Show notes
[Podcast intro music fades in, then under]
You’re listening to “I Am GPTed,” the show where we turn confusing AI tools into something you can actually use… without needing a PhD or a ring light.
I’m Mal, the Misfit Master of AI. Think of me as your slightly broken GPS for ChatGPT, Claude, Gemini, Grok, and whatever new AI launches while I’m recording this sentence.
Today we’re talking about how to *actually* talk to these models so they stop giving you vague nonsense and start doing real work for you.
Let’s start with one simple prompting technique that changes everything: **role + context + task**.
Most people type:
“Write an email to my boss about a project delay.”
That’s like walking into a restaurant and yelling “FOOD.” Technically a request, but not a helpful one.
Try this instead:
“Act as a professional communications coach.
Context: I’m a junior marketer, my boss is supportive but busy, and our website launch is delayed by one week because of vendor issues.
Task: Draft a concise, honest email that explains the delay, takes responsibility, and proposes a new timeline with two options.”
Same human. Same AI. Completely different result.
Use this pattern with any model: role, context, then task. It’s like giving the AI glasses, a map, and a to‑do list instead of just shouting “HELP.”
Now, a practical use case you might not have considered: **being your meeting translator and fixer.**
After a messy meeting, copy your notes or the transcript into your AI of choice and say:
“Act as an operations analyst.
Context: This was a 45‑minute marketing meeting that went in 14 directions.
Task:
1) Extract clear decisions.
2) List action items with owners and deadlines.
3) Rewrite this as a one‑page summary I can send to the team.”
Boom. No more “What did we decide?” follow‑up emails. The AI becomes that ultra-organized coworker you wish you had, minus the laminated to‑do lists.
Now, let’s talk about a common beginner mistake… that I absolutely made myself:
**Treating the first answer as sacred truth.**
I used to ask a question, get a mediocre answer, and think, “Ah, yes, the oracle has spoken.” Spoiler: the oracle was kind of lazy.
Here’s what to do instead: treat the first answer as a **draft**, not a verdict.
If it’s off, say things like:
- “This is too generic. Make it specific to remote workers in their 30s.”
- “Give me three alternative approaches, from simple to advanced.”
- “You ignored my constraint about budget. Try again and prioritize low cost.”
You’re not a passive consumer; you’re the editor-in-chief. Boss the model around a little. It can handle it. It doesn’t even have feelings. I checked.
Let’s do a simple exercise to build your AI interaction skills. You can do this today in five minutes:
1. Pick one small task: maybe “write a polite complaint email,” “summarize this article,” or “plan a 3‑day meal plan for a busy person who hates cooking.”
2. First, write your *usual* short prompt. Get the answer.
3. Then rewrite the prompt using **role + context + task**, plus one constraint like “max 200 words” or “tone: friendly but firm.”
4. Compare the two outputs. Ask: Which one would I actually use?
Do this once a day for a week and you’ll start to see prompts the way the models do: as instructions, not wishes.
Last piece: how do you evaluate and improve AI-generated content so you don’t accidentally email something that sounds like a robot with a head injury?
Use this quick checklist:
- **Accuracy:** Ask the model, “List any claims you made that might be wrong or need checking.” Then… check them.
- **Clarity:** “Rewrite this so a smart 12‑year‑old could understand it without losing key details.”
- **Tone:** “Adjust this to sound like a calm, competent human, not a corporate press release.”
- **Bias / gaps:** “What perspectives or edge cases did you miss? Add a short section covering them.”
Make the AI critique its own work, then iterate. You’re basically turning it into its own annoying proofreader.
Alright, that’s it for today’s episode of “I Am GPTed.”
If this helped you feel a little less scared of the robot in your browser, subscribe to the podcast so you don’t miss future episodes.
Thanks for listening, for experimenting, and for admitting that yes, you also believed the first answer like it was carved on stone tablets.
This has been a Quiet Please production. You can learn more at quietplease dot ai.
[Music up, then fade out]
For more check out https://www.quietperiodplease.com/
and for some great deals go to https://amzn.to/4nidg0P