Copilot Glossary

Computer systems capable of performing tasks that typically require human intelligence, such as understanding language, recognizing patterns, and generating text. Microsoft 365 Copilot uses AI to assist with writing, analysis, and content creation.

The process of introducing and integrating Copilot into an organization's workflows. Successful adoption requires infrastructure, training, community support, and clear policies. See Adoption & Governance for guidance.

Unintended preferences or assumptions in AI outputs, often reflecting patterns from training data. Can appear as inappropriate language, tone mismatches, or stereotypical assumptions. Always review outputs for bias and adjust as needed.

An enthusiastic early adopter who helps colleagues learn and troubleshoot Copilot. Champions multiply training efforts and create sustainable adoption momentum. See Adoption & Governance for champion program guidance.

Information provided in a prompt that helps Copilot understand the situation, audience, and requirements. Part of the prompting formula: Goal → Context → Expectations → Source.

Microsoft 365 Copilot is an AI-powered assistant integrated into Office apps (Word, Excel, Outlook, Teams, PowerPoint) that helps users draft, summarize, and analyze content using organizational data.

Policies and technologies that prevent sensitive data from being shared inappropriately. Copilot respects DLP policies, ensuring that protected information isn't included in outputs or shared with unauthorized users.

The unintended exposure of sensitive or confidential information. Can occur when sensitive data is included in prompts or when outputs containing sensitive information are shared inappropriately. Always review outputs before sharing.

A tool that helps create initial versions of documents, emails, or other content. Copilot is a drafting assistant—it creates drafts that require human review, editing, and verification before use.

Specifications for output format, tone, length, and structure. Part of the prompting formula: Goal → Context → Expectations → Source.

How you structure your prompt influences what the model generates. The same request can produce very different outputs depending on how it's framed. Effective framing includes role, format, tone, and scope specifications.

What you need from Copilot. Part of the prompting formula: Goal → Context → Expectations → Source. Should be clear and specific.

Connecting Copilot's responses to specific source files or data. Grounded prompts attach relevant documents, ensuring outputs are based on actual information rather than training data patterns.

When an AI model generates plausible but incorrect or fabricated information. Can occur when the model lacks sufficient context and falls back on training data patterns. Always verify outputs against source files.

A design principle where humans remain involved in decision-making processes, reviewing and verifying AI outputs. Essential for responsible AI use, especially in government contexts where accuracy and accountability matter.

The seamless connection between Copilot and Microsoft 365 apps. Copilot operates within your Microsoft 365 tenant, respecting existing security controls, permissions, and compliance frameworks. This integration provides security advantages over standalone AI services.

The process of refining Copilot outputs through follow-up prompts. Rather than accepting the first output, users can ask for revisions, additions, or format changes, treating Copilot like a collaborative conversation.

A type of AI system trained on vast amounts of text data to predict what words should come next in a sequence. Copilot uses an LLM to generate text based on patterns learned during training. LLMs are pattern-matching systems, not knowledge bases.

Your organization's dedicated Microsoft 365 environment, containing all users, data, and services. Copilot operates entirely within your tenant, ensuring data never leaves your controlled environment and respects all tenant-level security policies.

A computing system inspired by how neurons connect in the brain. LLMs use neural networks with billions of connections (weights) that encode patterns learned from training data. These patterns, not facts, determine how the model generates text.

The core mechanism by which LLMs work. Rather than retrieving facts or following rules, LLMs generate text based on statistical patterns learned from training data. This is why verification is essential—patterns can be wrong or outdated.

Access controls that determine what files and information users can access. Copilot respects the same permissions as you—it can only access files you have permission to view. This ensures security and prevents unauthorized data access.

The text instruction you provide to Copilot, describing what you want it to do. Effective prompts follow the formula: Goal → Context → Expectations → Source. Clear, specific prompts produce better results than vague requests.

The practice of crafting effective prompts to get desired outputs from AI systems. Involves structuring requests clearly, providing context, specifying expectations, and grounding prompts in source materials.

The process of verifying Copilot outputs before using them. Includes checking facts, numbers, dates, tone, completeness, and appropriateness. Essential for responsible AI use. See Mindset → QA checklist.

The phased introduction of Copilot across an organization. Typically starts with a pilot group, expands to early adopters, and gradually includes all users. Phased rollouts allow for learning, adjustment, and building confidence.

Files, documents, or threads that Copilot should use to ground its responses. Part of the prompting formula: Goal → Context → Expectations → Source. Always attach source files and name them explicitly in prompts.

The security boundary that defines your organization's Microsoft 365 environment. Copilot operates entirely within tenant boundaries, ensuring organizational data never leaves your controlled environment and respects all security policies.

The vast collection of text (books, websites, articles, etc.) used to train LLMs. The specific sources are typically not disclosed. Training data is frozen at training time—models don't automatically learn new information after training.

Confidence in Copilot's reliability and appropriateness for specific tasks. Built through understanding capabilities, using clear prompts, grounding in sources, verifying outputs, and applying domain expertise. See Trust & Safety.

The process of checking Copilot outputs against source materials to ensure accuracy, completeness, and appropriateness. Essential for all outputs, especially those with legal, financial, or public trust implications.

Numerical values in neural networks that encode patterns learned during training. Weights represent statistical correlations between words, concepts, and contexts—not facts. They determine how strongly different parts of the network connect.