Microsoft Copilot 365 Glossary A to Z — Microsoft Copilot 365

Microsoft Copilot 365 glossary — 106 definitions

Every term defined across the 26 sections of the guide, in alphabetical order.
Tip: press Ctrl + F to search for a term on the page.

"/" reference: In the Copilot prompt, / opens the recent-files picker: the chosen document becomes a source of the request.
"Explain this formula": The reflex that turns Copilot into a tutor: it details each part, you learn instead of copying.
Action extraction: "What tasks fall to me in this thread?": Copilot lists your commitments buried in the conversation.
Actions: An agent's ability to act: create a ticket, send an email, write to a CRM, launch a flow.
Advanced analysis in English: "linear regression", "segment my customers into 3 groups": Copilot generates and runs the Python for you.
Advisor, not decider: Coaching suggests but doesn't rule: a direct tone is sometimes justified — you keep the judgment.
Agent: A specialized, persistent and sometimes autonomous Copilot for a domain: a colleague already trained for a post.
Agent governance: The framework defining who creates, validates and controls agents, their sources and actions — against proliferation.
Anomaly detection: "Spot the outliers", "Are there duplicates?": an automatic first quality pass.
Assistant: The classic Copilot: you ask, it answers in the moment — a versatile temp you brief each time.
Automatic insights: Trends, highlights, PivotTables and charts suggested by Copilot in one click on a table.
Autonomy is earned through supervision: The more actions you delegate, the more you invest: minimal permissions, confirmations, logs, kill switch.
Calculated column: Copilot adds a column proper to the table, often with a modern function (LET, FILTER, XLOOKUP).
Calibrated summary: Specify the form: "in 3 sentences for an executive", "in 10 bullets", "as a table decisions/actions/owners".
Clarify live: "What are the points of disagreement?", "Have we talked about the budget?": following the thread without interrupting.
Clean data: Clear headers, no empty rows, one value per cell — the timeless rules, which Copilot makes mandatory.
Cloud condition: Only OneDrive/SharePoint files (indexed by the Graph) can be referenced — a local file can't be.
Coaching by Copilot: Before sending, an assessment of the draft on tone, clarity and the reader's sentiment, with suggestions.
Condense: "cut to the essentials", "merge the redundant slides": 40 slides become 15 punchy ones.
Connect business data: Give the agent its sources: SharePoint/OneDrive, websites, business systems — it answers on YOUR information.
Connectors: The links (Power Platform, API) connecting the agent to a system, with the description of when and how to act.
Contextual reply: On a received email, Copilot reads the thread and generates a coherent reply without re-reading the whole history.
Conversational description: "Create an agent that answers HR questions from these documents": the starting point of creation.
Copilot: Microsoft's AI assistant, available in Windows, the web (copilot.microsoft.com) and the Microsoft 365 apps.
Copilot (free): The general-purpose chatbot: web, Windows app, mobile. No link to your work files/emails.
Copilot Pages: A shared, persistent canvas where Copilot's answers become a living document, editable and shareable.
Copilot Pro: Personal subscription (~$20/month): Copilot in Word/Excel/PowerPoint/Outlook for personal accounts, priority access to recent models.
Copilot Studio: The no-code workshop where you create agents in natural language: describe, give sources, test — without coding.
Correlation ≠ causation: Copilot may flag that two columns move together; it's for you to interpret whether one explains the other.
Deck doctor: Copilot on an existing presentation: "summarize into 10 slides", "harmonize the style", "add notes".
Describe the message: Ask for the intent ("show the evolution by region") rather than the technical type ("stacked bar chart").
Draft from notes: Paste your rough notes then ask: "Turn these notes into a structured memo" — the most rewarding use.
Draft with Copilot: The prompt that appears in a blank document (or via the Copilot icon in the margin): describe the document you want, Word writes it.
Edit in Pages: The button that settles a good chat answer into a durable space, instead of losing it in the history.
Enterprise data protection: The Microsoft 365 Copilot commitment: prompts and answers not used to train models, data processed within the tenant's contractual framework.
Example (few-shot): Providing a model of the expected result ("following this template: …"): the most effective technique to get THE right format.
Form ≠ substance: Copilot improves structure and presentation, not the relevance of the content: reviewing the meaning stays yours.
Formal table: An Excel object (Ctrl+T / Insert > Table) with typed columns and named — not a mere range of cells.
Formula in natural language: Describe the calculation in plain English; Copilot writes the matching Excel formula (IF, AND, XLOOKUP…).
Free / consumer Copilot: Outside the enterprise framework: NEVER paste confidential company data into it.
Freshness: An agent plugged into outdated documents answers false with confidence: the corpus must be kept up to date.
Grounding: The mechanism that "anchors" the answer in your real data (files, emails) rather than in the model's memory alone.
Hallucination: A false answer stated with confidence — the number-one risk of generative AI: every output must be checked.
Implicit temperature: Asking for "3 options in different tones" turns the model's variability to your advantage.
Inherited permissions: The agent must respect access rights and never expose a document the user isn't allowed to see.
Instructions / knowledge / actions: An agent's three attributes: its role and rules, its dedicated sources, and what it can do.
Instructions + knowledge: Role/tone/scope on one side, documentary sources on the other: 90% of an agent's quality.
Interpretation skill: Reading an R², judging whether a model makes sense: human knowledge the tool makes necessary, not superfluous.
Iteration: Refining through successive exchanges ("shorter", "more formal", "as a table") rather than rephrasing from scratch.
Iteration in natural language: "put the regions in distinct colors", "add labels", "sort in descending order".
Keep / Regenerate / Refine: The three buttons after generation: keep, retry, or adjust with an instruction ("shorter", "add a budget section").
Language model (LLM): AI trained on vast text corpora, able to generate relevant text — but without "understanding" or guaranteeing accuracy.
Length & tone: Short/medium/long and formal/casual/direct: two settings to frame the message in two clicks.
Live catch-up: During the meeting: "What did I miss?", "Summarize the last 10 minutes" — Copilot answers privately.
Living analysis: The result lives in the workbook and recalculates with the data — not a frozen export.
Meaning-based search: "Where did we talk about supplier X?": Copilot finds the info far better than keyword search.
Meeting copilot: Copilot follows the thread for you — but doesn't replace your presence and depends on transcription being on.
Meeting recap: After a transcribed/recorded meeting, Copilot produces a summary: points, decisions, assigned actions.
Microsoft 365 Copilot: The enterprise licence (~$30/user/month, on top of Microsoft 365): AI built into the apps WITH your work data (Graph), tenant security.
Microsoft Graph: The map of your work data (emails, files, meetings, contacts): it's what makes Copilot 365 contextual.
Multi-file: Several references in one prompt: "Compare /Contract-v1 and /Contract-v2 and list the differences".
Natural-language question: "Which month dropped the most?", "Is there a correlation between discount and volume?": querying your data without a formula.
No transcript, no summary: The absolute prerequisite: without transcription (or recording) on, Copilot has no material to analyze.
One slide, one idea: The golden rule Copilot applies: spotting overloaded slides and lightening them.
Oversharing: Documents shared too broadly ("Everyone"): the number-one risk exposed by Copilot rollouts.
Pay-per-use: A cost model based on messages and actions: a popular or poorly-calibrated agent can generate a surprise bill.
Permissions respected: Copilot only sees what you have access to: never a private channel or an unauthorized message.
Persistence: An agent exists continuously, reusable by a whole team without re-briefing: created once, it serves a thousand times.
Prepare to speak: "Summarize the arguments for and against" before taking the floor: contribute rather than take notes.
Presentation from a prompt: "Make an 8-slide presentation on X": Copilot starts from a simple description — less reliable than a sourced document.
Presentation from Word: "Create a presentation from this document": Copilot turns a report into structured slides.
Prompt: Your instruction to the AI. The more precise and contextualized, the better the answer.
Python in Excel: Running Python code (pandas, matplotlib) in a cell, in the Microsoft cloud — enabled depending on the edition.
Query the meeting: "What did we decide on the budget?", "Which actions fall to me?": the meeting becomes queryable.
Queryable memory: Copilot cross-references conversations, files and messages: your Teams history is re-asked instead of lost.
Questioning the document: "What does it say about costs?", "What are the recommendations?": targeted extraction, without reading everything.
Range vs table: A range is a raw data rectangle; a table is a queryable object that Copilot knows how to analyze.
Read vs write: Reading a status is low-risk; writing or sending is risky (irreversible or costly) and needs safeguards.
Real chart object: The proposed chart inserts as a real Excel chart, editable normally afterward.
References (citations): The numbered callbacks in the answer: click to check the source passage — the anti-hallucination reflex.
Respecting permissions: Copilot only accesses content YOUR account already has rights to: it opens no new door — but it lights up poorly closed ones.
Responsible adoption: Train users, set usage rules, keep the human in the loop on the decisions that matter.
Rewrite (selection): Select a passage → Copilot icon → proposed rewrites, to keep, regenerate or refine.
Rewrite axes: Concision ("cut it in half"), tone (formal/warm/neutral), audience ("for a non-technical client"), structure ("as a bullet list").
RTCF structure: Role + Task + Context + Format: the 4 ingredients of the professional prompt.
Sensitivity labels: Microsoft Purview labels (Public, Internal, Confidential…): Copilot respects them and propagates them to generated content.
Skeleton: The breakdown into slides, titles, key points, visuals and notes — a 70% starting point, not a final deliverable.
Source scope: Connect only relevant and reliable sources: too many or contradictory sources degrade the answers.
Style from a template: "Write the minutes of /Meeting-notes in the style of /Minutes-March": substance from one source, form from another.
Summarize the thread: At the top of a long conversation: Copilot condenses decisions made, open questions and who's waiting on what.
Summarize this document: The basic request to Word's Copilot pane — on the open document, with references to source passages.
Table name: "Sales2025" rather than "Table1": a meaningful name makes the dialogue with Copilot natural.
Teams conversation summary: "Summarize what I missed in this channel": catching up a written discussion without reading everything.
Tenant: Your organization's Microsoft 365 space: your data stays there — Copilot 365 works inside this perimeter.
Test on a known case: Verify the formula on a row whose result you know: a plausible formula can be wrong.
Test pane: The built-in space to converse with the agent before publishing and fix its wrong or out-of-scope answers.
The 70/30 rule: The AI produces the material (70%); the human checks, adjusts, owns it (the decisive 30%). Never sent without review.
The accordion: Asking for several lengths of the same idea: tweet version, paragraph version, expanded version.
Three-way collaboration: Several people AND Copilot work in real time on the same Page: the AI becomes a team member.
Transform into table/list: "Turn this paragraph into a pros/cons table": the structural rewrite, very effective in a report.
Transparency & consent: Transcribing must be announced to participants — a matter of consent and often a legal obligation.
Validate the outline first: Check the slide sequence (the story) before polishing the form: reorganizing a skeleton is easy.
Validation cycle: A review (security, compliance, accuracy) before broad publication, especially for sensitive or acting agents.
Visual honesty: An axis that doesn't start at zero or a 12-slice pie misleads: it's for you to judge the visual's truthfulness.
Workspace: Pages serves to build progressively (brainstorming, plan, synthesis) — not a self-validated final deliverable.
Your voice: The email commits under your name: adjust the default smooth tone until it sounds like you, and always re-read.