How I Use AI to Speed Up Instructional Design Without Losing Quality
AI can draft outlines, scripts, scenarios and quiz items in minutes. It can also produce confident nonsense. Here is where I use AI in my instructional design workflow, where I don't, and the review rules that keep quality high.
AI is most useful in instructional design for first drafts and variations: turning source material into outlines, drafting scenario variants, generating quiz distractors, rewriting text for different reading levels and summarizing learner feedback. It's least useful for the decisions that define a course: what the business problem is, what behavior needs to change, and whether the content is accurate.
My rule of thumb is simple: AI drafts, humans decide. Every output that reaches learners gets checked by me or a subject matter expert.
Where AI fits in my workflow
Analysis: summarizing, not deciding
I use AI to summarize interview notes, cluster survey comments and pull themes from existing documents. In the Feedback Intelligence project, an AI pipeline grouped learner comments into themes and flagged recurring issues, which turned days of manual reading into a short weekly review. But choosing which problems to solve stayed a human decision.
Design: options to react to
Once the goal and key behaviors are clear (see action mapping), I'll ask AI for alternative module structures or scenario ideas. It's a fast way to get options on the table. I usually combine parts of two or three suggestions rather than taking one as-is.
Development: first drafts
This is where the time savings are largest:
- Storyboard text drafted from SME notes, then rewritten in the course's voice.
- Narration scripts adjusted for the ear: shorter sentences, contractions, natural rhythm.
- Scenario variations at different difficulty levels.
- Quiz items and feedback, reviewed carefully (more in writing quiz questions with AI).
- Job aids condensed from longer policy documents.
Evaluation: making sense of data
Summarizing open-ended survey responses and spotting patterns in assessment results are good uses, as long as someone checks the summaries against the raw data.
Where I don't use AI
- Defining the performance problem. That takes conversations with stakeholders and people doing the job.
- Final accuracy checks. AI can invent policy details, statistics and citations. Every fact gets verified against the source.
- Sensitive client content in unapproved tools. Confidential material only goes into tools the client has approved. My shadow AI guide covers the risks.
- Final voice and tone. AI drafts tend to sound generic. Courses should sound like the organization and speak to its people.
Review rules that keep quality high
- Ground every prompt in source material. Paste in the policy, process doc or SME notes rather than relying on the model's general knowledge.
- Check facts line by line against the source. Flag anything the source doesn't support.
- Read it aloud. Generic AI phrasing is easy to hear and easy to fix.
- Check for bias and representation in names, scenarios and examples.
- SME sign-off on anything technical, legal or safety-related.
Build a prompt library
The biggest long-term gain comes from saving prompts that work. I keep templates for scenario drafting, quiz generation, narration rewrites and feedback summaries, each with a slot for source material and a clear output format. My prompt writing guide explains how I structure them.
Sources
- NIST, AI Risk Management Framework
- OWASP GenAI Security Project, OWASP Top 10 for LLM Applications