# Glossary

> 41 terms for the field, defined plainly. Where a term has a home in the Framework or the Standards Map, it links there.

Source: https://agentrelations.org/learn/glossary

## A

### A2A

Agent2Agent, an open protocol that lets one agent discover another, learn what it can do and hand it tasks. An agent advertises itself with an agent card.

[Standards →](https://agentrelations.org/standards)

### ACP

Agentic Commerce Protocol, an open specification from OpenAI and Stripe that lets an agent complete a purchase with a merchant while the merchant stays the seller of record.

[Standards →](https://agentrelations.org/standards)

### AEO

Answer Engine Optimization: shaping content so that answer engines and assistants quote and cite it. The letters are also read as Agent Engine Optimization, and the two uses overlap, so check which one a writer means. In the Framework this work sits in Discover and Describe.

[Describe →](https://agentrelations.org/framework/domains/describe)

### Agent

Software that acts for someone: it takes a goal, decides on steps, uses tools and carries out tasks with little step-by-step direction. In this field, an agent is always acting for a principal.

[Principles →](https://agentrelations.org/framework/principles)

### Agent card

A short machine-readable file in which an agent describes itself: who runs it, what it can do and how to reach it. Defined by the A2A protocol.

[Standards →](https://agentrelations.org/standards)

### Agent endpoint

An address where an agent can ask an organization a question or take an action, such as an MCP server or an API described for agents. See Converse.

[Converse →](https://agentrelations.org/framework/domains/converse)

### Agent Relations

Everything an organization does to help external AI agents find, trust, and do business with it. It covers agents finding and presenting you, and you answering and transacting with them.

[The Framework →](https://agentrelations.org/framework)

### Agent visit

One request, or a short run of requests, made by an agent to an organization’s site or endpoints to fetch information or complete a task. A visit is counted per agent and per task, not per page.

[Measure →](https://agentrelations.org/framework/domains/measure)

### Agent-referred outcome

A result that an agent’s work led to: a booking, a purchase, an application or a lead. The Measure domain treats outcomes, not visits, as the number that matters.

[Measure →](https://agentrelations.org/framework/domains/measure)

### Agentic browser

A web browser with a built-in agent that can read pages, fill forms and click for the user. To a site, it can look like a person, so it needs identity signals to be told apart.

### Agentic commerce

Buying and selling where an agent does the shopping, comparing and checkout for the buyer. The merchant’s counterpart may be another agent.

[Transact →](https://agentrelations.org/framework/domains/transact)

### AI agent

The same as an agent, with “AI” added to say the software is driven by a language model. This site says “agent” unless the distinction matters.

### AI visibility

How often, how accurately and in what light an organization appears in the answers agents and assistants give. It is a result of Discover and Describe, and it is measured in Measure.

[Discover →](https://agentrelations.org/framework/domains/discover)

### Assistant

An agent that talks with a person and helps them over time, such as a chat assistant. All assistants are agents; not all agents are assistants, because some run with no person in the loop.

## B

### B2A

Business to agent: selling to, or serving, an agent that acts for a buyer, as opposed to selling to the buyer directly (B2C) or to a company (B2B). The term is informal.

## C

### Chief agent

An organization’s own agent that answers for it: it speaks for the organization to other agents and to people, within limits a named person sets. It is the agent-side counterpart of a front desk.

[Converse →](https://agentrelations.org/framework/domains/converse)

### Citation

A visible link from an agent’s answer back to the source it used. Citations are how a person verifies an answer, and how an organization is credited.

### Crawler

A program that fetches pages on its own schedule to build an index or a training set. It is not acting for a specific person at that moment. Compare user-agent fetch.

[Discover →](https://agentrelations.org/framework/domains/discover)

## F

### Failed agent task

A task an agent tried to complete with an organization and could not: a blocked request, a missing fact, an error or a dead end. Each one names the problem, the likely cause and the fix. It is the most useful number in Measure.

[Measure →](https://agentrelations.org/framework/domains/measure)

### Feed

A structured file an organization publishes and keeps current, such as a product catalog or a list of events, so that agents read it instead of scraping pages.

## G

### GEO

Generative Engine Optimization: improving how often and how well an organization appears in answers from generative AI systems. A narrower, content-focused cousin of Agent Relations.

[Discover →](https://agentrelations.org/framework/domains/discover)

### Grounding

Giving an agent trusted facts to base an answer on, so it quotes the source instead of recalling from training. An organization that publishes clear facts is easier to ground on.

[Describe →](https://agentrelations.org/framework/domains/describe)

## H

### Handoff

Passing a conversation or task from an agent to a person when an answer needs one. A clear handoff path is part of Converse.

[Converse →](https://agentrelations.org/framework/domains/converse)

## J

### JSON-LD

A way to embed structured data in a web page as JSON, usually using the schema.org vocabulary. It tells machines what a page is about: an organization, a product, an event.

[Describe →](https://agentrelations.org/framework/domains/describe)

## L

### llms.txt

A proposed convention: a Markdown file at /llms.txt that gives language models a short guide to a site and links to its most useful pages. It is a proposal, not a formal standard, and support varies.

[Standards →](https://agentrelations.org/standards)

## M

### Maturity level

One of six steps, from Level 0 (Closed) to Level 5 (Trusted), that says how far an organization has come with agents. Levels are assessed domain by domain.

[Maturity model →](https://agentrelations.org/framework/maturity)

### MCP

Model Context Protocol, an open protocol introduced by Anthropic in 2024 that lets an agent connect to tools and data through a standard interface. An organization runs an MCP server so agents can ask and act.

[Standards →](https://agentrelations.org/standards)

## O

### Organization

Any business, community, agency or institution that agents might deal with. The Framework uses the word for the party doing Agent Relations.

## P

### Policy for agents

A written rule about what agents may do with an organization: which are allowed in, what they may read and do, what data they may keep and who answers for the outcome.

[Trust →](https://agentrelations.org/framework/domains/trust)

### potentialAction

A schema.org property that says what a visitor can do on a page: reserve a table, buy a product, apply for a job. Agents can read it as a map of the actions an organization offers.

[Transact →](https://agentrelations.org/framework/domains/transact)

### Principal

The person or organization an agent acts for. A principal is accountable for what their agent does, and an agent’s authority comes from the principal.

[Principles →](https://agentrelations.org/framework/principles)

## R

### robots.txt

A plain text file at the root of a site that tells crawlers which paths they may fetch. It states a preference that well-behaved crawlers follow; it does not enforce anything.

[Discover →](https://agentrelations.org/framework/domains/discover)

## S

### schema.org

A shared vocabulary for describing things on the web: organizations, products, events, people. It is the most common source of structured data that agents read.

[Describe →](https://agentrelations.org/framework/domains/describe)

### Server card

A proposed machine-readable description of an MCP server, published at a well-known address, that lets an agent learn what the server offers before connecting. The format is still being settled.

[Standards →](https://agentrelations.org/standards)

### Structured data

Information published in a fixed, labeled format that a machine reads without guessing, such as JSON-LD, a feed or an API response, as opposed to prose on a page.

[Describe →](https://agentrelations.org/framework/domains/describe)

## T

### Tool call

A request an agent makes to an outside system to get information or take an action, such as checking stock or booking a slot. Each tool an organization exposes defines the calls it will accept.

[Transact →](https://agentrelations.org/framework/domains/transact)

## U

### UCP

Universal Commerce Protocol, an open standard for agentic commerce introduced by Google with retail and payments partners. It describes how agents discover what a merchant sells and complete checkout.

[Standards →](https://agentrelations.org/standards)

### User-agent fetch

A request an agent makes in real time because a person asked it something. Unlike a crawler, it is a one-off made for a principal right now, and it should be answered with that person’s task in mind.

[Measure →](https://agentrelations.org/framework/domains/measure)

## V

### Verified agent

An agent whose identity and operator a site can confirm, for example by a cryptographic signature, so that it can be given the access its identity earns.

[Trust →](https://agentrelations.org/framework/domains/trust)

## W

### Web Bot Auth

A proposal, developed in the IETF, for automated clients to prove who they are by signing their HTTP requests with HTTP Message Signatures. A site can then recognize a verified agent without relying on IP addresses.

[Standards →](https://agentrelations.org/standards)

### Well-known URI

A reserved path that starts with /.well-known/ where a site publishes a standard file, such as an agent card. Agents know where to look without being told.

Missing a term?

The field is young and the words are still settling. Where uses compete, as with AEO, we say so. Suggest a term at the [contact page](https://agentrelations.org/about/contact).
