132 | Engineering Conversations: Moving Beyond the Perfect Prompt

Image courtesy ChatGPT 4o

Imagine spending half an hour crafting a complex prompt for ChatGPT, following a popular and well-respected prompting framework. You rework each section (based on the acronym-shortened approach), sweating over whether you’ve provided enough context and captured the audience fully. You hit send, expecting brilliance, but the response is disappointingly ordinary.

Here's an idea: What if, instead of meticulously engineering every detail, you approach the task more naturally, as if you were going to have a chat with a colleague or friend?

In my experience, regularly chatting with AI—often just by talking to it—reveals that simply starting a conversation can lead to richer, more insightful results than elaborate instructions ever could.

The Prompt Engineering Trap

Since ChatGPT launched, we've been drowning in prompt engineering content. Everyone has an acronym: CLEAR, STAR, RACE—pick your favorite flavor of structured instruction-giving¹. There are entire websites dedicated to turning your messy human thoughts into "optimized prompts," and LinkedIn is full of posts about the "secret sauce" to AI communication.

Don't get me wrong: this approach has its place. When I'm using something like Deep Research to generate a comprehensive report, I absolutely want a well-structured, detailed prompt before the AI chatbot invests 10-20 minutes or so working on my request. And when I need specific formatting or structured data, traditional prompting works beautifully, and the time I spend carefully crafting my instructions pays off.

But for the vast majority of how we actually use AI day-to-day? All that engineering is overkill. Worse, it builds a significant "cognitive burden"—we're learning to speak computer instead of teaching computers to speak human.

The Simple Truth Most People Miss

Here's what I've learned: there are really three types of AI interactions, and each needs a different approach:

1. Simple questions → Just ask them. No engineering needed.

2. Complex exploration → Start a conversation.

3. Specific structured outputs → Traditional prompting has its place.

The problem is we've been treating everything like category 3 when most of what we need falls into category 2!

What is Conversation Engineering?

While everyone's been obsessing over prompt engineering, there's been surprisingly little written about what I think is far more important: conversation engineering. The concept comes from the field of conversational AI design, where researchers focus on "the holistic discipline that integrates technical implementation with human-centric principles of dialogue."

Let me translate that from academic-speak:

Instead of making us adapt to how the AI wants to be talked to, we figure out how to make the AI adapt to how we naturally want to communicate.

It's a shift from user-as-programmer to user-as-collaborator. Instead of front-loading all your cognitive effort into crafting the perfect instruction, you distribute that effort across an actual dialogue where both you and the AI build understanding together.

The Mindset Shift: Dialogue vs. Command

The difference is profound, and it starts in your head before you even touch the keyboard.

Traditional computer interaction works like this: You think through exactly what you want, formulate a precise command, send it, get a response, and you're done. You're essentially programming the machine. All the mental work happens upfront, and if you didn't think of everything, too bad—you get what you asked for, not what you needed.

Conversational approach works more like talking to a colleague: You start with your general direction, share some context, and then build understanding together. The AI can ask clarifying questions, explore unexpected angles, and even challenge your assumptions. The mental work is distributed throughout the conversation, and AI becomes a thinking partner, not just an execution engine.

I notice this difference most when I'm using voice interfaces. When I talk to ChatGPT, it feels natural to say something like, "I'm trying to figure out our organization's approach to AI adoption, but I'm not sure what questions I should even be asking." That's not a prompt—it's the start of a conversation.

Why Conversation Works Better

The magic happens when you stop trying to control every aspect of the interaction and start collaborating instead.

For exploration and discovery, conversation is unbeatable. The AI can surface questions you hadn't considered, explore connections you might have missed, and challenge positions you've taken for granted. I was recently working through a strategic planning issue, and the AI asked me something that sent my thinking off in a very different (outward-in) perspective: "What assumptions are you making about your stakeholders that might not be true?"

For practical problem-solving, from grant research to dinner planning, the conversational approach turns AI into a thinking partner rather than just a sophisticated search engine. The AI can pivot and adapt as your needs become clearer through the conversation. You don't have to know exactly what you want before you start—you can figure it out together.

Practical Conversation Engineering Techniques

So how do you actually do this? Here are the approaches I've come to employ:

Start with context, not commands. Instead of "Generate a list of potential funders for nonprofit AI initiatives," try "I'm exploring funding options for our nonprofit's AI pilot project, but I'm not sure where to start looking." The second approach invites dialogue; the first demands execution.

Leave room for and encourage the AI to ask questions. One of my favorite conversation starters is to add, just before I hit send: "What other information would help you give me better guidance?" or "What questions should I be asking that I haven't thought of?"

Use follow-up questions effectively. Don't just take the first response and run with it. Ask "What else should I consider?" or "How does this connect to what we discussed earlier?" or my personal favorite, "Challenge this assumption I'm making."

Guide without controlling. Instead of trying to script the entire interaction upfront, direct the conversation as it unfolds. Let the AI's questions and perspectives inform where you go next.

When to Choose Which Approach

I'm not suggesting we abandon structured prompting entirely. Like any tool, both approaches have their sweet spots.

Conversation Engineering shines when you're:

• Exploring new topics or complex problems

• Brainstorming and doing creative work

• Learning about unfamiliar subjects

• Making complex decisions

• Needing help figuring out what your goals are

Traditional Prompting still works best for:

• Specific formatting requirements

• Structured data extraction

• Translation or summarization tasks

• Getting a simple piece of information/insight (i.e., a replacement for Google or Wikipedia)

The Voice Interface Advantage

If you haven’t done this yet, step away from this article and try it now:

Talk to your AI instead of typing.

I know it feels weird at first—I felt ridiculous the first time I spoke to ChatGPT on my phone. But voice interfaces naturally encourage conversational interaction in a way that typing doesn't, and they’re improving massively, such that a “conversation” with ChatGPT today is amazingly close to one with a human (and maybe more fluid than with most non-native English speakers).

When you're typing, there's pressure to get everything right in one shot. When you're talking, it feels natural to pause, think, and say "Actually, let me add something to that" or "What do you think about this angle?" The medium shapes the interaction, and voice shapes it toward dialogue. Plus, when your brain only needs to put energy into the ideas you’re wanting to communicate, rather than the “form” (wording) that they’ll take, it frees you up to focus on what’s important—then content of the exchange.

Try This Experiment

Here's what I want you to do: Pick a topic you've been wanting to explore—maybe a work challenge, a learning goal, or even just something you're curious about. Try both approaches and see what happens:

First, spend time crafting what you think is the perfect prompt. Be specific, use formatting, include constraints—do the full prompt engineering thing.

Then, start fresh and just begin a conversation. Try something like:

• "I'm trying to understand..."

• "Help me think through..."

• "What questions should I be asking about...?"

I'm willing to bet the conversation approach will surprise you. Not just with better answers, but with better questions—questions that lead you places you wouldn't have thought to go with even the most perfectly engineered prompt.

The future of AI interaction isn't about becoming better prompt engineers. It's about becoming better conversationalists. And honestly? That's a much more human skill to develop.

And of course, regardless of how you engage with generative AI, be sure that you’re mindful of the potential for hallucinations, biases, violation of data privacy, etc. As you become more comfortable using something like ChatGPT as a conversation partner, it’s even more crucial always to…

Make Good Choices

1. Popular prompt engineering frameworks include CLEAR (Concise, Logical, Explicit, Adaptive, Reflective), STAR (Situation, Task, Action, Result), and dozens of others