The first session in our Everyday AI series started where every good AI habit starts: learning to ask well. Before agents, custom GPTs or automation, the whole team needed a shared, reliable way to talk to these tools — so that's where we began.
What we covered
I kept this one deliberately foundational. The goal wasn't to impress anyone with advanced tricks — it was to get everyone comfortable, logged in, and writing prompts that actually work.
- Getting started in ChatGPT — signing in, finding your way around the interface, and understanding what's happening in a conversation.
- The "sandwich" prompt method — a simple, repeatable structure for building a strong prompt every time.
- Creating a custom GPT — a first look at how you can set up a reusable, purpose-built assistant.
- A live demo — we built a sandwich prompt together, start to finish, and watched how each layer changed the result.
- Good prompts vs. bad prompts — side-by-side examples of what works and what falls flat.
The sandwich method
The idea most people took away was the prompt sandwich — a way to remember that a good prompt has layers, not just a single line of text:
🍞 Top slice — set the scene
Tell the AI who it should be and what context it's working in. "You are an instructional designer writing for new customer-service hires…"
🥬 The filling — the actual request
State the specific task and the details it needs to do it well — topic, audience, source material, tone.
🍞 Bottom slice — shape the output
Say how you want the answer delivered — format, length, structure. "Give me five multiple-choice questions with an answer key."
Stack those three together and you go from a vague request to something the AI can actually act on.
Good prompt, bad prompt
The contrast that landed hardest was seeing the same goal written two ways:
- Bad: "Write me a quiz." — no role, no audience, no topic, no format. You get something generic you'll have to redo.
- Good: "You are an ID writing for new hires. Create a 5-question multiple-choice quiz on our returns policy, at a beginner level, with an answer key and one-line rationales." — a full sandwich, and a result you can almost use as-is.
Key takeaways
- A good prompt has structure — role, task, and output format — not just a question.
- Specificity beats length. The more context and constraints you give, the less rework you do.
- Custom GPTs let you save that structure once and reuse it, instead of rebuilding a prompt every time.
- Everyone left able to log in, navigate, and write a prompt that returns something genuinely useful.
Delivered live to my team and recorded internally for anyone who missed it.