Prompt Writing for Instructional Designers: A Practical Guide

A good prompt for instructional design work gives an AI model five things: a role, the context (audience, goal and source material), a specific task, constraints (length, tone, reading level, what to avoid) and an output format. Prompts with all five consistently produce drafts that need light editing. Prompts with only a task, like "write a course on data privacy," produce generic text that needs rewriting.

This guide covers the structure I use and examples for the most common instructional design tasks.

The five-part structure

1. Role

Tell the model what perspective to take. "You are an experienced instructional designer writing workplace scenarios for frontline retail staff" frames vocabulary and judgment better than no role at all.

2. Context

This is the part most people skip, and it matters most. Include:

• Who the learners are (role, experience level, language needs).
• The learning objective or behavior you're targeting.
• Source material: paste the policy, process steps or SME notes. Tell the model to use only this material.

3. Task

Be specific. "Write a branching scenario" is better than "write some content." "Write a three-decision branching scenario where the learner handles a customer asking for a refund outside the return window" is better still.

4. Constraints

Length, reading level, tone, terms to use or avoid, and anything that must or must not appear. For example: "Plain English, under 120 words per screen, no jargon, no humor about the customer."

5. Output format

Ask for the format you'll actually use: a table matching your storyboard columns, numbered steps or JSON for import into a tool.

Example: branching scenario prompt

You are an instructional designer creating workplace security training. Learners are finance staff with no technical background. Objective: verify any change to vendor bank details by calling the vendor using the number on file. Using only the process below, write a branching scenario with three decision points. At each point give three realistic choices: one best, one partly right, one risky. Write consequence feedback for every choice explaining why. Keep each screen under 100 words, plain English. Output as a table: Screen ID, Situation, Choices, Feedback, Next screen. [Paste process here]

This is close to how I drafted early versions of finance scenarios similar to those in my BEC training guide.

Example: narration rewrite prompt

Rewrite the on-screen text below as narration for a voiceover. Audience: new employees. Use second person, contractions and sentences under 20 words. Don't repeat the on-screen bullet points word for word; explain them conversationally. Target 45–60 seconds when read aloud. [Paste text here]

Techniques that improve results

• Ask for options. "Give me three different opening scenarios" lets you choose the strongest.
• Show an example. Paste one screen you've already written and ask the model to match its style.
• Iterate, don't restart. "Make choice B more plausible; right now it's obviously wrong" refines the draft.
• Ask it to flag gaps. "List any steps where the source material is unclear" catches missing information early.
• Ask for a self-check. "Check each feedback message against the source and list any claim it doesn't support."

What prompts can't fix

Even a perfect prompt won't make the model know your organization's real policies, culture or edge cases. That's why source material and human review are non-negotiable, and why sensitive content should only go into approved tools. See how I use AI in my design workflow for the review rules I follow.

Sources

• Anthropic, Prompt engineering overview
• OpenAI, Prompt engineering guide