The Question Behind the Question: What AI Taught Me About Prompting

By Kelsey Little, Administrative Assistant - Project Management Office

Have you ever typed a frantic question into the Chat GPT or Claude and been disappointed with the result? At first, I thought ChatGPT was just bad at helping me. My questions produced generic or downright off-the-wall responses. These results did absolutely nothing for my particular combination of perfectionism and desire for instant gratification. So, I started wondering if this was a ‘me’ problem or a robot problem.

What I couldn’t have predicted is that learning how to ask better AI questions would force me to actually understand what I wanted in the first place

This might have turned into a mild existential crisis.

It turns out AI is only as useful as the information you give it, which meant I had to stop typing vague words like “help me plan a meal” and actually think through what success looked like.

What did I learn? AI exposes how often we ask incomplete questions.

Has this ever happened to you? It’s Friday, you’re heading home in less than 30 minutes and remember that you promised to bring a dish to pass at your friend’s party TONIGHT. Some people plan ahead, some people remember right before they leave work while they’re having their afternoon bite of cheese (ok, fine, I’m “some people”).

In your panic, you turn to AI and type:

“Give me an appetizer recipe.”

ChatGPT enthusiastically suggests a complicated baked dip that requires forty-five minutes in the oven, seventeen ingredients and what can only be described as emotional preparedness.

The party is outdoors, you don’t have much time to shop and prep your dish, and you really don’t want to ask the host to borrow their oven. Also, seventeen ingredients -- grocery prices are currently testing everyone’s character development right now.

So instead, let's try again with this prompt:

“I need an appetizer for an outdoor party with 12–15 people. It should cost less than $25, take under 20 minutes to prepare and have a pop of color. Give me two options, an estimated cost breakdown and a grocery list organized for shopping at Meijer so that I can shop as quickly as possible.”

Suddenly the results are weirdly fantastic.

ChatGPT suggests:

Watermelon, feta & mint bites

Strawberry tomato bruschetta

The watermelon bites win because they sound effortlessly fancy while still requiring almost no work, (which is honestly my ideal personality trait in recipe form).

AI isn’t actually good at reading minds -- it’s good at responding to details. The clearer you are about what you want, the better the results become.

I’ve learned that my most successful prompts include four things:

A Goal: What do I actually want?
Context: Who is this for and why?
A Role: What role do I want AI to play?
Output: How do I want the information to be organized?

I’ve started carrying this prompting philosophy into areas of my life that have nothing to do with AI.

I think about it when considering long-term career goals, planning vacations, saving money or preparing for difficult conversations.

I don’t use AI in those situations, but I do ask myself the same questions:
What do I actually want here?
What outcome am I hoping for?
What matters most?

Really understanding what your goals are at their core helps you remove that vague overwhelm of, I don’t know what I want, but I know it isn’t this.

A few AI specific lessons I’ve also learned:

AI confidence ≠ accuracy.

Specificity improves relevance, not truthfulness.

Users still need judgment and verification.

AI can be wrong in a way that feels almost inspirational. Interrogate and verify the information it presents. I swear, these chatbots are so confident – hilariously so at times. I’ve had AI recommend local restaurants that are made up (a waffle/taco fusion restaurant is too good to be true, sadly). I once asked AI for book recommendations, and it confidently suggested a title that did not exist. Which honestly felt extra rude as a library worker.

Asking incomplete questions isn’t a new problem for us humans. The funny thing is that library workers have been helping people do this forever. Sometimes the hardest part of finding answers isn’t finding information — it’s figuring out what question you’re really trying to ask.

As it turns out, that’s useful advice whether you’re using AI, searching for library resources or just trying to figure out what kind of appetizer to bring to a party you forgot about until the last minute.