Glossary

The disciplines and what they are called

The field has several names in circulation. They overlap heavily; the differences below are the ones practitioners actually observe.

Agent Experience Optimization (AXO)

The practice of structuring a website, its content, and its services so that AI agents can find, understand, cite, and act on them reliably. AXO covers both halves of the problem: being cited by third-party answer engines, and being a good host to agents that visit your site directly.

AXO is the umbrella term used across this site. The citation half of AXO is what the wider industry calls AEO or GEO.

What is AXO? →

Answer Engine Optimization (AEO)

Optimizing content so that answer engines — systems that reply directly rather than returning a list of links — surface and cite it. In practice AEO and GEO are used interchangeably.

AEO is common in job titles, usually bolted onto SEO rather than replacing it. It also has a second, unrelated 2026 meaning in circulation: "agentic engine optimization", proposed for structuring content so agents can act on it.

Which acronym should you use? →

Generative Engine Optimization (GEO)

Optimizing content for inclusion in generative AI answers. GEO is the most widely used trade name for this discipline in 2026, and is generally treated as a synonym for AEO rather than a distinct practice.

The term originates in a 2023 academic paper. Some early published "GEO lift" percentages have since been criticized in survey work for measuring gains that were conditional on a source already being retrievable.

Which acronym should you use? →

AI visibility

How often, and how prominently, a brand or domain appears in AI-generated answers. "AI visibility" is the settled label for the metric itself, and most trackers describe themselves that way.

It did not displace AEO and GEO, which several enterprise vendors still use for the practice — the market split the job into two words rather than picking one. See our post on the acronyms for how the vocabulary divided up.

Measurement →

Agentic Experience Design (AXD)

The design discipline for creating usable, explainable, and trustworthy interactions between humans, AI agents, and the systems those agents act on. Where UX concerns direct manipulation, AXD concerns delegated autonomy — supervising, directing, and correcting an agent acting on your behalf.

What is AXD? →

Agent experience (AX)

The overall quality of an agent’s experience of a product, site, or service — the agent-facing counterpart to user experience (UX) and developer experience (DX).

How answer engines find and cite content

The mechanics that determine whether your page ends up in an answer.

Citation

A reference to a source inside an AI-generated answer, usually rendered as a link or footnote. Citations are the unit of visibility in AI search the way ranked positions were the unit in classic search.

Query fan-out

The practice of decomposing one user question into several parallel sub-queries, retrieving for each, then synthesizing a single answer. Google’s AI Mode has been documented running roughly 9–11 sub-queries per prompt; ChatGPT runs far fewer.

Fan-out is why a page can be cited for a question it does not obviously target: it matched a sub-query rather than the original prompt.

Retrieval-augmented generation (RAG)

Generating an answer by first retrieving relevant documents and then conditioning the model’s output on them. Most citing answer engines are RAG systems, which is why retrievability matters more than any single on-page tactic.

Chunking

Splitting a document into smaller passages for embedding and retrieval. Retrieval systems generally operate on passages rather than whole pages, which is why self-contained sections tend to be extracted more cleanly than sections that depend on earlier context.

Chunking happens on the retrieval side and is not something you control directly. Writing sections that stand alone helps; rewriting pages into artificial fragments to "pre-chunk" them is not supported by current evidence and can hurt readability.

Zero-click search

A search that ends without the user visiting any website, because the result page answered the question. Roughly 68% of US Google searches ended without a click between January and April 2026, up from about 60% in 2024 (SparkToro/Datos).

Grounding

Constraining a model’s answer to retrieved source material so claims can be attributed. Grounding is what makes citation possible; an ungrounded answer has nothing to cite.

Content freshness

How recently content was published or meaningfully updated. Freshness is a measurable factor in AI citation: analysis of millions of AI citations in 2026 found cited URLs were meaningfully fresher on average than the web at large.

Trust signals →

Crawlers, agents, and access control

Who is fetching your pages, under what identity, and what you can actually do about it.

AI crawler

An automated fetcher operated by an AI company. The 2026 roster that matters includes GPTBot, OAI-SearchBot, ChatGPT-User and OAI-AdsBot (OpenAI); ClaudeBot, Claude-SearchBot and Claude-User (Anthropic); Google-Extended and Google-Agent; PerplexityBot and Perplexity-User; Applebot-Extended; meta-externalagent; Amazonbot; Bytespider; and CCBot.

AI crawler controls →

User-triggered fetcher

A fetch made because a person asked for it in the moment, rather than as part of bulk crawling — ChatGPT-User, Claude-User, Perplexity-User and Google-Agent are examples. Providers generally treat these as the user acting, so they may not follow robots.txt directives aimed at crawlers.

This distinction matters legally as well as technically: in August 2026 the Ninth Circuit vacated an injunction against Perplexity’s Comet browser on the reasoning that a user-directed agent is the user acting.

Crawl-to-refer ratio

How many times an AI company crawls your site for each visitor it sends back. Cloudflare’s 2026 measurements put Anthropic around 1,800:1 and OpenAI around 850–1,300:1, against roughly 5:1 for Google Search — the clearest quantification of why the traditional crawl-for-traffic bargain has broken down.

Content Signals

A Cloudflare-originated extension to robots.txt that lets a site state permitted uses — search, AI input, and AI training — separately, rather than only allowing or denying access.

Web Bot Auth

An emerging IETF effort to let bots cryptographically prove their identity using HTTP Message Signatures, rather than relying on an easily-spoofed user-agent string. No RFC has been published yet, but it is already deployed in production by Cloudflare, Vercel, Shopify and AWS.

llms.txt

A proposed plain-text file at the root of a site that lists its most important pages in a model-friendly format. Adoption grew sharply through 2026, but consumption did not follow: a 137,000-site study found 97% of valid llms.txt files received no requests at all in May 2026, and Google has stated it does not use the file for Search or AI Overviews.

Treat llms.txt as tooling for coding agents and developer documentation workflows, not as a lever for AI citation. It is cheap to publish and harmless — just not what will get you cited.

Really Simple Licensing (RSL)

A machine-readable licensing standard letting publishers attach terms and compensation requirements to their content. It has broad publisher backing — roughly 1,500 media organizations — but as of August 2026 no AI company had signed an RSL payment agreement.

The agentic web

The protocols and interfaces through which agents act on services rather than just read them.

Model Context Protocol (MCP)

An open protocol for connecting AI models to external tools and data sources. MCP was donated to the Linux Foundation’s Agentic AI Foundation in December 2025 and is the de facto standard for exposing capabilities to agents.

The specification revision of 28 July 2026 moved the protocol core to a stateless model, removing the initialize handshake and protocol-level sessions. Material written against the 2025 spec describes a handshake that is no longer part of the core.

WebMCP

A W3C Web Machine Learning Community Group draft, co-edited by Google and Microsoft, that lets a website expose structured, callable JavaScript tools to agents running in the browser — so an agent invokes a defined function instead of guessing at the DOM. It was a headline announcement at Google I/O in May 2026 and entered a Chrome origin trial in Chrome 149.

For sites that want to be good hosts to agents rather than merely readable by them, WebMCP is the most consequential standard to watch.

Agent-to-Agent (A2A)

A protocol for agents built by different vendors to discover and delegate work to one another, using signed Agent Cards to describe capabilities. It reached a stable version 1.0 in April 2026.

Agentic commerce

Buying and selling where an AI agent performs part of the transaction on a person’s behalf. Two coalitions dominate: the Agentic Commerce Protocol (OpenAI and Stripe) and the Universal Commerce Protocol (Google with Shopify and major retailers).

The 2026 lesson from OpenAI scaling back in-chat Instant Checkout is that discovery moved into the assistant while the transaction largely stayed on the merchant’s own site.

x402

A payment protocol built on the long-dormant HTTP 402 "Payment Required" status code, allowing machine-to-machine payment for resources such as API calls or content access. Its foundation was formalized under the Linux Foundation in April 2026.

Prompt injection

An attack where instructions embedded in content an agent reads are treated by the model as commands from its user. For agentic browsers this is a structural problem rather than a bug: there is no reliable boundary between trusted user instruction and untrusted page content.

Generative UI

Interfaces assembled by a model at runtime rather than laid out in advance by a designer — typically by selecting and composing from a constrained component library.

Generative UI →

On-page structure and trust

The parts you control directly on your own pages.

Structured data (JSON-LD)

Machine-readable markup describing what a page is about, usually expressed as Schema.org vocabulary in a JSON-LD script tag. It remains foundational for rich results and for agent-facing commerce feeds.

Google has stated that structured data is not a lever for inclusion in AI Overviews, a claim some practitioners dispute. Implement it for the things it demonstrably does, and do not expect it to buy citations on its own.

Schema markup →

Semantic HTML

Using HTML elements according to their meaning — headings, lists, tables, articles — so that structure is machine-readable without inference from styling.

Agent crawlability

The degree to which an agent can reach and extract a page’s content. The most common failure is content that only exists after client-side JavaScript execution: most AI fetchers do not run JavaScript, so server-rendered or statically generated content is markedly safer.

Citation readiness

How easily a specific fact can be lifted from your page and attributed. Self-contained statements, explicit subjects instead of pronouns, named sources, and visible dates all raise it.

Trust signals

Evidence of credibility that both readers and models can evaluate: named authors with verifiable credentials, resolvable citations to primary sources, publication and modification dates, and consistent identity across the wider web.

Trust signals →

The CLEAR framework

This site’s five-part checklist for agent-ready content: Concise, Logical, Explicit, Accessible, Referenceable.

The CLEAR framework →

Canonical URL

The address you declare as the authoritative one for a page, so that duplicates consolidate rather than compete. A canonical tag that points every page at the same URL tells crawlers the whole site is one page — a common and expensive misconfiguration.

Acronyms at a glance