I am here with another opinion. Let’s be honest for a second.
You’ve seen the bad AI content. We all have. The eLearning module that sounds like it was written by someone who has never been inside an organization. The facilitator guide that hits every best practice checkbox and misses the actual audience entirely. The course overview paragraph that is so generic it could be about anything.
If that’s your reference point for AI, your skepticism makes complete sense. And I’m not going to try to talk you out of it.
What I want to do instead is make a distinction. Because there’s a difference between AI replacing your judgment, which is a real problem with real consequences, and AI handling the parts of your workflow that don’t require your judgment in the first place. The first is a legitimate concern. The second is just efficiency.
Here’s where that line actually falls.
Where AI earns its place in the in-house workflow
These are the tasks where AI gives you a useful starting point, a faster first draft, or a pressure-test you wouldn’t otherwise have time to run. They’re not tasks where the output goes straight into your course. They’re tasks where the output gives you something to work with. And I am only sharing text-based tasks. We could go on for days on image and code vibing tasks.
1. Drafting SME interview questions before a kickoff
You know what you need to find out. You have a vague brief and a 30-minute window with a subject matter expert who may or may not show up prepared. AI can give you a full set of focused, behavior-centered interview questions in about three minutes. You then edit them for your specific context, your specific SME, and the specific performance gap you’re trying to understand.
What used to take 30 minutes now takes 10. And your SME meeting is better for it.
2. Building out scenario variations from a single example
You’ve written one branching scenario. It’s good. It captures the situation accurately. Now you need three more variations that cover different decision points, wrong turns, and real-world consequences. That’s the part of scenario writing that’s
technically correct but genuinely tedious, and AI is solid at it when you give it your original example and the outcome parameters.
You still write the first one. You still edit all of them. But you’re not starting from scratch four times.
3. Writing first-pass sections of a facilitator guide
Facilitator guides are important. They’re also the kind of writing that takes real time without always requiring your best thinking. The section on how to debrief an activity, the timing notes, the “watch for this” callouts. AI can draft a usable first pass if you give it the activity, the learning objective, and what you want facilitators to know.
Then you make it sound like you and you can reuse these over and over again.
4. Reviewing your own content for learning design gaps
This one surprises people. Paste your learning objectives and the content you’ve built around them and ask AI to identify gaps, misalignments, or places where the design doesn’t match the performance goal. It will find things. Not always things you agree with, but things worth considering.
It’s the closest thing to a peer review you can get at 9pm before a deadline.
5. Turning compliance text into readable learning objectives
You know the document. Legal, Compliance, or HR sends over forty pages of policy language and asks you to ‘turn it into a training.’ AI can extract the core behavioral requirements from dense regulatory text faster than you can and give you a starting list of objectives to work from.
You still must decide which of those behaviors matter, which ones can be addressed through job aids instead of training, and which ones represent a compliance requirement that doesn’t actually change how anyone does their job. That part is yours.
The pattern: In every case, AI is handling the part that’s technically correct but time-consuming. You’re still making the decisions that require knowing your learners, your organization, and the actual performance problem. That’s not AI replacing you. That’s AI giving you more time to do the work that actually requires you.
Where it still needs you. Every time.
Here’s what AI cannot do, regardless of how good the prompt is.
Know your learners
AI doesn’t know that your manufacturing floor audience reads at a sixth-grade level and will disengage immediately if the language sounds too corporate. It doesn’t know that your sales team is experienced and impatient and will click through anything that doesn’t get to the point in the first 30 seconds. It doesn’t know that your new hire cohort is anxious and needs psychological safety built into the onboarding experience before they’ll engage honestly with reflection activities.
You know that. That knowledge changes everything about how you design.
Identify the actual performance problem
The request that comes in is rarely the problem. ‘We need a training on communication’ is a request. The actual problem might be that managers aren’t having hard conversations early enough, or that the performance review process is so complicated nobody understands what they’re being evaluated on, or that one team has a culture of silence that training won’t fix.
Diagnosing the real problem, before you build anything, is the most strategically valuable thing you do. AI can help you ask better questions. It cannot do the analysis for you.
Push back when training isn’t the answer
This is the part of the work that makes you a strategic partner instead of an order taker. AI won’t tell your stakeholder that what they’re asking for won’t work. You have to do that. And you need the language, the evidence, and the professional confidence to make that case effectively.
Edit AI outputs so they don’t sound like AI outputs
This is more important than it sounds. AI writing has a recognizable voice. It’s smooth. It’s slightly flat. It hedges in specific ways and uses certain phrases that appear over and over. Your audience may not consciously identify it, but they’ll feel the difference between content that sounds like a real practitioner wrote it and content that sounds generated.
Your job isn’t just to check AI’s work for accuracy. It’s to make it yours.
Straight talk: The skepticism you have about AI content quality is correct. Bad AI content exists because people skip the editing step. They treat the first output as the final output. That’s not how this works, and it’s not how any practitioner who cares about quality should be using it.
The frame that makes this useful
AI doesn’t replace good instructional design. It amplifies it, but only if you understand the fundamentals first.
If you know what a well-constructed learning objective looks like, you can tell when AI has given you one. If you know what good scenario design requires, you can edit AI’s variations toward it. If you understand the difference between a training problem and a performance problem, you can use AI to help you make that case to a stakeholder.
If you don’t have that foundation, AI will give you content that looks right but isn’t. And you won’t catch it.
That’s the honest version of this conversation. AI is a capable collaborator for people who already know what they’re doing. It’s not a shortcut around the expertise. It’s a multiplier of it. Come learn with us in #IgniteLearning this month. Our VIP workshop this month is “Build It with Me: How I Use Claude from Design to Rise Development” and our free template is “Real Work, Real Prompts: An AI Conversation Guide for In-House Instructional Designers.”