AI can help instructional designers work through large amounts of source material and create stronger first drafts. It cannot decide what people truly need to learn, whether a process is correct or how learning will transfer to the job.
Start with a performance problem worth solving
Before prompting any tool, meet with the stakeholders who own the work. Ask what people should do differently, where mistakes occur and what evidence would demonstrate success. A polished AI-generated module still misses the mark if it teaches information that employees can already find in a job aid.
Translate the business request into a practical design brief: audience, task, decisions, source-of-truth documents, constraints, accessibility requirements, completion expectations and review ownership. This brief is the guardrail for every AI-assisted step that follows.
Step 1: Audit and map the existing content
AI can be useful for identifying headings, summarizing recurring themes and locating possible duplication across existing training materials. Ask it to produce a content inventory that points back to the original module, unit, heading and source passage. That traceability matters more than an impressive-looking summary.
Separate knowledge that should live in a job aid or searchable resource from decisions that benefit from practice. A click-by-click process that changes every few months usually needs a maintainable reference; an ambiguous judgment call may deserve scenario-based learning.
Human review: Have a subject-matter expert verify every process, exception and reference. The model may combine similar procedures that should remain separate.
Step 2: Draft learning objectives and scenarios
Once the content is verified, use AI to brainstorm measurable learning objectives and realistic scenario options. Ask for the decisions a learner must make, the information needed to make them and the likely consequences of each response. Then revise the outputs against your actual workflow and policy.
Good prompts establish audience and context. For example: “Using only the approved process notes below, draft three customer-service scenarios. For each, identify the decision point, plausible options, consequence-based feedback and the exact source passage supporting the correct action. Flag anything the source does not resolve.”
AI drafts are raw material. They are not approved policy, evidence or ready-to-publish assessment items.
Step 3: Build a repeatable design system
One of the strongest uses of AI is accelerating routine production work without compromising consistency. Create templates for storyboards, assessments, feedback, alternative text drafts and review checklists. Define naming conventions, tone and visual standards before a team begins generating content.
For Articulate Rise 360, a repeatable lesson pattern might combine a concise explanation, a worked example, a learner decision and targeted feedback. For Storyline 360, prototype the key interaction and document the variables, triggers, states and accessible focus order. These decisions should be made by the learning designer, not left to the generator.
Step 4: Verify facts, privacy and permissions
Every AI-assisted production workflow needs a formal verification pass. Check that the content matches the latest approved materials, that terminology is consistent and that examples don't add unsupported rules. Watch for invented citations, incorrect product behaviour and subtle meaning changes in rewritten procedures.
Do not paste confidential customer, employee or client information into an AI system without permission and the correct organizational controls. Confirm how the service handles uploaded content, retention, licensing and access. When uncertain, use sanitized examples or an approved internal system.
Record who reviewed critical content and what version of the source was approved. Clear governance protects both the learner and the client.
Step 5: Test the experience with real learners
AI-generated narration, images and quizzes still need accessibility and usability review. Check captions, transcript quality, alternative text, keyboard access, logical focus order, contrast and small-screen layouts. A technically valid course can still be confusing if its questions feel artificial or its feedback lacks context.
Invite a few representative learners to attempt a task without coaching. Observe where they hesitate. Replace unnecessary interactions with simpler explanations when that improves clarity. Use feedback to revise the design rather than treating the first generated output as finished.
Step 6: Measure and maintain
Define what success means before launch: a better decision, faster task completion, more consistent adherence to a process or fewer avoidable questions. Use LMS results as one input, not the only measure. Gather qualitative feedback from learners and business partners, then revisit the course when policy or technology changes.
Keep the editable source, prompt notes, review log and approval decisions together. That makes maintenance possible when content owners change or when AI tools produce different results later.
The human-led AI checklist
- Business problem and learner decisions are defined before prompting.
- Every source-based statement can be traced to approved material.
- Processes and exceptions are reviewed by an accountable expert.
- Sensitive data is handled within authorized tools and policies.
- Accessibility and real-device testing are included in QA.
- Learning success is measured against a job task, not just production speed.
The opportunity is to spend less time creating repetitive first drafts and more time on the work that needs expertise: understanding the audience, shaping effective practice, verifying accuracy and designing for real-world performance.