A useful AI workflow, from brief to tested lesson
Give AI a bounded job. Keep evidence, instructional decisions and final approval in human hands.
1 · Start with approved evidence
Assemble a short design brief, a verified source pack and one target behavior. Remove personal, confidential or regulated information unless your organization explicitly permits that use in the selected service. Check the tool’s data terms and your organization’s approved configuration.
Ask AI to identify missing information, conflicting statements and questions for the subject-matter expert. Require source references. Inspect the original source for every consequential claim: a citation can be present and still fail to support the statement.
2 · Generate options with constraints
Ask for two or three practice designs for the same objective. Supply the audience, context, reading level, time limit, source boundaries and accessibility needs. Compare how much meaningful practice each option creates. Do not select an option merely because it is easy to produce.
A useful prompt: “Using only the supplied policy, draft a refund decision scenario for a new support agent. Include one correct option and two plausible errors. Explain each consequence using a source section. Mark unsupported details as NEEDS REVIEW. Do not invent policy.”
3 · Review the design before production
Check factual accuracy with an expert, alignment with the objective, realistic distractors, feedback quality, bias and accessibility. Ask whether the assessment requires the skill or merely rewards clues in the wording. AI can help generate critique questions, but it cannot certify that its own answer is correct.
Keep a small review log: draft claim, source, reviewer, correction and approval date. Label synthetic examples. If you use generated narration, check pronunciation and captions against the approved script.
4 · Test with people and revise
Observe representative learners using the prototype. Separate interface friction from difficulty with the actual task. Ask them to explain their decision, then compare their work with the rubric. Revise based on observed behavior.
Evaluate AI’s value using time saved after review, defect rate, revision effort and learner performance. Faster drafting with more factual errors may not be a net improvement. Never substitute simulated learners for all real learner testing.
These guides combine original editorial recommendations with the resources below. Tools and features can change.
ATD: AI in instructional designArticulate: AI Assistant