AI course builders can produce a full course in minutes. But skipping the setup wizard is not the same as skipping instructional design, and too many teams are learning that the hard way. In this tactical session, Traci Synatschk shares what happened when she fed the same prompt to four AI course builders, scored the outputs against an ID quality framework, and used AI to build reportable, xAPI-tracked simulations. You'll leave with a repeatable workflow that puts human judgment at three specific checkpoints: before the prompt, during curation, and before publish.
Event: Human-Led AI for Instructional Designers
Speaker: Traci Synatschk
Host: Luis Malbas
Luis Malbas:Alright. Welcome back, everybody. Thank you so much for being here for our virtual event, Human-Led AI for Instructional Designers, hosted by The Training, Learning, and Development Community. Let me make sure that we have folks coming into the session. Yes, very nice. You can just type a hello in chat; I want to make sure that's working okay. And maybe where you're logging in from—I love seeing that as well.
Jennifer, I think you were the first one last time, posting hi from Illinois. Laura's here from Arizona. Nice to see you, Laura. Carrie from Ottawa. Tanya from Atlanta. Excellent. Hi, Maria in Oakland! I'm over in Sebastopol in Sonoma County, so not too far away.
Alright, well, we're gonna get started. I'm happy to introduce you to Traci Synatschk, the founder of MentoringU, bringing over twenty years of L&D experience to the stage. Traci specializes in building scenario-based simulations, treating AI as a powerful build partner rather than a replacement for human design thinking. From creating AI-coded interactions for foster parent courses to developing the LMS Genie Suite for migrations, she knows how to make technology work for the learner. She's here today to share a core philosophy: AI can speed up the build, but humans still have to decide what problem we're actually solving. Everybody, if you can, click that emoji with some hand clapping there and welcome, Traci! I'm gonna hide my screen from the interface and let you take over.
Traci Synatschk:Sounds great. Hi, everybody. It's good to be here today. Super excited to be a part of this whole environment and virtual conference. This is obviously a passion project for me, so you'll hear me get really excited as we go through. I also love Q&A, so drop them in the chat.
As was mentioned, my name is Traci Synatschk. I'm the founder of MentoringU. It's a woman-owned instructional design and LMS consultancy. I've been in the industry for about 24 years, both in corporate work and as a freelancer running my own business for the last ten.
A heads-up on what this session is and what it isn't. I know everybody has big feelings about AI. This is not a hype talk. I'm not here to convince you about AI, and it's not an anti-AI talk either. I use AI as a tool in my toolbelt, and I think that is the best way to look at it. If we can leverage it to do what we do better, that's where I want to see things land. Today is a working playbook. I want to show you how I use AI, how it fits into my workflows, and then switch gears to talk about the fun part: what I do with the time that these tools save me to elevate the learning I build.
Let's start with a little poll. I'm curious how you guys are using AI course builders right now—tools like Rise course builder, Mindsmith, etc. Are you using them? Are you interested in them? Is this new to you? If you don't know what a course builder is, drop that in the chat, and we'll talk about it too.
Alright, we have a lot of people who are not using them yet. I think that is a really great place to be right now. We're gonna talk about the pros and cons today, so hopefully, you'll leave with some good insight. For those drafting full courses with them, I'm gonna show you what I do when I do those same tasks, and how I validate to keep it human-centric.
Here is the whole talk in one line: Skipping the instructional design wizard in these course builders is not the same as skipping the instructional design.
When you log into Rise and it prompts you to load a source document or asks for your learning objectives, are you hitting the skip button? We, as an industry, have always been pushed to build more content faster with very few resources. I've always said L&D teams are tiny but mighty. So when we see AI course builders, they lead with how efficient they are: "This is gonna save you time and build content faster."
If you're clicking "skip" and jumping past the wizard, it's not necessarily a bad thing if the instructional design has already been done. Pre-AI, we met with SMEs, did needs assessments, looked at learning objectives, and storyboarded. A lot of work went into development before we actually started building the course.
We need to pay special attention to this: is that instructional design still happening before we hit "generate course"? Are we still having those conversations with SMEs? Are we pushing back when someone asks for a training that really just needs to be a job aid? If those conversations are happening, skipping the wizard is fine. If they aren't, that wizard is there to push you to think about those things.
AI can generate a polished course in minutes. But if we step back and look at how well it built the course, that's when the conversation of human-led AI surfaces. We really need a human in the driver's seat at key points to make sure we're meeting our goals.
I use tools like Rise and CreateLMS to create drafts of content quickly. But how do I make sure that content is the same quality as if I built it from scratch? I use rubrics and measurements. I took a course brief—a seven-page document covering the structure, tone, and content—and fed it into four different AI course builders. I didn't tweak anything. I let them build.
Then, I exported those as SCORM files and loaded them into a tool I built called IDA (Instructional Design Analyzer), which acts like a rubric for cognitive load, transfer of knowledge, chunking, and accessibility. Nobody was perfect. Two tools came close to being ready with small tweaks, and a couple came in lower.
What this meant was no AI tool was good to go without human interaction. They all needed me to address chunking or scenarios, and none of them did well with alt-text for accessibility. The tools were good at structure and following the outline, but weak at the language required to actually change behavior. None of the four fully stated the course-level problem or outcome.
This is where we need human intervention. Where do we step in to be most efficient?
First: Before the Prompt. I spend a good bit of time on my course brief. I answer two or three questions before AI writes a single word: What problem are we solving? Who is this for? What should a learner be able to do when they finish? Use clear, observable verbs. For example, instead of "understand empathy," use "calm an upset coworker using our three-step response." I paste those answers into every prompt as a guardrail.
Second: During Curation. AI is great at volume and variation, but you must own your judgment. Read the draft. AI will never underwrite; it will give you too much content. Often, your job becomes figuring out what to cut to avoid cognitive overload. Make sure the tone hits the mark and that scenarios feel real, not AI-generated.
Third: Before Publish. I am a huge fan of scoring a course before it ships using a QA framework. Check outcomes (are objectives observable?), alignment (does every activity map to an objective?), practice (are learners making decisions or just clicking next?), cognitive load, accuracy, and accessibility (alt-text, captions). If you give AI the reins to draft it, you must objectively review the output.
When you use a course builder with a good process, you save a lot of time generating that draft. I reinvest that time to elevate the learning experience. This is where custom simulations come into play.
I build custom xAPI simulations. For example, a client wanted an SOP refresher training on why tasks weren't getting done over the weekend (the "weekend gap"). Instead of just building a course, I built a 3D, librarian-style simulation in Claude. Learners investigate clues, interview stakeholders, determine the real problem, and prototype solutions.
What's really cool is the xAPI backend. It tracks everything. Did they interview more people? Did they skip steps? Did they assume the answer without investigating? I can sit back and see where the actual gaps are with my learners based on real data, not just a pass/fail completion score.
I built this simulation using Claude Pro. I am not a coder. I do a lot of prep before I start. I use Claude's "plan mode" first, mapping out the scenarios, then iterate until it's what I want. AI did the heavy lifting on the HTML code, but I edited all the feedback, the language, and the scenarios to make sure they made sense. Getting the xAPI to tie back to a Google Sheets dashboard was the best part, giving me line-by-line data on what learners were actually doing.
AI will generate a first draft that looks fine but teaches nothing (e.g., "Empathy is important. True or False?"). We have to step in and apply our instructional design framework. Create a draft, spend the 15 minutes to let AI build it, but be intentional about reiterating, QAing, and keeping yourself centric to the process.
I have a kit with the pre-prompt template and QA scorecard. I'll drop the link in the chat now, and then let's get to questions.
Luis Malbas:That was amazing, Traci. Wow. So much stuff. Let's see, Nicole's asking for the link.
Traci Synatschk:Yep, I am grabbing a Google Drive link. The whole kit is also loaded in my downloads on my Thinkific site, which I can share too.
Luis Malbas:Awesome. Here's a question from Ummer: "Since you're building with AI, does this mean you understand coding?"
Traci Synatschk:I try to be very mindful that I am not a coder; I come to this with just enough knowledge to be dangerous. I am careful about what I code. I use a lot of Google Apps Scripts and trust Claude to write code for me, but sometimes I take that code into Gemini or Copilot to ask if there are bugs. Mostly, I am building SCORM or xAPI files that I test in SCORM Cloud and my LMS. I am not a coder; I guess we're calling it a "viv-coder" now.
Luis Malbas:And just to confirm, you're using Claude Pro, not Claude Code?
Traci Synatschk:I use Claude Pro for the design feature to build these. I did upgrade to the $100/month plan eventually just because of the volume of simulations I am building for clients, but I built several on the $20 plan beforehand. My best advice is to have your plan mapped out first and use Claude's plan mode before having it write the HTML, which eats up tokens.
Luis Malbas:Can you tell us a little bit more about MentoringU?
Traci Synatschk:I was in corporate L&D right out of college, got my master's, and then left to start a freelance business about ten years ago. MentoringU creates content for nonprofits, small businesses, and global companies. I also specialize in LMS migrations, using Claude to develop API interactions that allow me to move content quickly. For anyone interested in that, Coursera has amazing courses specifically on API interactions so you don't have to go back to school for it.
Luis Malbas:Amira has a question here: "Is the dashboard connected with an API key meaning Claude is gathering the info, or is it done automatically?"
Traci Synatschk:Great question. API keys are the keys to the castle; you do not want them public. This dashboard does not actually have an API key. I created a Google Apps Script project with a public web app enabled. Once I finalize the xAPI file, I add one line of code in the HTML with that web app URL. It allows the xAPI to send those results to the dashboard. It is not connected to Claude at all, which is important so I can safely hand it off to clients.
One crucial tip for building with AI: If I have a Rise course, I can go back and edit it six months later. If we build coded simulations with AI, we risk losing that archive. So, at the end of every Claude build, I ask Claude to give me a "Handoff Document." It creates an MD file with all the code specifics. I download it and save it with the final files, so if I ever need to revise it, I just upload that document back to Claude to pick up right where we left off.
Luis Malbas:That's amazing! How did you figure that out?
Traci Synatschk:I accidentally started a build in ChatGPT and needed to move the work to Claude. I asked ChatGPT how to do it, and it offered a handoff document! Now it's a standard part of my process.
Luis Malbas:Fantastic session, Traci. Thank you so much for being a part of the event. Please post your LinkedIn and newsletter links in the chat!
Traci Synatschk:Absolutely. Thank you, everybody. I hope you dive into the kit and I'd love to hear your feedback.
Luis Malbas:Thanks, everyone. Don't forget we've got two more sessions coming up tomorrow. Goldie Love is opening, and Donna Boucher is closing out the event. We'll see you again. Thanks again, Traci!
