Since ChatGPT launched in November 2022, I’ve been equally curious, intrigued and skeptical about what can be achieved in this magical chat window. I started by poking it with a metaphorical stick, trying to work out if it was alive or dead, friendly or hostile. Through lots of learning and experimentation I’ve now created a place for AI in my day to day. It’s my 24/7 personal assistant for life.

There’s so much I want to share about what I’ve learned, but I figured the best place to start is with how I talk to AI. Or more precisely: how I think about talking to AI before I even open the chat window.

Here’s the process I use to get in the right headspace and set myself up for the best chance of having a great conversation.

Before I prompt, I ask myself:

  1. What role do I need it to play?
  2. What context does it need?
  3. What kind of conversation do I want?

How I actually talk to AI

There’s an ever-growing pool of advice out there about prompting—and I’m always learning from it. For example, frameworks like RACE are well worth exploring, and OpenAI and Google have both recently released guides on prompting.

But I’ve found that trying to craft the perfect prompt can become a barrier of its own. What’s worked better for me is running through a few simple questions that help me frame the interaction before I start typing.

What role do I need it to play?

I don’t mean a detailed persona (though that might come later). I’m talking about the type of interaction I need right now. Am I looking for:

  • A task helper – to answer a question, review code, or tidy up some text
  • A thought partner – to explore ideas I haven’t fully formed yet
  • An expert – to advise me in an area I don’t know much about

Just having a mental label for the role I need sets the tone and helps me frame my opening message.

What context does it need?

Sometimes, it really does feel like AI just “gets” me. And with memory features improving, this ability will only become more powerful. But I’ve learned not to rely on that. Because even when an AI has years of memory about your style, your projects, or your preferences, it still doesn’t automatically know what matters for this specific situation, in this specific moment.

I always try to ask:

What does it need to know to be helpful right now?

If I’m asking for expert advice, context is essential—what kind of expertise I need, what my goal is, what I’ve already tried. For simpler tasks or thought partnering, something a bit more brief might be enough.

Once I’ve started the conversation, if the response feels off, it’s usually because the AI doesn’t have the full picture. Large language models rarely ask clarifying questions (unless specifically instructed to do so), so they often make incorrect assumptions.

Fortunately, I can usually just layer in context as I go to improve the outcome. But even if things have drifted too far off track, I’ve usually gained enough insight to write a much better starting prompt. So I just open a new window and start again.

What kind of conversation do I want?

This one’s trickier to pin down, because it’s more about feel than mechanics. But, before I start typing, I try to imagine the kind of conversation I want to have:

  • Do I want empathy or challenge?
  • Am I looking for fast answers or careful thinking?
  • Should it push back, or just help me clarify?

This helps me provide better instructions in that initial prompt, or get the conversation back on track if it starts to drift. For example, if it’s being too agreeable, I’ll ask it to be more critical. If it’s giving me information I can’t make sense of, I’ll clarify what kind of outcome I’m looking for.

Bonus Tip: Pacing

I’ve also found that pacing the conversation (i.e. carefully considering the scope of each question I ask) makes a big difference—especially when I’m working through something complex. Instead of diving straight into “give me the solution,” I’ll ask exploratory questions first. This helps flesh out the context together, and often leads to a better outcome than jumping to an answer too soon.

Examples

So, what does this look like in action? Here are some actual prompts that I’ve used with ChatGPT

A simple question with just enough context

Can you help me summarise the different game modes in NBA 2K25? I want to understand which one lets me play as just one team through a season.

Why it worked

I didn’t need a long background—just a concise question with a little extra context (“as one team through a season”). That was enough to guide the answer and avoid a generic list of game modes.

Establishing an expert thought partner, then switching to voice mode

This one was a two step process. I had started a conversation that actually resulted in me deciding to start this website. At the end, I said this.

Can you please create a prompt that I can use to start a new conversation with a gpt that will help me clarify this vision and plan a post that provides value to my target audience

The result was a VERY in depth summary. I won’t share the entire thing, just the basic structure.

My Background and Context:
...

My Goals for this Blog:
...

My Audience:
...

Your Task (ChatGPT):
Help me:

- Define a clear, engaging, and authentic angle that aligns with my goals and resonates with my target audience.
...

At the time, I didn’t have specific posts in mind, just some vague ideas I needed to better articulate. So, just before leaving work for my commute home, I pasted the above with the addition of the following line…

Please dont respond, just wait for me to switch to voice mode so we can have a conversation about this

Then, I proceeded to dump all of my random thoughts verbally until some clear ideas started to form.

Using pacing to step through something complex

As part of a marketing campaign for my business, Freely, my wife (and Freely co-founder) was in the process of creating a freebie and I wanted to start thinking about how we approach the marketing. I started with this…

We are currently developing a print at home card game... What are your thoughts on how to market this freebie ie audience, creative, retargeting etc

It gave me an extremely detailed marketing plan with a lot of assumptions built in. So I said this in order to slow the conversation down and start stepping through the points one at a time…

Let’s just focus on the creative... How should we grab their attention? Maybe just show the game being played to capture curiosity?

When prompting matters

Obviously, better prompting = better results. But, for general usage, I can usually get a good outcome through conversation, rather than dwelling on prompting perfection.

However, once I establish an ongoing need for a certain type of interaction with AI, that’s when it’s worth putting the effort into crafting a great prompt. That way, I can copy and paste or set up a Custom GPT, to get straight into a specific type of interaction quickly.

If you’ve got questions, ideas, or just want to compare notes, I’d love to hear from you! Feel free to reach out on LinkedIn or via email.