your citations as a proportion of all brands cited for a query set — a competitive share metric. Citation rate is the proportion of answers that cite you — an absolute presence metric. Prime Sentia uses citation rate as the GEF Score's target outcome.

The four GEF dimensions

Accessibility: Whether an engine can reach and parse the page at all: AI-bot access, HTTPS, server-rendered or static HTML, valid JSON-LD, correct canonical. In the GEF Score it acts as a gate — a page a bot cannot read caps the whole score regardless of other strengths.
Extractability: Whether content is structured for atomic extraction: an answer-first passage, a regular heading cadence, appropriate length and reading level — passages a model can lift whole. It carries the most weight in the GEF Score, because in the generative era the unit of citation is the passage.
Authority: Verifiable reasons for an engine to trust the source it is about to quote: author and byline markup, sameAs entity links, freshness, and internal-link importance — E-E-A-T extended with machine-verifiable identity.
Citability: Whether the page carries citable units ready to be lifted into an answer — definitions, statistics, named entities, and FAQ or HowTo structured data. Citability means citation-readiness, not realized citations.

Signals & mechanisms

Source of truth: The principle that the rendered page itself — not a side-channel file or self-declared metadata — is the authoritative content an engine should read. It is why the GEF Score measures the page rather than declared signals.
Structured data (JSON-LD @graph): Machine-readable schema.org annotations embedded in the page that declare entities (Organization, Article, DefinedTerm, FAQPage) and their relationships. A connected @graph links nodes by @id so engines resolve them as one entity graph instead of disconnected fragments.
Server-side rendering (SSR)SSR: Generating a page's HTML on the server so the full content is present in the initial response, readable by crawlers that do not execute JavaScript. A precondition for Accessibility.
Canonical URL: The rel="canonical" declaration of the preferred URL for a piece of content, removing duplicate-content ambiguity for crawlers and consolidating signals on one address.
E-E-A-TEEAT: Experience, Expertise, Authoritativeness, Trustworthiness — the framework for source quality, here extended with machine-verifiable author and entity identity so an engine can confirm who is behind a page.
Entity clarity (sameAs): Making a brand's or author's identity unambiguous to engines by linking the on-page entity to authoritative external references via schema.org sameAs, so the engine knows exactly which entity the page is about.
Answer-first passage (direct answer): A complete, self-contained answer to the page's core question placed early — typically within the first ~120 words — so an engine can extract it without parsing the whole document.
Chunking (chunk-friendly content): Structuring content into self-contained passages an engine can retrieve and cite independently. Contested as a lever: Google states it is not required, because its systems comprehend multiple topics on a page, so the effect is treated as engine-dependent and measured rather than assumed.
Freshness: Signals that content is current — dateModified, sitemap lastmod — used by engines as a relevance and quality cue when choosing which source to cite.
llms.txt: A proposed plain-markdown file at /llms.txt that points machine consumers to clean markdown versions of key pages, paired with llms-full.txt for the full corpus. Consumed in practice by some coding tools and answer-engine retrieval; not required by, and given no special treatment by, Google Search.
Content negotiation (markdown for agents): Serving a clean markdown representation of a page to machine consumers via the HTTP Accept: text/markdown header or .md endpoints, while humans receive HTML. A standards-based practice (RFC 9110), not cloaking, since it is the same content in a different format.
AI crawlers (user agents): The bots generative engines use to fetch pages — GPTBot (OpenAI), ClaudeBot (Anthropic), Google-Extended (Google), PerplexityBot (Perplexity). Whether they are allowed or blocked in robots.txt determines whether your content is eligible to be cited at all.
Retrieval-augmented generation (RAG)RAG: The pattern where an engine retrieves source documents and grounds its generated answer in them, citing the sources. It is the mechanism by which a generative-engine-friendly page turns into an actual citation.
Academic Repository PresenceARP: An exploratory off-page variable measuring whether a brand or author has presence in academic repositories such as OSF or arXiv. Studied as a possible authority signal, it is not part of the core GEF Score.

Agentic web readiness

Agent readiness: The degree to which a site is not just citable but operable by autonomous AI agents — discoverable, authenticable, transactable, and tool-callable. It extends GEO beyond “can an engine quote you” to “can an agent find, trust, pay, and act on your services.”
Web Bot Auth: A Cloudflare/IETF method, built on RFC 9421 HTTP Message Signatures, by which a bot cryptographically signs its requests with an Ed25519 key and publishes its public keys as a JWKS at /.well-known/http-message-signatures-directory. It proves a request's identity — not its authorization — replacing the spoofable User-Agent string.
Content Signals: A robots.txt extension (introduced by Cloudflare) that lets a site declare, beyond allow or deny, how crawled content may be used — for example search, ai-input (use in AI answers), or ai-train (use for model training) — as machine-readable directives.
Model Context Protocol (MCP)MCP: Anthropic's open standard for connecting AI models to external tools and data through a client–server JSON-RPC protocol. An MCP server card advertises which tools a server exposes, so an agent can discover and call them.
Agent2Agent (A2A)A2A: Google's open protocol for interoperability between AI agents. Each agent publishes an Agent Card — a JSON descriptor at a well-known URL advertising its skills, endpoints, and authentication — so other agents can discover it and delegate tasks. (Prime Sentia's own auditor is built on A2A internally.)
WebMCP (navigator.modelContext)WebMCP: A W3C Community Group standard from Google and Microsoft that exposes a browser-native navigator.modelContext API. A website registers structured tools through it that in-browser AI agents can discover and invoke directly — a client-side counterpart to MCP, with no server component.
x402: An open payment protocol from Coinbase and the x402 Foundation that revives the HTTP 402 “Payment Required” status code: a server answers 402 with payment terms, and a client — often an AI agent — retries with a signed stablecoin payment, settling without accounts or API keys.
Agentic Commerce Protocol (ACP)ACP: An open checkout standard from OpenAI and Stripe (Apache 2.0) that lets an AI agent complete a purchase on a buyer's behalf — cart, fulfillment, delegated payment, and OAuth-delegated authorization — without the buyer leaving the agent's interface. It powers ChatGPT Instant Checkout; sites can expose discovery at /.well-known/acp.json.
Universal Commerce Protocol (UCP)UCP: A decentralized agentic-commerce standard from Google and Shopify in which each merchant hosts its own profile as JSON at /.well-known/ucp, so any AI shopping agent can discover the merchant, browse its catalog, and check out — with no central directory or gatekeeper.
Agent discovery files: Well-known endpoints that tell agents what a site offers and how to use it: an api-catalog (RFC 9727, /.well-known/api-catalog) listing the site's APIs, OAuth 2.0 authorization-server and protected-resource metadata for authentication, an agent-skills index, and emerging conventions such as auth.md (how to authenticate) and DNS-AID (agent interfaces advertised via SVCB/HTTPS DNS records under the _agents namespace).

Engines & models

ChatGPT: OpenAI's conversational AI and one of the primary answer engines GEO targets. In search mode it retrieves the web and cites roughly three to six sources per answer, leaning on high-authority domains such as Wikipedia; in memory mode it answers from the model with no citations.
Perplexity: An answer engine that runs a real-time web search (RAG) on every query, retrieving five to ten pages and citing three or four. It rewards extractability, first-hand experience, and freshness, and cites community sources such as Reddit heavily.
Gemini: Google's family of generative models, powering the Gemini app and AI Overviews. It draws heavily on existing organic search results, so classic SEO authority strongly influences whether a page is cited.
Anthropic: The AI safety company behind the Claude family of models. Its models power the assistants and agents that read, retrieve, and cite web content across the agentic web.
Claude: Anthropic's family of large language models, used across chat, coding, and agentic products. Claude models both generate content and, as agents, retrieve and cite sources from the open web.
Opus: Anthropic's most capable Claude tier, built for deep reasoning, long-context work, and the hardest agentic tasks.
Sonnet: Anthropic's balanced Claude tier, tuned for the speed-to-quality trade-off across everyday reasoning, writing, and coding.
Haiku: Anthropic's fastest, most compact Claude tier, for low-latency, high-volume tasks where responsiveness matters most.
Fable: A model in Anthropic's Claude family.

Search engines & SERPs

Bing: Microsoft's search engine and web index. It supplies the web results behind ChatGPT search and Copilot, so a Bing-indexed page is a prerequisite for being cited by those answer engines.
Yandex: Russia's dominant search engine, with its own crawler (YandexBot) and independent index — the primary discovery and citation surface for Russian-language and regional markets.
Brave Search: A privacy-first search engine built on its own independent index (not resold from Google or Bing), with a built-in AI answer feature — an alternative index that AI answers can draw from.
SERPSERP: Search Engine Results Page — the classic ranked list of links a search engine returns. GEO's contrast case: a SERP ranks pages, while an answer engine extracts and cites them. Google now stacks AI Overviews above the organic SERP.

Get your GEF Score

See how your pages measure against the vocabulary above. A free audit scores your site across the GEF dimensions.

Run a free audit