I want to be upfront about something before we get into this.
This is not a polished walk-through of the times AI made my work effortless and brilliant. This is the real version, including the outputs I had to rewrite significantly, the prompts that missed because I didn’t give enough context, and the moments where I had to step back in and use my actual expertise to fix something that looked right but wasn’t.
I’m sharing it this way because I think the messy version is more useful to you than the highlight reel. And because if you’re going to try this in your own workflow, I need to set appropriate expectations.
Here’s how I use AI across an ID project, from analysis through delivery prep.
Analysis: getting smarter about the problem before I design anything
This is where AI earns the most credit in my workflow, and it’s also the phase most people skip when they’re talking about AI for instructional design.
Before I write a single learning objective, I use Claude, ChatGPT, or Microsoft Copilot (it depends on the nature of the project and content I am working on) as a thinking partner on the performance problem. That looks like this:
- I typically create a new Claude Project, upload a skill file and/or instructions and pertinent files. The skill file will contain the client brand colors, brand voice, audience analysis, company information, etc. The project files and instructions include all notes I have taken in calls with the client (I use my Plaud or Fathom to record), the project proposal/statement of work we agreed to, and then any source materials that I can share with AI. Side note: I actually use AI to draft my proposal as well, based on a custom skill file I created, so the conversation may have started before this initial conversation and here I am just adding more details once I get awarded the project.
- I also upload a skill called the LLM Council. This is a skill that instead of asking a question to your favorite LLM provider, it actually asks several LLMs (think of it like your Local Multi-Model AI Advisory Board (e.g. OpenAI GPT 5.1, Google Gemini 3.0 Pro, Anthropic Claude Sonnet 4.5, xAI Grok 4, eg.c).
- I open a conversation, ask it to review my files, instructions, and skills, share (like I was talking to a colleague) what I know about the project, the stakeholder request, the audience, the performance gap I’m hypothesizing. I try to share any nuances that the files, instructions, and skills might not contain.
- I ask Claude to push back on my assumptions.
A real prompt looks something like: ‘My stakeholder wants a workshop on manager communication. Here’s what I know about the actual situation, “…”
What am I probably missing? What questions should I be asking before I make my recommendations?
The response isn’t always right. But it consistently surfaces questions I hadn’t thought to ask, and it helps me go into a stakeholder kickoff better prepared to diagnose instead of just execute.
I also use Claude to build out my SME interview question banks. I describe the performance gap, the audience, and what I most need to understand, and I ask for a full question set organized by category. Then I edit. The SME questions I end up using are never Claude’s verbatim, but they’re built faster and they’re better organized than what I would have drafted from scratch.
Honest caveat: Claude doesn’t know your organization. It doesn’t know the VP who torpedoed your last needs assessment or the culture that makes certain approaches a non-starter. You bring that. What it gives you is a thinking partner available at 10pm who will engage seriously with your problem and never tell you your idea is fine when it has a gap. These are the details I add to the Project files and instructions as well.
Design: Using AI to Pressure-test the Approach
Once I have a working hypothesis about the performance problem, I use Claude to pressure-test my design decisions before I commit to them.
This is the phase where I’m building out learning objectives, deciding on instructional approaches, and starting to think about the scenario architecture. I’ll share my draft objectives with Claude and ask: ‘Given this performance gap, what are these objectives missing? Are there any that don’t connect to the actual behavior change?’
I’ll also use it to build out design documents, especially for Rise. I create what I call a “Content Design Document (CDD)” that details the outline of content that I will create as Rise lessons and topics and how I will most likely treat that content (block types).
I also use it for scenario variations. I write the first scenario myself, because I need it to reflect the real culture and the real decision points my audience faces. Then I ask Claude to generate three more variations that cover different wrong turns, different context factors, and different consequence sets. I edit all of them, but I’m not building from nothing four times.
Because I started with a Claude Project, which gives our conversation persistent context. I share any additional files like the project brief, the performance gap analysis, the audience
profile, and any relevant organizational constraints in the skill file or instructions. Once that context is there, I don’t have to re-explain it every conversation. The quality of the collaboration improves significantly when Claude already knows what matters.
Try this early: build a project context file (either a skill file or instructions) before you start prompting for outputs. Five minutes of setup saves a lot of re-explaining, and it changes the quality of everything downstream.
Development: First Drafts, Not Final Outputs
Here’s the phase where I see the most over-reliance and the most frustration, so I want to be specific about how I use AI here.
I use Claude to write first drafts of content that I’m going to significantly edit. Not sections I’m going to lightly review. Significantly edit. The distinction matters.
For Instructor Led Training, or Virtual Instructor Led Training, I will use Canva AI to draft my slides, and then use its built in AI tool to create custom notes on every slide. If I am creating a separate facilitator guide, I will use Claude to help me draft the facilitator guide using the finalized slides, and the Facilitator Guide template I have uploaded.
Additionally, I use it for activity ideas, discussion prompts and debrief (using the xchange choreographies I am trained on), transition language, and timing notes.
Finally, I will use Canva AI to create a job aid or my workbook from my slides . I have used Claude for this as well, especially when I’m adapting dense policy language into something field-usable.
For eLearning (Rise or Storyline), I am using it to help me write custom code HTML blocks, JavaScript, or helping me brainstorm content treatment ideas. If it is Vyond, I am collaborating with Claude or Vyond AI to help me write narration scripts.
What I don’t use AI to draft: anything where the organizational context is so specific that a generic first pass would be more misleading than helpful. For those pieces, I write from scratch and use AI to review instead.
When AI gives you a draft that misses your context, don’t just fix it and move on. Tell Claude what you changed and why. ‘I rewrote this section because the original tone was too formal for our audience. Here’s what I kept and what I changed.’ That feedback improves the next output and builds a better working context over time.
Delivery Prep: Making My Design Rationale Visible
This is the phase nobody talks about in AI workflow conversations, and it’s one of the places I’ve found the most practical value.
Delivery prep, for me, means two things: preparing for the stakeholder review, and if it is ILT or VILT, making sure the facilitators or instructors who will deliver the learning are set up to succeed.
For stakeholder reviews, I’ve started using Claude to help me anticipate objections. I describe the stakeholder, the organizational context, and the design decisions I made and why, and I ask: ‘What questions or pushback is this stakeholder most likely to have? What would I need to be prepared to defend?’ The responses aren’t always right, but they’re often right enough to help me go into a better prepared review meeting.
I’ve also started using Claude to write what I think of as a design rationale one-pager: a short document that explains, in plain language, the decisions I made and the evidence behind them. It’s written for a non-L&D audience, which means it connects design choices to business outcomes. Stakeholders who don’t understand instructional design suddenly understand why a 47-slide information dump isn’t a learning solution when they can see the connection to the performance goal they care about.
For facilitators, I often give a little background on the facilitators’ skills and background and then from there create a one-page document that highlights “Things to Remember Before Delivering this Course” or “Gotchas,” things they should watch out for when delivering the content.
This is one of the clearest ways AI has helped me show up as a strategic partner instead of a production resource. And if you’ve been trying to shift that dynamic inside your own organization, it’s worth trying.
The Honest Summary
Claude is in my workflow at every stage now. Not because it does the work for me, but because it does the parts that don’t require my judgment faster, which gives me more of myself left for the parts that do.
The analysis is better because I have a thinking partner who will push back on my assumptions at any hour. The design is tighter because I pressure-test it before I commit. The development is faster because I’m editing first drafts instead of building from nothing. And my stakeholder conversations are stronger because I’ve anticipated the pushback.
None of that is magic. It’s just workflow.
If you want to see this entire process in action, not as a description but as a live build, that’s exactly what VIP members inside #IgniteLearning are getting this month.
The June VIP workshop is called Build It With Me: How I Use Claude From Design to Rise Development. We build a real course together, from source documents through a live
custom HTML block in Rise, with Claude in the room at every step. The session is recorded, so if you can’t make it live, it goes straight into the learning library.
VIP members also get the Real Work, Real Prompts prompt toolkit this month: 15 ready-to-use AI prompts organized around the exact workflow I described above, written for the in-house reality, with a ‘what to do with the response’ note for every single one.
The blog describes the workflow. The workshop shows it. The toolkit puts it in your hands.
Ready to see it live?
Join #IgniteLearning. Free 7-day VIP trial. https://zps.circle.so/checkout/ignite-learning
About Dani Watkins
Dani is the founder of #IgniteLearning and the owner of Zenith Performance Solutions. She’s an instructional designer, trainer, and eLearning developer who creates practical resources for in-house L&D professionals. She presents regularly for Training Magazine and believes deeply that good learning design changes outcomes, and that the right tools make those possible inside real organiz