For the past 18 months, I have been deep in the trenches of integrating AI into our L&D workflows. I’ve seen the magic—drafts created in seconds, storyboards that actually make sense, and assessments that feel like a decent starting point. But I have also seen the catastrophes. I’ve seen AI confidently hallucinate compliance regulations, invent internal terminology that doesn't exist, and create assessment distractors that are actually arguably correct.
If you aren’t running a "gotchas" document for your AI outputs, you are playing with fire. In our industry, "AI-assisted" does not mean "AI-automated." To keep our learners safe and our stakeholders happy, we must move past the sloppy "looks good to me" feedback loop. It is time to clearly define the roles and responsibilities between the instructional designer and the subject matter expert (SME) to ensure that every asset reaching production is bulletproof.
The Risk-Based QA Framework
Not all training is created equal. Before you send a single document to an SME, you need to conduct a risk-based assessment of the content. A micro-learning module on how to use the new printer in the breakroom requires a different level of scrutiny than a compliance module on financial reporting regulations.
Low-Stakes Content: AI-drafted scripts or generic soft-skills scenarios. The ID acts as the primary validator. The SME provides a quick sign-off on tone and context.
High-Stakes Content: Product knowledge, technical training, legal/compliance, or safety protocols. Here, the SME is the primary validator for accuracy, while the ID acts as the steward of learning efficacy and pedagogical integrity.
What the Instructional Designer (ID) Must Verify
If you are an ID, your job isn’t just to check for typos. Your responsibility is to act as the gatekeeper of the learner’s experience. You are the one who knows how people learn, how to write a distractor that isn't a "gimme," and how to maintain a consistent instructional architecture.
Pedagogical Integrity
Does the AI output align with our learning objectives? Often, AI will generate content that looks like a Wikipedia entry—factual, but pedagogically hollow. It is the ID's job to strip out the "fluff" and replace it with application-based learning. Are we showing, or are we telling?
Assessment Mechanics (The "Breaker" Approach)
I treat every assessment draft like I’m trying to break the course. When an AI generates a multiple-choice question, I ask: Can I arrive at the correct answer through pure logic without knowing the material? Are the distractors plausible enough to test for deep understanding? The ID must verify that the here test questions are aligned with Bloom’s Taxonomy and free of "trick" questions that frustrate learners.
Clarity and Cognitive Load
AI has a tendency to be overly wordy or, conversely, too terse to be meaningful. I spend hours rewriting sentences just to remove ambiguity. If the AI provides a explanation that takes four lines to say what could be said in one, it’s coming out. The ID is the editor-in-chief of the learner’s cognitive load.
What the SME Must Verify
The biggest mistake in L&D is asking the SME to "review the content." This is too vague and invites the dreaded "looks good to me" response. Instead, you need to provide the SME with a specific, targeted review checklist. Their focus should be on the truth, not the presentation.
Technical Accuracy and Nuance
The SME is the ultimate authority on what is factually true within the company’s context. They must verify that the AI hasn’t hallucinated a policy or used outdated terminology. They are responsible for flagging nuances that only someone in their role would understand—the "in-the-wild" reality that AI often smooths over.
Source Validation
I require my SMEs to point to the source material for any claims the AI makes. If an AI writes: "According to company policy, we should approach the client in X manner," the SME must verify that this policy exists and that it hasn’t been superseded by a more recent memo. No citation, no approval.
Tone and Cultural Fit
AI often defaults to a weirdly formal, robotic, or overly corporate tone. SMEs are the best judges of whether the content feels like "us." Does the language align with how we actually talk to clients? Does it reflect the culture of the team?
Defining the Hand-off: Improving Review Ownership
The hand-off is where most projects lose time. If you send an AI draft to an SME without explicit instructions, you are inviting disaster. Use a structured hand-off that forces accountability.
- Version Control: Never email a Word doc. Use a collaborative platform where changes are tracked. The "Specifics" Request: Instead of "What do you think?", ask "Please confirm the accuracy of section 2.4 and clarify if the terminology used here is current for Q3." The "No" Button: Empower your SMEs to say, "This is wrong," rather than feeling pressured to edit every single line themselves.
The SME vs. Instructional Designer Responsibility Matrix
Responsibility Instructional Designer (ID) Subject Matter Expert (SME) Fact-checking Secondary (Spot checks) Primary (Full validation) Learning Objectives Primary (Alignment) Secondary (Relevance) Tone/Voice Primary (Consistency) Secondary (Cultural fit) Assessment Logic Primary (Pedagogy) Secondary (Technical accuracy) Compliance/Policy Secondary (Formatting) Primary (Accuracy)Final Thoughts: Avoiding the AI "Gotcha"
The most important takeaway for my fellow practitioners is this: AI does not have a conscience, and it does not have a memory of your internal processes. It is a probabilistic engine, not an expert. As L&D professionals, we are the human-in-the-loop.
If you are frustrated with your AI outputs, stop blaming the tool and start refining your review ownership. By clearly splitting the duties—with IDs managing the structure and pedagogy, and SMEs managing the technical truth—you create a high-functioning loop that actually saves time rather than adding a layer of frustration. Stop accepting vague feedback, stop trusting the AI Visit this page implicitly, and start testing your content like a learner who wants to catch you in a mistake. Your learners will thank you for it.