Stop Trying to Get the Right Answer From AI

You know that feeling when you're going back and forth with ChatGPT? You ask something, get a response, then find yourself saying "no, more like this" or "actually, from a different angle" or even "that's not quite right, but it makes me realize..." After a few rounds of this, you start to feel like you're being indecisive or that the AI just isn't getting it.

But here's what's actually happening: you're discovering what you meant to ask.

The chat interface made AI familiar and approachable - it looks just like texting a friend. But that familiarity hides what the interface is actually good for. When you're three messages deep, refining and adjusting, you're not failing to communicate clearly. You're using the conversation to figure out what you actually want.

The goal isn't getting the AI to output the right answer. The goal is getting you to output the right question.

And once you see it that way, everything changes. You stop trying to get the AI to produce the final answer on the first or second try. Instead, you deliberately use it to generate responses that provoke you to clarify what you really want. The important information in this exchange isn't the AI's draft strategies or proposed frameworks - it's your reactions to them. "Yes, and..." or "No, but..." or "That's closer, except..."

This matters because for most knowledge work, the hard part isn't execution anymore. AI can draft strategies, write summaries, generate options. The hard part is knowing what you want in the first place - what problem are you actually trying to solve, what constraints matter, what good looks like for your specific situation. And most of us aren't naturally good at articulating all of that upfront. We figure it out by seeing something and reacting to it.

Use chat deliberately for question discovery

When you see yourself going back and forth like this, instead of asking the AI to solve your problem directly, ask it to help you figure out what you're asking. Paste something like this into the chat:

"Restate my initial question in 4 different ways. Make each interpretation as specific and as distinct from the others as possible."

Then respond to what you see: "I like the way 2 addresses the technology but also the way 3 addresses the business side. Let's zoom in based on that feedback."

Or try this:

"Define an spectrum for possible answers that captures an important aspect of the problem. Show me what extreme answers at each end would look like, plus something in the middle."

Then use that to define what you were really asking: "I want something in between the middle and the last scenarios, and more focused on the way the customer will see it."

After a few rounds of this, ask the AI to summarize what it's learned about what you actually want by restating your original question and trying to answer it.

What makes this work?

First, don't jump to solutions. When you see an initial answer as an incomplete solution that just needs fixing, you're locked into one way of framing the problem. Back up. Define the question first.

Second, recognize that infinite patience is actually one of AI's superpowers. You can (and should) do this same process with colleagues to talk through ideas. But at some point, even your most understanding coworker will say "okay, that's enough for me." With AI, it keeps going until you tell it "okay, I know what to do now."

The back-and-forth isn't a bug. It's the feature, but the feature for the specific use case of problem discovery. Focus on using it intentionally and deliberately.