I am Mike Meyers and I just published my first custom GPT—a learning advisor designed to guide you through a streamlined version of a process I’ve relied on for over a decade to shape meaningful, action-oriented learning experiences.
The Story First Design Learning Advisor GPT blends Cathy Moore’s action mapping framework—which starts with a measurable goal and works backward to identify the key actions learners need to take—with a storytelling layer that uncovers the deeper context: the motivations, tensions, and consequences surrounding those actions.
This combined approach doesn’t just identify what people need to do, it helps clarify why it matters and how it plays out in real life. The result is a learning experience that’s both practical and meaningful. After about 10 minutes working with this learning advisor, you’ll have a macro design anchored in a realistic story tailored to your audience.
Wayan built custom GPTs and failed at it too!
I came into this project with a solid foundation in generative AI, but I still picked up a few lessons the hard way. If you’re thinking about building your own GPT, here are 8 practical tips that can help you get started faster and smarter.
1. Lock in your conversation starters—manually
When you create a custom GPT, the system automatically generates suggested conversation starters (those clickable buttons that initiate a chat). If you decide to change them, as I did, know this: every time you update your GPT, those starters will revert back to the original ones it suggested unless you explicitly tell the GPT to use the ones you want.
To avoid this, tell the GPT to lock in your preferred conversation starters. Add a line to the Instructions field or the Create tab that says something like: “Always use the following conversation starters and do not revert to default options after updates.”
This tells the GPT to preserve your custom starters and keeps them from resetting.
2. Rule-based logic for consistent replies
One of my conversation starters was “How does this learning advisor work?” During testing, I noticed it gave me a different answer every time. I wanted a clear, consistent response. To fix this, I added a rule: “If the user selects this conversation starter, respond with this exact message.” This locked in the answer using rule-based logic.
3. Watch the 8,000-character limit
After hours of refining my instructions in the Create tab, I suddenly couldn’t save any updates. I’d unknowingly hit the 8,000-character limit in the instructions field. I asked the GPT to shorten the text without losing key details—but unfortunately, much of the nuance we’d built together was lost.
To prevent this, add a rule up front: “Never exceed 8,000 characters in the instructions field when iterating.” It’s a simple safeguard that can save hours of work.
4. Prioritize publicly accessible content
During testing, some users had trouble getting the GPT to respond properly. In troubleshooting, I discovered that GPTs can struggle to reliably reference uploaded documents. When possible, train your GPT on content that’s either pasted directly into the Create tab or hosted on publicly accessible websites. It’s more stable—and easier to update.
5. Be careful how you structure templates
In one section of my GPT, I gave it a template to use for email summaries, with placeholder text like “Subject Line: Insert subject line here.” During testing, the GPT delivered emails that literally said “Insert subject line here.” To avoid this, add a prompt like: “Replace all placeholders with relevant, context-specific content. Do not display template instructions or placeholder text in the final output.”
6. Don’t let your GPT overpromise
Some of my pilot testers were delighted when the GPT offered to create a PowerPoint summary of their conversation—until it confessed that it couldn’t actually do that. While ChatGPT can generate downloadable content, custom GPTs currently can’t.
To avoid confusion, tell your GPT not to offer downloadable outputs. Instead, guide it to present information in a copy/paste–friendly format. In my case, I had it summarize the session in an email-ready format and encouraged the user to send it to me or another learning advisor.
7. Just start—and iterate as you go
If this all feels overwhelming, try what worked for me: pay for a month of ChatGPT Plus, create a custom GPT, and use the “Create” tab to tell it what you’re trying to achieve. Then, just start iterating. Getting to a usable product quickly taught me more than any tutorial ever could and gave me the momentum to keep going.
That said, there are a ton of resources you can lean on before you get started and along the way. There are even GPTs available on the GPT store to help you build your own. Spend 30 minutes with one of those and you’ll walk away with a strong draft of your instructions.
8. Build a GPT around a process you know well
Ask yourself: What process do I know inside and out—one others often struggle with? For me, it was blending action mapping with storytelling to design engaging learning experiences.
I trained the GPT using publicly available resources on action mapping and the hero’s journey, then layered on my own approach for combining the two. The result was a guided experience to help designers draft an action-oriented, story-driven macro design.
Now it’s your turn.
What’s a process do you know inside and out that others could benefit from? What sources, steps, and frameworks shape that process? If there’s no custom GPT that does it yet, go build it. And if you learn lessons along the way, please share!
