Microsoft Copilot Glossary
Microsoft Copilot combines generative AI with Microsoft 365 apps, organisational data, web information and increasingly capable AI agents. This glossary explains the language used across Microsoft 365 Copilot, Copilot Chat, Copilot Studio, Work IQ, Cowork, agents, prompting, grounding, governance and responsible AI. Use the A–Z menu to jump directly to a letter.
A
Action
A task an agent or Copilot workflow can perform, such as creating a document, sending a message, updating a record or calling another service.
Adaptive Card
A structured card format used in Microsoft products to display information, choices and interactive controls inside conversational or workflow experiences.
Agent
An AI-powered assistant designed to pursue a goal, use instructions and knowledge, call tools and complete one or more steps on a user's behalf.
Agent 365
Microsoft's management and governance layer for discovering, securing, controlling and observing AI agents used across an organisation.
Agent Builder
A Microsoft 365 experience for creating focused agents using natural-language instructions, selected knowledge and configured capabilities.
Agentic AI
AI that can plan and carry out multi-step work rather than only producing a single response to a single prompt.
AI Agent
A software-based AI worker that can reason over a task, access approved information, choose tools and take authorised actions.
AI Model
A trained computational model that recognises patterns and generates or predicts outputs from inputs such as text, images, audio or data.
AI Safety
The practices used to reduce harmful, misleading, insecure or unintended outcomes when designing and operating AI systems.
AI Skill
A reusable capability or set of instructions that extends what an agent can do, often packaged so it can be added to multiple agents.
AI-Generated Content
Text, images, summaries, analysis, presentations, code or other material produced by an AI model from user instructions and available context.
Analyst
A Microsoft 365 Copilot agent designed for data analysis, calculations, structured reasoning and producing insights from business information.
App Context
Information available from the Microsoft 365 app in which Copilot is being used, such as the current document, email, workbook, presentation or meeting.
Authentication
The process of verifying a user's or service's identity before allowing access to Microsoft 365 data, Copilot features or connected systems.
Authorisation
The rules that determine what an authenticated user, Copilot or agent is permitted to read, create, change or execute.
Azure AI Foundry
Microsoft's Azure platform for building, evaluating and operating AI applications and agents using models, tools, data and enterprise controls.
B
Background Task
Work that an agent or automation continues without requiring the user to remain in the immediate conversational flow.
Business Chat
An earlier Microsoft term for the cross-app conversational Copilot experience that could reason over work information. Current Microsoft 365 Copilot terminology has evolved beyond this name.
Business Data
Organisational information such as documents, emails, meetings, records, reports and line-of-business data that can provide context for AI responses.
Business Process
A repeatable sequence of work steps that can be assisted, automated or orchestrated using Copilot, agents, Power Automate or Copilot Studio.
Business System
An operational application such as CRM, ERP, HR or service-management software that may be connected to Copilot through approved integrations.
Bring Your Own Data
An approach where an organisation connects its own approved information sources so an AI experience can answer questions using relevant business context.
Browser Context
Information from a webpage or browser session that an AI assistant may use when a supported feature is allowed to reference the current web content.
C
Citation
A reference included with an AI response to show which document, webpage or other source supported a statement.
Connector
An integration that gives Microsoft tools controlled access to external applications, services or data sources.
Context
The information supplied to an AI model alongside the prompt, including conversation history, files, work data, instructions and retrieved sources.
Context Window
The amount of information a model can consider at one time when processing a request and generating a response.
Conversation History
Previous prompts and responses retained within a chat so Copilot can understand follow-up questions and maintain continuity.
Copilot
Microsoft's broad name for AI assistants embedded across products to help users create, analyse, search, communicate, automate and complete work.
Copilot Agent
An AI agent accessible through Microsoft's Copilot ecosystem and configured for a particular role, task, process or body of knowledge.
Copilot Chat
Microsoft's secure workplace chat experience for asking questions, creating content and using AI, with capabilities depending on licensing and available grounding.
Copilot Control System
Microsoft's collection of administrative, security, management and measurement capabilities for governing Copilot and agents in organisations.
Copilot Dashboard
A reporting experience that helps organisations understand Copilot adoption, usage patterns and workplace impact using available Microsoft 365 signals.
Copilot in Excel
Copilot capabilities within Excel for analysing data, explaining trends, creating formulas, generating insights and assisting with spreadsheet work.
Copilot in Outlook
Copilot capabilities in Outlook for drafting, summarising and understanding email and, where supported, assisting with calendar-related work.
Copilot in PowerPoint
Copilot capabilities in PowerPoint for creating presentations, restructuring slides, summarising content and generating speaker-support material.
Copilot in Teams
Copilot capabilities in Teams for understanding meetings and conversations, summarising discussions, surfacing actions and assisting with collaboration.
Copilot in Word
Copilot capabilities in Word for drafting, rewriting, summarising, comparing and transforming document content.
Copilot Notebooks
A Microsoft Copilot workspace that brings selected files, notes and sources together so AI can work with a focused body of reference material.
Copilot Pages
A collaborative canvas for turning useful Copilot responses into editable, shareable working content that teams can develop further.
Copilot Studio
Microsoft's platform for building and managing agents, connecting knowledge and tools, creating workflows, testing behaviour and publishing AI experiences.
Cowork
A Microsoft 365 Copilot experience that can carry out multi-step work across Microsoft 365, including documents, email, meetings, Teams and files, subject to permissions and available capabilities.
D
Data Boundary
A defined geographic, tenant or service boundary that governs where organisational data is processed or stored under applicable Microsoft service commitments.
Data Grounding
Using trusted organisational or external information to give an AI model relevant facts and context before it generates an answer.
Data Loss Prevention (DLP)
Policies that identify and restrict inappropriate sharing or handling of sensitive information across supported Microsoft services.
Data Minimisation
A privacy principle of using only the information necessary for a defined task or purpose rather than exposing excessive data to a process.
Data Source
A location from which Copilot or an agent can retrieve information, such as SharePoint, OneDrive, email, Teams, a website or a business application.
Deep Research
An AI-assisted research process that investigates a topic across multiple sources, reasons over findings and produces a more developed result than a quick chat answer.
Declarative Agent
An agent configured primarily through instructions, knowledge and available capabilities rather than requiring a fully custom application to be coded.
Document Grounding
Providing one or more documents as reference material so the model can answer or generate content using those documents as context.
Draft
AI-generated working content intended for human review and refinement before it is treated as final or sent to others.
Dynamic Grounding
Retrieving relevant information at request time so the model works with current context instead of relying only on static instructions or training data.
E
eDiscovery
Microsoft Purview capabilities for identifying, preserving, collecting and reviewing organisational content for legal, regulatory or investigative needs.
Embedding
A numerical representation of meaning used by AI systems to compare text or other content and retrieve semantically similar information.
Enterprise Data Protection
Microsoft protections applied to eligible work AI experiences to help secure prompts, responses and organisational information under enterprise terms and controls.
Environment
A managed boundary in services such as Power Platform and Copilot Studio that contains resources, settings, data connections and security configuration.
Evaluation
Testing an AI system or agent against defined criteria such as accuracy, safety, usefulness, latency, reliability and task completion.
Excel Copilot
Another common way of referring to Copilot features inside Microsoft Excel for formulas, analysis, insights and spreadsheet assistance.
Extensibility
The ability to expand Copilot with additional agents, data sources, tools, connectors, actions, APIs or custom business logic.
External Data
Information stored outside Microsoft 365 that may be made available to an agent or Copilot experience through an approved integration.
Microsoft Entra ID
Microsoft's cloud identity and access service, used to authenticate users and enforce access controls across Microsoft 365 and connected applications.
F
Fabric IQ
A Microsoft intelligence layer associated with Microsoft Fabric that helps agents and AI experiences work with governed data and analytical context.
Feedback
User or evaluator input about the quality of an AI response, often used to identify gaps, improve instructions and assess agent performance.
File Grounding
Using an uploaded or referenced file as a source of factual context for a Copilot response or agent task.
Fine-Tuning
Further training a model on selected examples to adapt its behaviour or performance for a specialised use case.
Foundation Model
A large, broadly trained AI model that can be adapted or prompted for many different tasks rather than being built for only one narrow purpose.
Function Calling
A capability that lets a model request a defined software function or tool so a system can retrieve data or perform an action outside the model itself.
Forms Copilot
AI assistance in Microsoft Forms for creating, refining or working with forms and surveys where Copilot capabilities are available.
Freshness
How current the information supplied to an AI system is, particularly important when responses depend on recent organisational or web data.
G
General Availability (GA)
A product-release stage indicating that Microsoft considers a feature ready for broad production use under its published service terms.
Generative AI
Artificial intelligence that creates new content such as text, images, code, summaries or analysis in response to instructions and context.
Generative Answer
A response composed by an AI model from available instructions and knowledge rather than selected from a fixed set of prewritten answers.
Generation
The process in which an AI model produces an output token by token from the prompt and context supplied to it.
Governance
The policies, responsibilities, controls and oversight used to manage how Copilot and agents are created, accessed, monitored and used.
Graph Connector
A Microsoft integration mechanism that can bring selected external content into Microsoft search and AI experiences while preserving configured access rules.
Graph Grounding
Grounding an AI response using work information available through Microsoft Graph, such as files, mail, meetings, chats and people data the user may access.
Grounded Response
An AI answer generated with supporting source material or live context rather than relying only on the model's general learned knowledge.
Grounding
The process of supplying relevant, trusted information to an AI model so its response is more specific, current and connected to the user's task.
Guardrail
A technical or policy control designed to restrict unsafe, inappropriate, unauthorised or out-of-scope AI behaviour.
GPT
A family of generative pre-trained transformer models used for language and multimodal AI tasks. Microsoft Copilot experiences may use OpenAI models alongside Microsoft orchestration and controls.
H
Hallucination
An AI output that sounds plausible but contains invented, unsupported or incorrect information.
Hallucination Mitigation
Techniques such as grounding, citations, evaluation and human review that reduce the risk of unsupported AI-generated claims.
Handoff
The transfer of a task or conversation from one agent, system or automated step to another agent or to a human.
Harmful Content
Material that may be unsafe, abusive, discriminatory, illegal or otherwise inappropriate and is typically subject to safety controls and moderation.
History
Stored records of previous conversations or interactions that can help users revisit work and, where applicable, provide continuity for AI experiences.
Human-in-the-Loop
A design approach in which a person reviews, approves, corrects or takes responsibility for important AI-generated outputs or actions.
Hybrid Search
Search that combines multiple retrieval methods, such as keyword matching and semantic or vector similarity, to improve the relevance of results.
I
Identity
The verified user, service or agent identity used to determine authentication, access rights, audit information and permitted actions.
Image Generation
The use of generative AI to create images from natural-language instructions, reference material or other supported inputs.
Implicit Context
Relevant information supplied automatically by the application or service, such as the active document or meeting, without the user manually attaching it.
Inference
The stage at which a trained AI model processes new input and produces a prediction, answer, classification or generated output.
Information Barrier
A Microsoft Purview control that can restrict communication and collaboration between defined groups to meet organisational or regulatory requirements.
Input
Information given to an AI system, including prompts, files, images, data, instructions, conversation history and retrieved context.
Instruction
A rule or direction that tells Copilot or an agent what role to take, what goal to pursue, how to respond or what constraints to follow.
Intent
The user's underlying goal or requested outcome that an AI system attempts to infer from the words and context of a request.
Interaction History
A record of exchanges between users and an AI experience that can be used for continuity, support, evaluation or governance depending on the service.
Isolation
Separating environments, data, workloads or permissions so an agent or AI process cannot access information outside its authorised boundary.
J
Jailbreak
An attempt to manipulate an AI model into ignoring its safety instructions or producing content that would normally be restricted.
Job Automation
Using AI and workflow tools to automate repeatable parts of a role or process while leaving judgement, accountability and exceptions with people where appropriate.
JSON
JavaScript Object Notation, a structured text format widely used to exchange data between agents, APIs, tools and software services.
Just-in-Time Access
A security approach that grants privileged access only when required and typically for a limited period instead of maintaining permanent elevated permissions.
K
Key Phrase
A short phrase that captures an important concept in content and may be extracted or used to improve search, classification and summarisation.
Keyword Search
Retrieval based on matching words or phrases, often combined with semantic techniques in modern AI search systems.
Knowledge
The approved information an agent can reference, which may include documents, websites, Dataverse records, SharePoint content or connected business data.
Knowledge Graph
A structured representation of entities and relationships that helps systems understand how people, concepts, documents or business objects connect.
Knowledge Grounding
Retrieving relevant information from an approved knowledge source and supplying it to the model before an answer is generated.
Knowledge Source
A repository or service that an agent is configured to consult when answering questions, such as SharePoint, a website, files or business records.
KQL
A Microsoft query language acronym used in different services, most notably Kusto Query Language for analysing large telemetry datasets and Keyword Query Language in Microsoft search scenarios.
L
Label
A classification marker applied to content or data, such as a Microsoft Purview sensitivity label used to indicate handling and protection requirements.
Language Model
An AI model trained to process and generate human language, including tasks such as summarisation, question answering, drafting and classification.
Large Language Model (LLM)
A language model trained on very large datasets and capable of handling a broad range of natural-language and reasoning tasks.
Latency
The time between submitting a request and receiving a response or completed action from an AI service or agent.
Least Privilege
A security principle that users, apps and agents should receive only the permissions necessary to perform their authorised tasks.
Licence
The subscription entitlement that determines which Microsoft Copilot features, agents, administration tools and usage allowances are available to a user.
Lifecycle Management
The processes used to create, test, approve, deploy, monitor, update and retire agents or AI solutions in a controlled way.
Line-of-Business Data
Operational data held in business applications such as CRM, ERP, HR, finance, support or industry-specific systems.
Log
A recorded event or activity entry used for auditing, troubleshooting, monitoring, security investigation or performance analysis.
Loop Component
A portable Microsoft 365 collaborative component that can stay synchronised across apps and may be used alongside Copilot-generated working content.
M
MCP
Model Context Protocol, an open protocol used to connect AI systems to tools and data sources through a consistent interface.
Meeting Recap
A summary of a meeting that may include key discussion points, decisions, action items and references to the transcript or recording where available.
Memory
Stored contextual information that helps an AI experience maintain useful continuity, preferences or working context across interactions where the feature is supported.
Metadata
Descriptive information about content, such as author, date, file type, location, permissions or labels, that can help search and retrieval systems interpret it.
Microsoft 365 Copilot
Microsoft's enterprise AI experience integrated with Microsoft 365 apps and organisational context to help users create, analyse, communicate, search and complete work.
Microsoft 365 Copilot App
The Microsoft 365 application experience that brings together Copilot Chat, files, apps, agents, Pages, Notebooks and other work-related AI capabilities.
Microsoft Graph
Microsoft's API and data layer for accessing authorised information across Microsoft 365 services such as users, mail, calendars, files, Teams and organisational relationships.
Microsoft IQ
Microsoft's broader intelligence approach for agents, bringing together work, data and model context through services including Work IQ and related intelligence layers.
Microsoft Purview
Microsoft's data security, governance, compliance and risk platform, providing controls such as sensitivity labels, DLP, audit and eDiscovery.
Microsoft Search
Microsoft's enterprise search capabilities for finding authorised people, files, sites, messages and other work content across Microsoft 365.
Model
The trained AI component that transforms input into generated or predicted output. Copilot adds Microsoft orchestration, grounding, security and product experiences around underlying models.
Model Context Protocol (MCP)
A standardised way for AI applications and agents to discover and use external tools, resources and contextual data through MCP servers.
Moderation
Automated and human processes used to detect or restrict unsafe, inappropriate or policy-violating content in AI inputs and outputs.
Multimodal AI
AI that can work with more than one type of information, such as text, images, audio, video or structured data.
N
Named Entity Recognition
A natural-language technique that identifies entities such as people, organisations, locations, dates and products within text.
Natural Language
Ordinary human language used to communicate with Copilot and agents instead of requiring specialist programming syntax.
Natural Language Processing (NLP)
The field of AI concerned with understanding, analysing and generating human language.
Neural Network
A machine-learning architecture made of interconnected computational layers that learn patterns from data.
No-Code Agent
An agent created largely through configuration and natural-language instructions rather than traditional software development.
Non-Deterministic Output
The characteristic of generative AI where the same or similar prompt can produce different valid responses on separate runs.
Notebook
A focused workspace that groups selected source material and AI interactions around a particular subject, project or research task.
Notification
A message or alert informing a user that an agent, workflow or Microsoft 365 process has completed work or requires attention.
O
OneDrive
Microsoft's cloud file storage service for individual work files, commonly used as a source and destination for Microsoft 365 Copilot tasks.
OpenAI Model
An AI model developed by OpenAI. Microsoft uses OpenAI technology in a number of Copilot and Azure AI experiences alongside Microsoft services and controls.
Orchestration
The process of coordinating model calls, instructions, knowledge retrieval, tool use and multi-step actions to complete an AI task.
Organisational Context
Work-related information about the user's organisation, role, relationships, files, meetings and business activity that can make AI responses more relevant.
Organisational Data
Information owned or managed by an organisation, including Microsoft 365 content and approved data from connected business systems.
Organisational Search
Searching across work information such as people, documents, messages and sites while respecting the requesting user's permissions.
Outlook Copilot
A common term for Copilot features within Outlook, including assistance with drafting, summarising and understanding email and related work.
Output
The content, decision, recommendation or action produced by an AI system after processing a prompt and its available context.
Oversight
Human and organisational supervision of AI systems to ensure they remain useful, secure, compliant and aligned with intended purposes.
P
Pages
Short for Copilot Pages, the editable collaborative workspace used to develop and share useful content generated during Copilot conversations.
Permission Trimming
Filtering retrieved content so a user or agent receives only information that the underlying identity is authorised to access.
Personalisation
Adapting AI responses using relevant user context, preferences, role or work patterns while operating within available permissions and privacy controls.
Plugin
A term used for software extensions that add external capabilities to an AI assistant. Microsoft's newer agent and tool terminology has replaced some earlier plugin-based approaches.
Policy
An organisational rule or technical control that governs access, data handling, AI behaviour, security, deployment or acceptable use.
Power Automate
Microsoft's workflow automation service, often used with Copilot Studio and agents to execute business processes and connect applications.
Power Platform
Microsoft's low-code platform spanning Power Apps, Power Automate, Power BI, Copilot Studio and Dataverse for building business solutions and automation.
PowerPoint Copilot
A common term for Copilot capabilities in PowerPoint that help create, summarise, reorganise and refine presentations.
Preview
A release stage in which Microsoft makes a feature available before general availability. Preview capabilities may change and can have different support or production-use terms.
Prompt
The instruction, question or request a user gives an AI system to describe the desired task or output.
Prompt Chaining
Breaking a complex task into a sequence of prompts or AI steps where the output of one stage informs the next.
Prompt Engineering
The practice of designing clear, structured instructions and context to improve the relevance, accuracy and usefulness of AI outputs.
Prompt Gallery
A collection of reusable prompt examples that helps users discover effective ways to ask Copilot to perform common tasks.
Prompt Grounding
Adding reference information, files, context or retrieved facts to a prompt so the generated response is tied to relevant source material.
Prompt Injection
A security attack that places malicious instructions in user input or retrieved content in an attempt to alter an AI system's intended behaviour.
Protected Material
Content subject to copyright or other protections that AI systems may need to identify and handle carefully when generating or transforming material.
Q
Q&A
A question-and-answer interaction in which Copilot retrieves or reasons over relevant information and generates a response for the user.
Quality Evaluation
The assessment of AI outputs against criteria such as correctness, completeness, relevance, groundedness, tone and task success.
Query
A request for information sent to a search, retrieval or data service, often generated automatically as part of AI grounding.
Query Expansion
Adding related concepts, synonyms or alternative wording to a search request to improve the chance of retrieving useful information.
Quota
A defined usage allowance or technical limit, such as requests, capacity or credits available to an AI service, agent or tenant.
Quick Action
A predefined shortcut that starts a common Copilot task without requiring the user to write a full prompt from scratch.
R
Reasoning
The model and orchestration processes used to interpret a task, compare information, plan steps and produce a considered result.
Recap
A concise AI-generated summary of a meeting, conversation, document or other body of information.
Recommendation
An AI-generated suggestion based on available information, which should be reviewed against business context before important decisions are made.
Red Teaming
Structured testing that deliberately challenges an AI system with adversarial or difficult scenarios to identify safety and security weaknesses.
Reference File
A document supplied to Copilot as source material for drafting, summarising, answering questions or creating a related output.
Response
The answer, content or status message returned by Copilot or an agent after processing a request.
Response Grounding
Connecting an AI answer to supporting organisational, file, database or web sources so the result is based on retrievable evidence.
Responsible AI
The principles and practices used to develop and operate AI fairly, safely, transparently, securely, inclusively and with appropriate accountability.
Retention
Rules governing how long organisational content, records, conversations or audit information must be kept before deletion or review.
Retrieval
The process of finding relevant information from a search index, knowledge source or connected system for use in an AI task.
Retrieval-Augmented Generation (RAG)
An AI pattern that retrieves relevant information from external sources and includes it in the context supplied to a generative model.
Researcher
A Microsoft 365 Copilot agent designed for complex, multi-step research across available sources and for producing structured findings with supporting references.
Risk Management
The process of identifying, assessing, treating and monitoring risks arising from AI use, including security, privacy, legal, operational and accuracy risks.
Role-Based Access Control (RBAC)
An access-control method that assigns permissions according to defined roles rather than granting rights individually to every user.
S
Safe Completion
A response pattern in which an AI system remains helpful while avoiding or redirecting unsafe, disallowed or inappropriate requests.
Security Trimming
Restricting search and retrieval results so users and agents receive only information their underlying identity is authorised to access.
Semantic Index
An AI-enhanced index that represents meaning and relationships in organisational content to improve search, retrieval and grounding beyond exact keyword matching.
Semantic Search
Search based on the meaning and intent of a query rather than only matching the exact words entered.
Sensitivity Label
A Microsoft Purview classification label that can identify sensitive content and apply handling, encryption or access requirements.
Session
A period of interaction between a user and Copilot or an agent, often containing related prompts, responses and contextual state.
Shadow AI
AI tools or services used without formal organisational approval, visibility or governance, creating potential security, privacy and compliance risks.
SharePoint
Microsoft's platform for team sites, intranets, document libraries and managed organisational content, and a major source of grounding data for Microsoft 365 Copilot.
Sharing Link
A link granting defined access to a Microsoft 365 file or resource, with the actual access determined by its sharing configuration and permissions.
Single Sign-On (SSO)
An authentication approach that allows a user to access multiple authorised applications using one organisational sign-in session.
Skill
A reusable set of instructions or capabilities that can be attached to an agent to help it perform a defined type of work.
Source
The document, webpage, message, record or other information item from which an AI answer retrieves supporting context.
Source Citation
A visible reference linking an AI-generated statement to the source used to support it, helping users verify important information.
Speech-to-Text
AI technology that converts spoken audio into written text, enabling transcripts, voice prompts and meeting-analysis features.
Summarisation
Using AI to condense a longer document, email thread, meeting, conversation or dataset into its most important points.
Supervised Learning
A machine-learning method in which a model learns from examples paired with known target outputs or labels.
Synthetic Data
Artificially generated data designed to resemble real data, often used for testing, simulation or model development without relying solely on production records.
System Prompt
High-priority instructions supplied by the AI application to define behaviour, constraints, role and response expectations before the user's prompt is processed.
T
Task
A unit of work that Copilot or an agent is asked to complete, ranging from a simple summary to a multi-step business process.
Task Orchestration
Coordinating multiple AI, data and tool steps in the correct sequence to achieve a broader user goal.
Teams Copilot
A common term for Copilot capabilities in Microsoft Teams that help users understand meetings, chats, decisions and action items.
Temperature
A model-generation setting that influences how predictable or varied outputs are. Higher values generally increase variation where the underlying service exposes this control.
Tenant
An organisation's isolated Microsoft cloud environment containing its users, identities, licences, policies, applications and data.
Tenant Data
Information associated with an organisation's Microsoft cloud tenant, including content, configuration, identity and service data subject to applicable permissions.
Token
A small unit of text processed by a language model. Model limits and usage are often measured partly in tokens rather than words.
Tool
An external capability an agent can invoke to retrieve information, perform a calculation, create content or take an authorised action.
Tool Calling
The process by which an AI system decides to invoke an available tool and provides the required parameters for that tool to run.
Topic
A conversational concept used in agent design to organise recognised user intentions, dialogue paths or specialised areas of behaviour.
Trace
A diagnostic record showing how an agent or AI workflow processed a request, including steps, tool calls or decisions useful for troubleshooting.
Transcript
A written record of spoken conversation, commonly used by Copilot to analyse and summarise supported Teams meetings.
U
Unified Audit Log
A Microsoft Purview audit source containing activity records from supported Microsoft services for investigation, governance and compliance purposes.
Unstructured Data
Information without a fixed tabular schema, such as documents, emails, presentations, chats, images and free-form text.
Upload
Providing a file directly to an AI experience so it can be used as temporary or persistent reference material depending on the product and workspace.
Usage Report
Administrative reporting that shows adoption or activity metrics for Copilot, agents or related Microsoft 365 services.
Usage-Based Billing
A charging model where consumption is billed according to actual use, such as Copilot Credits, API activity or agent capacity rather than only a fixed user licence.
User Context
Information relevant to the current user, such as identity, role, recent activity, permissions, files and working relationships.
User Grounding
Using authorised information related to the current user to make an AI response more relevant to their work and circumstances.
User Permissions
The access rights assigned to a user that determine which files, messages, sites, systems and actions are available to them and to AI acting in their context.
V
Variable
A named value used within agent logic or workflows to store information such as user input, retrieved data or the result of a previous step.
Vector
A numerical array used to represent features or semantic meaning so AI systems can compare the similarity of content mathematically.
Vector Database
A database optimised to store and search vector representations, commonly used in semantic retrieval and RAG solutions.
Vector Search
Searching for content whose vector representation is semantically similar to the meaning of a query.
Verification
Checking an AI output against reliable sources, calculations, policies or human expertise before relying on it for important work.
Version History
A record of earlier versions of a Microsoft 365 file that can help users review changes or restore previous content.
Viva Insights
Microsoft's workplace analytics and wellbeing service, which also contributes reporting experiences used to understand collaboration and Copilot adoption.
Voice Input
Using spoken language rather than typing to provide a prompt or instruction to an AI experience where voice interaction is supported.
W
Web Data
Public or otherwise accessible information retrieved from the web and supplied as context to an AI system.
Web Grounding
Using current web search results or webpages as supporting context for an AI-generated response.
Web Search
Searching online sources to retrieve current information that can supplement the model's general knowledge and organisational context.
Whiteboard Copilot
AI assistance in Microsoft Whiteboard for supported ideation, organisation and collaborative visual-work activities.
Word Copilot
A common term for Copilot capabilities in Word used to draft, summarise, rewrite, compare and transform document content.
Work IQ
Microsoft's intelligence layer for grounding Copilot and agents in real-time organisational context across Microsoft 365 and connected business information.
Work IQ API
Microsoft interfaces that allow approved applications and agents to access Work IQ capabilities such as contextual intelligence, tools and long-running workspaces.
Work IQ Data Layer
The Work IQ layer that unifies relevant signals from files, email, meetings, chats and business systems to represent how work happens across an organisation.
Work IQ Inference Layer
The Work IQ layer that brings together models, skills and tools so agents can reason over organisational context and take permitted actions.
Work IQ MCP
Microsoft's MCP-based access to Work IQ tools, enabling supported agents to retrieve and act on Microsoft 365 work information through governed interfaces. Some capabilities may be in preview.
Work IQ Memory Layer
The Work IQ layer designed to build persistent understanding of work patterns, priorities and context for supported agent experiences.
Workflow
A defined sequence of steps that moves work from an input to an outcome, potentially combining people, applications, agents and automation.
Workflow Agent
An agent designed to carry out or coordinate process steps rather than only answer questions conversationally.
Workplace Intelligence
Contextual understanding derived from organisational data, relationships, activities and business systems to make AI assistance more relevant to work.
Workspace
A persistent environment that groups relevant content, tools, context and activity around a piece of work, project or long-running agent task.
X
XAI (Explainable AI)
Approaches that make AI decisions, predictions or outputs easier for people to understand, interpret and scrutinise.
XLSX File
The standard modern Microsoft Excel workbook format, commonly used as a source or output when Copilot analyses or creates spreadsheet content.
XML
Extensible Markup Language, a structured data format used by many applications and document standards, including components of Microsoft Office file formats.
Y
Yammer
The former name of Microsoft's enterprise social-networking service, now Microsoft Viva Engage. The older term may still appear in historical Microsoft 365 documentation and discussions.
YAML
A human-readable structured-data format frequently used for configuration files, automation definitions and developer workflows connected to AI solutions.
Z
Zero-Shot Classification
Classifying new content into categories without training the model on labelled examples for those exact categories first.
Zero-Shot Learning
A model's ability to perform a task or recognise a concept without task-specific training examples supplied during model training.
Zero-Shot Prompting
Asking a model to perform a task using instructions alone, without including worked examples in the prompt.
Zero Trust
A security model based on verifying explicitly, using least privilege and assuming breach rather than automatically trusting users or devices because of network location.
Explore related Microsoft Copilot training and resources
Develop practical AI skills with our Artificial Intelligence training and Microsoft Technical courses. You can also explore Microsoft certification pathways, read our Microsoft Copilot guide, or browse the full course catalogue.
.png){kind=small}
