Signal is not outcome
Finding WebMCP, structured data or llms.txt records an implementation state. Success still requires a real task test.
Agentic web glossary
The terms behind agent-ready websites
A practical vocabulary for teams testing how AI agents discover, understand and operate websites.
Use precise language
Three distinctions prevent misleading conclusions
A technology can exist without helping a task. A run can succeed without proving broad readiness.
Readiness is not confidence
Readiness describes observed capability. Confidence describes how strongly repetitions and evidence support that result.
Agentic web is not one standard
The field combines established web foundations, open protocols, product behavior and emerging proposals with different maturity.
A shared vocabulary for audit decisions
Each definition is scoped to practical website testing. Related terms make the measurement model traceable.
01 / 04
Foundations
The concepts that describe agents, browser interaction and the semantic interface exposed by a page.
- AI agent
- A software system that interprets a goal, chooses actions, uses available tools and evaluates progress toward an outcome.
- RelatedBrowser agentTask contract
- Agentic browsing
- Goal-directed use of a browser by an agent to observe pages, operate controls and complete multi-step web tasks.
- RelatedBrowser agentAccessibility tree
- Agentic web
- An umbrella term for web experiences that agents can discover, understand and operate through browser or structured interfaces.
- This is a descriptive industry term, not a single formal web standard.
- RelatedAgentic browsingWebMCP
- Browser agent
- An agent that uses browser observations and input actions such as navigation, clicking, typing and scrolling.
- RelatedAgentic browsingBrowser baseline
- Accessibility tree
- A semantic representation derived from the page that exposes roles, names, states and relationships to assistive technologies and some agents.
- RelatedSemantic controlAgentic browsing
- Semantic control
- An interactive element whose role, accessible name, state and behavior make its purpose programmatically understandable.
- RelatedAccessibility treeDeterministic assertion
02 / 04
Discovery and context
Signals that help agents retrieve, interpret and attribute public information without guaranteeing a citation or ranking.
- AI search
- A search experience that synthesizes or answers with model-generated output based on retrieved sources.
- RelatedCitation readinessCanonical source
- Citation readiness
- The degree to which a source is accessible, specific, current and easy to attribute when an AI system retrieves it.
- Readiness can be tested, but no website can guarantee a citation from an external product.
- RelatedCanonical sourceStructured data
- llms.txt
- A proposed Markdown file at a well-known site path that offers language models a concise map of useful public content.
- It is an auxiliary proposal, not a replacement for crawlable canonical HTML or a ranking requirement.
- Relatedllms-full.txtCanonical source
- llms-full.txt
- An optional companion file that provides expanded public context for language models in one text resource.
- Support and conventions vary. Its presence is an inventory signal, not proof of agent readiness.
- Relatedllms.txtCanonical source
- Structured data
- Machine-readable facts embedded in or linked from a page, commonly expressed as Schema.org vocabulary in JSON-LD.
- RelatedCanonical sourceCitation readiness
- Canonical source
- The authoritative public URL selected to represent a fact or document across duplicates, translations and machine-readable summaries.
- RelatedCitation readinessllms.txt
03 / 04
Execution and interfaces
Contracts, controls and structured channels used to run a task safely and verify its final state.
- Task contract
- A versioned scenario definition containing a goal, prerequisites, allowed actions, fixtures, stop gate and success assertions.
- RelatedStop gateDeterministic assertion
- Deterministic assertion
- A machine-checkable condition that establishes whether the expected final state was reached without relying only on model judgment.
- RelatedTask contractTime to verified outcome
- Stop gate
- An explicit boundary that ends a production run before payment, submission, deletion or another irreversible action.
- RelatedTask contractModel Context Protocol
- Browser baseline
- A deterministic automation run used as a stable reference for separating website behavior from agent-specific behavior.
- RelatedBrowser agentFailure attribution
- Model Context Protocol
- An open protocol for connecting AI applications to external data sources, tools and workflows through defined capabilities.
- RelatedWebMCPTask contract
- WebMCP
- An emerging proposal for websites to expose structured browser tools that agents can discover and invoke with explicit contracts.
- Audit its lifecycle and implementation when present. Absence should not automatically fail a website.
- RelatedModel Context ProtocolTask contract
04 / 04
Evidence and measurement
Outcome, stability and diagnostic measures used to explain what happened during an agent run.
- Agent session replay
- A time-ordered, privacy-controlled record of observations, actions, page states and technical events from an agent run.
- RelatedTime to verified outcomeFailure attribution
- Time to verified outcome
- Elapsed time from the accepted task start until deterministic assertions confirm the expected final state.
- RelatedDeterministic assertionAgent session replay
- Cumulative Layout Shift
- A Core Web Vital that measures unexpected layout movement over a page visit and can expose target movement during agent interaction.
- RelatedTarget movementAgent session replay
- Target movement
- A change in an intended control's position between observation and action that can cause a click to miss or hit the wrong element.
- RelatedCumulative Layout ShiftFailure attribution
- Failure attribution
- Classification of a failed run as a website defect, agent-specific limitation, policy block or test-infrastructure problem.
- RelatedBrowser baselineConfidence score
- Confidence score
- A separate measure of how strongly the available repetitions, coverage and evidence support a readiness conclusion.
- Confidence describes evidence quality. It does not increase the website's readiness score.
- RelatedFailure attributionDeterministic assertion
Primary references
Definitions grounded in current specifications and guidance
Where a technology is proposed or emerging, the definition says so. Aiscovery does not turn optional technology into a pass requirement.
- Schema.org DefinedTermSetPrimary reference for Schema.org DefinedTermSet.
- W3C Core Accessibility API MappingsPrimary reference for W3C Core Accessibility API Mappings.
- Chrome accessibility for agentsPrimary reference for Chrome accessibility for agents.
- web.dev Cumulative Layout ShiftPrimary reference for web.dev Cumulative Layout Shift.
- Model Context ProtocolPrimary reference for Model Context Protocol.
- WebMCP documentationPrimary reference for WebMCP documentation.
- llms.txt proposalPrimary reference for llms.txt proposal.
Turn vocabulary into a testable audit
Define task contracts, collect evidence and separate website failures from agent-specific behavior.