If you’re here from Amy’s podcast, welcome! I’m Becca Harpain, a Squarespace designer turned educator, and I recently launched Custom Codey, an custom GPT trained on my years of coding experience. When I launched this as a digital product, I had over 700 users sign up in the first few weeks... but let me tell you, it didn’t start out that smooth. 😅

I spent 6 months going down the wrong rabbit holes, and only 30 days doing what actually worked. So I created this roadmap to help you skip the overwhelm and build your own Custom GPT with confidence.

How I built and launched my Custom GPT in 30 days

✅ Step 1: Make Sure AI Is Actually the Right Fit

Before you dive into building anything, get crystal clear on whether your idea actually needs intelligence. Some tools are better off as quizzes or calculators; AI isn’t always the answer.

Key Questions To Ask Yourself:

  • Does this require real-time interpretation or feedback?

  • Would a flowchart or form-based tool solve the same problem?

  • Will users ask questions in unpredictable ways?

Your Action Step:
Write out 3–5 real problems your tool will solve. For each one, decide: Does this require a smart response that evolves in a conversation? If yes, you're AI-ready.

⚠️ Step 2: Don’t Overbuild (Like I Did)

I wasted months trying to host a private GPT on my own servers, hire a dev, build a login system, and analyze feedback... it was a mess. You probably don’t need all of that. 😬

Key Questions:

  • Do you need enterprise-level security and analytics?

  • Will a basic “anyone with the link” setup be enough?

  • Are you trying to build a product or an experience?

Action Step:
Start with OpenAI’s built-in Custom GPT tool. Unless your audience needs logins or monthly billing, the built-in tools are more than enough. If you do need something more advanced, check out some of the new AI tools out there, like Lovable.

📂 Step 3: Curate Your Training Data

Your GPT is only as smart as what you feed it. I used real questions, real answers, and real language from my tutorials, not just polished teaching content or AI generated info that wouldn’t resonate with my community of Squarespacers.

Key Questions:

  • What are the 20 most common questions you get asked?

  • Do you already have transcripts, docs, blog posts, or course material?

  • How do you want your GPT to sound?

Action Step:
This is more of an action dance than a step 😅 Here is an overview of what you’ll likely need to train your custom GPT.

  • Gather 10–20 sample Q&As.

  • At least 3 examples of how you write/talk

  • Glossary or cheat sheet of terms from your niche

  • Optional: tutorial transcripts, blog posts, code examples, etc.

🧠 Step 4: Train + Test with Real Questions

Once you upload your files, don’t assume it “just works.” Test it with real, messy questions—especially the kind your audience asks when they’re confused or stuck.

Key Questions:

  • Can it handle beginner phrasing and expert language?

  • How does it respond to typos or vague requests?

  • Can it explain things clearly in your voice?

Action Step:
Ask your GPT at least 25 real user questions from comments, emails, DMs, etc. Don’t clean them up; use their actual wording. It’s improtant to leave leave as much of the natural dialog as you can. Tweak or retrain if the answers aren’t solid & remember that you are supposed to chat with ChatGPT! Provide feedback in a conversational style.

🚀 Step 5: Launch Before You’re Ready

Once I had over 100 good answers in a row, I launched. And yes, I was terrified. But perfection wasn’t the goal: usefulness was. And my community responded with so much love and support! They knew it wouldn’t be perfect because I was as up front about that as possible. AI makes mistakes and Custom Codey wont always get it right, but the only way it would get better was in real conversations.

Key Questions:

  • Is it answering questions with 80% accuracy or better?

  • Have you clearly explained that it’s not always right, but still useful?

  • Do you trust your audience to grow with you?

Action Step:
Set your GPT sharing to “Anyone with the link.” and definitely be transparent about it’s abilities with a friendly disclaimer like: This is powered by AI and still learning. It may not always be perfect, but it’s here to help! Then hit publish. 🎉

🔁 Step 6: Keep Getting Better

AI tools aren’t “set it and forget it.” I still chat with Custom Codey every single weekend to review how it’s doing, refine its responses, and train it on new patterns.

Key Questions:

  • How will you track what’s working (and what’s not)?

  • Can you make time each week or month to train your GPT?

  • Are there topics your audience struggles with most?

Action Step:
Once a week, take real client questions from emails, comments, and DMs and test them in your custom GPT. Refine as needed and keep it improving 💪

💬 Let’s Keep the Conversation Going

If you’re curious how Custom Codey’s evolving, want to see behind-the-scenes of my new AI-powered business model, or just want to nerd out about GPTs, come hang out with me on Instagram!

I share updates, experiments, and the real journey over at @thinkinsidethesquare.

Feel free to DM me any time if you’re building your own GPT; I’d love to cheer you on. ❤️