What AI Asks You Is More Useful Than What It Answers

Amy Westlake
Jun 30
3 min read

The Situation

I had a long prompt loaded in ChatGPT. I'd been careful with it. Then I added one line at the end, something I'd picked up from a creator I follow.

Ask me 5 questions that will improve the response you'll give me.

I learned this from Jeff Su, whose work on AI productivity is worth following closely. The technique is simple: at the end of any prompt, add that line.

That's it. That's the whole thing.

The AI Move

The first time I used it, I was building a prompt for a ChatGPT session. I'd given it a lot of context: my role, my programs, the company I was working with, what I was trying to get out of it. I asked for 10 prompts that would help me be more productive, delivered at three levels: basic, complete, and creative.

Then I added the line.

AI tools want to answer. It's their default state. Give them a prompt, they start generating. Adding this one statement forces a break in that reflex. Instead of an immediate response, what came back was a set of questions.

Some of them I expected. A few I didn't.

There was one I don't remember the exact wording of, but when I read it, I thought: that's a really good question. I hadn't even considered that. And when I answered it, I added context I'd forgotten to include in the original prompt. Context that changed what came back.

The output had more depth. More specificity. It was closer to what I actually needed.

The Shift

There are two things this technique does, and they're different.

The first is obvious once you've seen it: it creates space. AI processes fast. Faster than the part of your brain that's still loading everything in. The pause gives you a moment to catch what you meant to include and didn't. You answer the questions and think, oh yeah, I also should have mentioned that. Then you add it.

Your answers don't need to be long. Sometimes a few words is enough. A short answer still gives AI more context than it had before.

The second thing is subtler. The questions themselves are sometimes better than what you would have asked. There's a difference between questions you anticipated and questions that surface something you hadn't thought to examine. When AI asks a question that stops you, when you genuinely have to think before answering, that's the part worth paying attention to.

A few weeks after that first experiment, I used the same technique to draft a video script for a program update. I loaded in all my raw notes: messy, dense, unstructured. Added the line. Answered the questions. And when the script came back, it was right on the first try. I tweaked maybe four or five sentences across the entire thing. I was honestly surprised. The questions had already done the work of clarifying what mattered.

The Pattern

Most people treat AI like a search engine with better sentences. You type in a question, you get an answer, you decide if it's useful.

The "ask me 5 questions" prompt changes the sequence. You type in a question, AI asks you questions, you answer those, and then you get a response. It's slower by maybe two minutes. But the quality of what comes back reflects all the context that the back-and-forth forced you to surface. And it reduces the follow-up exchanges you'd otherwise have. Instead of getting a generic response and then asking a follow-up, and then another, the questions front-load that exchange. Two minutes upfront instead of ten going back and forth.

The questions don't just improve AI's response. They improve the prompt you thought you'd already written.

The Implication

At the end of your next prompt, a draft you care about, a plan you're trying to build out, something where a generic response would be useless, add this:

Ask me 5 questions that will improve the response you'll give me.

Answer them. Include anything you realize you forgot. Then see what comes back.

Tip: Turn on your microphone and respond verbally. You'll answer faster, and when you're not trying to be concise, the answers tend to be richer.

The questions will tell you more than the answer would have.

What I'm Testing Next

I've started building this into Gems so it runs automatically, no remembering required. That's a different post. But if you want to try it before setting anything up, you don't need a Gem. You just need the line.