Microsoft 365 Copilot

Complete Guide

City of Lethbridge Training Resources

Key Questions:

How do I craft an effective prompt for Copilot?
How do I ensure the quality of the information that Copilot generates for me?

Overview

Microsoft 365 Copilot is an AI-powered assistant that helps you draft, summarize, and analyze content across Word, Excel, Outlook, Teams, and PowerPoint.

Key Points

Works inside your Microsoft 365 apps
Uses your organization's data and files
Respects permissions and security boundaries
Requires human oversight and verification

Remember: Copilot drafts; you decide. Always verify outputs before sharing.

What is Copilot?

What it does

Summarizes long threads and meetings
Drafts first versions of documents
Highlights trends in spreadsheets
Keeps tone consistent
Organizes information into tables

What it's not

Not an autopilot
Not a fact-checker
Not a replacement for judgment
Not always accurate
Not a live database

                Think of Copilot as: A capable junior colleague who works 24/7 and never gets tired of revisions—but you still need to review their work.
            

Copilot Mindset

Copilot is a drafting assistant, not an autopilot

Human oversight stays in charge

Works inside Word, Excel, Outlook, Teams

Best results come from clear intent + context + expectations + sources

Session Basics

Copilot drafts; you edit and verify
Always ground prompts in your own files
Specify audience, tone, and format upfront
Iterate and refine outputs

The Prompting Formula

Goal → Context → Expectations → Source

Goal

What you need

"Draft an invite"

Context

Audience, topic, constraints

"Parks managers, workshop"

Expectations

Tone, length, format

"One paragraph, friendly"

Source

File/thread names to ground answers

"Use the Q3 budget workbook"

Good vs. Vague Prompts

❌ Vague Prompt

Write about the road closure

Problem: No context, no format, no source files

✅ Good Prompt

Draft a two-paragraph resident update on the Main Street closure this Saturday 7am–6pm. Include detours, transit options, and a contact number. Friendly, concise tone.

Clear details: audience, format, content, tone

Effective Prompting Best Practices

Be specific and positive: Say what to do, not just what to avoid
Order matters: Describe the task first; mention files at the end
Show the format: Tell Copilot the structure ("3 bullets with rationale")
Iterate: Treat it like a conversation—"Shorten to 150 words"

Quick Prompt Template"Draft a [format] for [audience] about [topic]. Include [must-have details], use a [tone] tone, and keep it to [length]. Base it on [file/thread name]."

Before You Run a Prompt

Scope

Use work files/threads you can verify; avoid sensitive/PII unless policy allows.

Sources

Attach the right docs so answers are grounded; name them in the prompt.

Audience

Set tone, length, and format (bullets/table) up front.

Quality Assurance Checklist

Always verify before sharing:

✓ Verify names, dates, and amounts against the source file
✓ Check tone for residents vs. council vs. colleagues
✓ Confirm every required section landed; add missing details yourself
✓ Remove sensitive data; regenerate if the answer feels speculative

Remember: Copilot drafts; you decide. Every output is a draft until verified.

Common Workflows: Outlook

Resident Notice

Draft a friendly notice to residents about the Main Street closure on [DATE]. Include detours, transit options, and a contact number. Keep to 150 words and use a courteous tone.

Meeting Follow-up

Write a follow-up email to the Parks leadership team summarizing today's meeting decisions. Bullet owners and due dates, and add a short "next steps" paragraph. Keep concise and professional.

Common Workflows: Word

Policy Summary

Rewrite this policy section for residents in two short paragraphs, plain language, and include one call-to-action at the end. Preserve all key requirements.

Council Briefing

Create a one-page briefing for council on [TOPIC]. Include background, current status, risks, budget implications, and a recommendation. Use a formal tone.

Common Workflows: Excel

Budget Variance

In this budget workbook, summarize year-to-date variances, highlight the top 3 over-budget lines, and propose one corrective action per line. Return bullets and a small table.

Data Cleanup

Scan this table for likely data quality issues (duplicates, missing values, odd formats). List fixes and, where possible, generate cleaned columns.

Common Workflows: Teams

Meeting Recap

Summarize today's Teams meeting. List decisions, owners, and due dates; add open questions. Keep bullets under 8 words each.

Pre-read Outline

From this agenda and attached docs, create a pre-read outline for the committee. Highlight the 3 choices to be made and what inputs members should review.

Trust & Safety: Common Risks

Hallucinations

Copilot can generate plausible but wrong facts. Always attach source files and verify.

Numerical Errors

Double-check all numbers and calculations. Verify against source spreadsheets.

Data Leakage

Avoid including sensitive data in prompts. Review outputs before sharing.

                Human-in-the-loop: Copilot drafts; you decide. Keep humans involved in critical decisions.
            

Understanding How LLMs Work

Key Concept: Pattern Matching, Not Fact Retrieval

LLMs are neural networks trained to predict what words should come next. They use statistical patterns learned from training data, not verified facts.

Training data: Unknown sources, frozen at training time, quality varies
Weights: Statistical correlations, not facts—represent patterns, not knowledge
Non-probabilistic: Pattern-based generation, not rule-based logic
No source attribution: Can't cite where information came from

                Why this matters: Always ground prompts in your own files. The model doesn't "know" facts—it matches patterns.
            

How Framing Works

Framing refers to how you structure your prompt influences what the model generates.

Poor Frame

Tell me about the budget

Generic information, possibly from training data

Better Frame

Summarize the Q3 budget variance report attached, focusing on departments over 10% variance

Specific analysis of your actual document

Framing techniques: Role framing, format framing, tone framing, scope framing, constraint framing

Adoption Strategy

Infrastructure

Licenses
MFA enabled
DLP policies
Permissions

UX & Discoverability

Quick-start guides
Prompt library
In-app tips

Communities

Champions
Office hours
Teams channel

Policy & Trust: Set clear expectations, boundaries, and verification processes

Progressive Skill Building

Level 1: Low-risk

Summarizing meeting notes
Drafting internal emails
Creating outlines

Level 2: Medium-risk

Drafting public content
Analyzing spreadsheets
Creating reports

Level 3: Higher-risk

Drafting council materials
Budget analyses
Legal/regulatory documents

Always involves human review

Key Takeaways

Copilot is a drafting assistant—human oversight is essential

Always ground prompts in your own files

Use the formula: Goal → Context → Expectations → Source

Verify all outputs before sharing

Start with low-risk tasks and build confidence

Iterate and refine—treat it like a conversation

                Remember: Trust in Copilot = (Understanding) × (Clear prompts) × (Source grounding) × (Verification) × (Your expertise)
            

Next Steps

Review the detailed guides on each topic
Practice with low-risk tasks first
Join the Copilot community channel
Attend office hours for questions
Share success stories and tips

Questions? Check the full guides or contact your Copilot champions