Effective Prompting: Copilot Guide

60-minute session for City of Lethbridge managers

Focus: how to ask Copilot clearly and how to verify what it returns.

Session goals
  • Show Copilot as an in-app assistant across Word, Excel, Outlook, and Teams.
  • Teach a repeatable prompt pattern: Goal → Context → Expectations → Source.
  • Practice quality checks so every draft is verified before sharing.

Agenda at a glance (60 minutes)

Prompting fundamentals

Think of Copilot like a capable assistant. Clear, plain language wins. Use the four-part prompt frame:

Effective prompting is both an art and a science. Research in natural language processing and human-computer interaction has shown that the way users structure their requests significantly impacts the quality of AI-generated outputs. Studies on prompt engineering demonstrate that prompts that provide clear context, specific instructions, and relevant examples produce more accurate and useful results than vague or ambiguous requests (Reynolds & McDonell, 2021). This is particularly important in professional contexts where precision and accuracy matter.

For municipal government work, effective prompting becomes even more critical because outputs often have legal, financial, or public trust implications. Research on AI-assisted writing in professional contexts shows that users who invest time in learning effective prompting techniques achieve better outcomes and develop more trust in AI tools (Zhang et al., 2023). The four-part framework—Goal, Context, Expectations, and Source—provides a structured approach that helps users create prompts that produce reliable, relevant outputs.

References: Reynolds, L., & McDonell, K. (2021). Prompt Programming for Large Language Models: Beyond the Few-Shot Paradigm. CHI '21 Extended Abstracts. Zhang, Y., et al. (2023). Evaluating Verifiability in Generative Search Engines. arXiv preprint arXiv:2304.09848.

  • Goal: What you need (e.g., “Draft an invite to department heads”).
  • Context: Audience, subject, constraints (e.g., “Parks managers, workshop on new budgeting software”).
  • Expectations: Tone, length, format (e.g., “One paragraph, friendly, include date/location”).
  • Source: Files or threads to ground the answer (e.g., “Use the Q3 budget workbook and last week’s thread”).

Prompt best practices

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].”

Quality assurance checklist

Copilot drafts; you edit. Use this quick QA pass before sending or publishing.

Quality assurance is essential when using AI-generated content, particularly in government contexts where accuracy and appropriateness directly impact public trust and organizational credibility. Research on AI-assisted writing shows that systematic verification processes significantly reduce errors and inappropriate outputs (Bansal et al., 2021). The QA checklist serves as a structured approach to verification, ensuring that multiple dimensions of quality are addressed: factual accuracy, completeness, tone appropriateness, and policy compliance.

Studies on human-AI collaboration emphasize that verification should be treated as an integral part of the workflow, not an optional step. Research shows that users who develop systematic verification habits catch significantly more errors than those who review outputs casually (Zhang et al., 2023). For municipal governments, this verification process also serves compliance and risk management functions, ensuring that outputs meet legal requirements, accessibility standards, and organizational policies.

References: Bansal, G., et al. (2021). Updates in Human-AI Teams: Understanding and Addressing the Performance/Compatibility Tradeoff. AAAI '21. Zhang, Y., et al. (2023). Evaluating Verifiability in Generative Search Engines. arXiv preprint arXiv:2304.09848.

Need the full guardrails? See Mindset → QA checklist.

Spotting hallucinations
  • Overly specific facts you never provided (dates, amounts, quotes) are red flags.
  • Summaries without access to the source document are likely guesses—double-check.
  • If unsure, regenerate or split the task into smaller prompts and compare answers.

Live demo scenarios (use during the session)

1) Summarize a council report (Word/Teams)

Prompt: “Summarize the key decisions and action items from the April 5 council meeting transcript on downtown housing. Keep it to one page, highlight deadlines and budget impacts, use neutral report tone.”

QA: Verify budget numbers and deadlines against the source transcript.

2) Draft a resident email about a road closure (Outlook)

Prompt: “Draft a friendly, informative email to residents about the Main Street closure next Saturday for the City Marathon. Include time, detours, transit options, and a contact number. Keep it concise.”

QA: Check date/time, contact info, and that tone is courteous.

3) Budget trends in Excel

Prompt: “In this budget workbook, summarize year-to-date variances, highlight top three over-budget lines, and propose one action per line. Return bullets plus a simple table.”

QA: Confirm calculations match the sheet; rerun with “show the formulas you used.”

4) Meeting recap in Teams

Prompt: “List decisions, owners, and due dates from today’s Parks leadership meeting. Add open questions. Keep bullets under 7 words each.”

QA: Ensure owners/dates match what was actually agreed; adjust tone before sharing.

Practice prompts for participants
  • “Rewrite this policy summary for residents in two paragraphs and plain language.”
  • “Create three options for a community outreach plan; include pros/cons for each.”
  • “Turn these meeting notes into an action list with owner and due date columns.”
  • “Draft a press release about our new playground opening; add one quote from the mayor.”
Based on the "Microsoft 365 Copilot Lunch-and-Learn: Effective Prompting & Quality Assurance" session.
See the original DOCX in this folder for the full slide content.
Related resources

← Back to Index