Copilot as a design partner
The fastest way to improve your prompting is to understand what Copilot is doing when it responds. It is not searching for information — it is generating the most statistically plausible next text given what you wrote. That is a useful description, because it tells you what the prompt needs to contain.
If your prompt is vague, Copilot generates plausible text for that vague space. If your prompt is specific — audience, objective, format, tone, constraints — Copilot generates plausible text for that specific space. The specificity of the input governs the usefulness of the output.
By the end of this unit- Construct a structured Copilot prompt for each of the four ID prompt types.
- Apply the five-element prompt anatomy to a real ID task.
- Identify the revision a piece of AI-generated ID content most needs before it is used.
The prompting mindset shift
Most people begin by prompting Copilot the way they would search a database: short, keyword-based queries. "Write a learning objective about data security." This produces generic output because the brief is generic.
The prompting mindset for ID is closer to briefing a writer: you tell them the audience, the objective, the format, the tone, the constraints, and the context. Every missing element is a gap the AI fills with a generic average.
The anatomy of an ID prompt
A well-structured ID prompt has five elements. Each element closes a gap that would otherwise be filled with a generic average.
Role and context
Who are you, and what is the design project? "I am an instructional designer building a 45-minute onboarding module for new Copilot users in a financial services organisation." This sets the frame for every word that follows.
Audience
Who is the learner? Be specific. "My audience is mid-level account managers with 3–8 years of experience. They are not technical users. They are comfortable with Teams and Outlook but have not used Copilot before." Generic audiences produce generic examples.
Task and objective
What do you want Copilot to produce, and what learning objective should it serve? "Write a 200-word scenario-based opening for a section on using Copilot in Teams meetings. The learning objective is: the learner will identify three types of meeting task where Copilot can save preparation time."
Format and length
What should the output look like? "Write it in trainer-talk voice — direct, second person, no bullet points in this section. Under 200 words. No jargon." Without format constraints, AI defaults to bullet-heavy, formal text.
Constraints and exclusions
What should Copilot not do? "Do not mention specific Copilot features by product name — the feature set changes frequently. Do not use the word 'leverage'. Do not open with a question." Exclusions prevent the most predictable problems before they appear.
The four ID prompt types
Different ID tasks call for different prompt strategies. These four types cover the majority of generative work in a standard ID project.
Type 1 — Outline generation
Use for: Getting a section-by-section structure from an objective and audience description.
Key inputs: The learning objective, the audience description, the total time available, and the format (linear or branching).
What to watch for in the output: Generic sections that describe the topic rather than teach the objective. Sections that are missequenced (concept before context, not context before concept). Sections that cover adjacent topics rather than the core one.
Example prompt fragment: "Generate a five-section outline for a 30-minute module. Objective: the learner will select the right Copilot prompt strategy for three types of writing task. Audience: marketing coordinators, no prior Copilot experience. Include: a worked example section, a practice section, and a short review. Do not include a general introduction to AI."
Type 2 — Screen script generation
Use for: Turning a storyboard section description into screen-ready instructional text.
Key inputs: The section objective, the audience, the block type (text, accordion, process), the word limit, and the tone.
What to watch for: Bullet lists when you asked for prose. Passive voice. Hedging language ("it is important to note that…"). Generic examples not grounded in the learner's actual context.
Example prompt fragment: "Write a 150-word trainer-talk explanation of why context matters in a Copilot prompt. Voice: direct, second person, active. Audience: HR managers. Include one specific example from an HR context — performance review season. No bullet points. No jargon."
Type 3 — Knowledge check generation
Use for: Generating multiple-choice questions that test the section objective.
Key inputs: The learning objective, the content the question tests, the cognitive level (recall, application, evaluation), and the number of plausible distractors.
What to watch for: Questions that can be answered by re-reading the screen. Distractors that are obviously wrong. Questions that test topic familiarity rather than the objective.
Example prompt fragment: "Write a multiple-choice question that tests this objective at application level: 'Identify the context element missing from a given Copilot prompt.' Provide one correct answer and three plausible distractors. Each option under 20 words. The question should require the learner to apply the principle, not recall a definition."
Type 4 — Scenario and case generation
Use for: Generating realistic worked examples and branching scenarios grounded in a specific role context.
Key inputs: The role, the task, the AI tool involved, the setting, and what a good outcome looks like versus a poor one.
What to watch for: Scenarios set in obviously fictional or generic organisations. Scenarios where the right choice is too obvious. Characters without constraints (the fictional persona has unlimited time and perfect information).
Example prompt fragment: "Write a 200-word scenario in which a project manager at a construction firm uses Copilot to prepare for a stakeholder meeting. Show them making one strong decision about how to prompt (good context, clear output specification) and one weak one (too vague, no constraints). Do not resolve the scenario — end it at the decision point."
Keep a personal library of prompts that worked. The context-setting opening of a successful prompt is often reusable across projects. Build a template for each of the four types and refine it with every project.
Review, refine, own
AI output is a starting point. The review step is not optional, and the ownership step is not theoretical — your name will be on this course, not Copilot's.
The editorial review checklist for AI-generated ID content
Does it teach the objective?
Read the section against the objective you set. If the content explains the topic but never builds toward the observable skill, it fails the first test.
Is the example real?
Generic examples — "a team member in a large organisation" — do not build transfer. Replace any non-specific example with one from the actual role context.
Is the voice right?
AI defaults to a slightly formal, hedging register. Check for passive voice, hedging phrases ("it is often the case that"), and third-person constructions. Rewrite in second person, active voice.
Is the length right?
AI will almost always produce more words than you need. Cut. The working maximum for a prose paragraph in a workplace learning screen is 45 words.
Is the knowledge check doing real retrieval work?
Read the question, then look at the screen it relates to. If the answer is visible on the screen, rewrite the question so it requires application or analysis of already-learned content.
An ID prompt produces this output: "AI tools can help teams work more effectively by automating routine tasks and providing intelligent suggestions." What is the most important revision this sentence needs?
End of Unit 3
You should now be able to:
- Build a structured five-element prompt for any ID task.
- Select the right prompt type for outline, script, knowledge check, and scenario generation.
- Apply an editorial review checklist to AI-generated instructional content before using it.
Unit review
Which element of the five-part prompt anatomy is most commonly missing from first-attempt prompts?
You need to generate a 200-word worked example for a Copilot prompting module. Which prompt type applies?
What is the most reliable sign that a knowledge check question needs rewriting?
What does it mean to "own" AI-generated content?
End of module
You have completed Course 03: Prompting for Learning Design. Next: Storyboarding with Copilot — using AI to accelerate the planning stage without losing the design rigour that makes the build stage work.