I Kept Asking AI the Wrong Question

Why Better Context Gets Better Answers Than Better Prompts

I typed one sentence into ChatGPT, hit enter, and got back an answer that was technically responsive and completely useless. Vague, generic, the kind of thing that could have applied to almost anyone asking almost anything remotely similar. I closed the tab, mildly annoyed, and filed it away as more proof that AI was overhyped.

A few days later, I tried again, this time with the same basic question, but I actually explained myself. What I was trying to accomplish. Where I was starting from. What constraints I was working within, what budget I had to work with, who the answer actually needed to make sense to, and when I needed it done by. The response that came back wasn't just better. It was almost unrecognizable as coming from the same tool. Nothing about the underlying AI had changed between those two conversations. What changed was how much I'd actually told it.

Why AI Isn't Reading My Mind

It sounds obvious written out like that, and yet I hadn't really internalized it until I saw the difference firsthand. AI doesn't know anything about my situation unless I say it out loud, the same way a new coworker wouldn't know the history behind a project you've been quietly circling for months. If I ask a coworker "how should I handle this," and give them nothing else to go on, I'd expect a fairly generic answer back too — not because they're incapable, but because I haven't given them anything to actually work with.

One sentence carries almost no information. "Help me write a résumé" tells a tool nothing about my industry, my experience level, the specific role I'm targeting, or what's already working versus what isn't. "Help me plan a workout" says nothing about my current fitness level, the equipment I have access to, or how much time I actually have in a week. The vague version of a question gets a vague version of an answer, every time, not because the tool is weak, but because there was never enough there to build something specific.

Why Prompt Lists Didn't Solve My Problem

For a while, I thought the fix was finding the right prompt — one of those viral lists that promises a magic phrase that unlocks dramatically better answers. I tried a few. Some helped marginally. Most didn't, because they were built around someone else's situation, not mine. A prompt engineered for a different industry, a different goal, a different constraint set doesn't automatically become useful just because it worked well for whoever originally shared it.

The problem was never really about finding clever wording. It was about the fact that a perfectly worded question with no actual context behind it is still just a vague question wearing a nicer outfit. No amount of phrasing trickery replaces the specific details that make an answer genuinely useful to my actual situation.

Where AI Actually Helped

Once I understood that context, not cleverness, was the real lever, everything shifted. This is where ChatGPT and Claude actually became useful conversation partners, rather than tools I kept quietly blaming for disappointing answers.

I started letting it ask me follow-up questions instead of demanding a complete answer from a single, underspecified request. That alone changed the quality of what came back, because the tool was finally working from an actual understanding of my situation instead of guessing at it. Within a single conversation, it remembered the constraints I'd already laid out, which meant I didn't have to keep repeating myself every time I asked a follow-up. As I gave more detail, the suggestions adapted accordingly, shifting away from generic advice and toward something that actually reflected my goal, my experience level, my budget, my deadline, my audience.

The improvement was never about discovering magic wording that unlocked a better version of the tool. It came entirely from giving it enough to actually work with. A well-informed conversation partner and a poorly informed one aren't operating with different intelligence. They're operating with different amounts of context, and that difference shows up in everything they're able to offer back.

What AI Couldn't Know

There's a real limit here worth naming honestly. AI still can't know anything I never actually tell it. It doesn't know my specific circumstances unless I describe them, and it can't infer details I've left out just because they seem obvious to me. If I leave out my budget, it has no way of knowing whether I'm working with fifty dollars or five thousand. If I leave out my deadline, it has no sense of whether I need something today or next month.

That's not a flaw exactly. It's just the nature of the tool — it works with what's actually in front of it, and nothing more. The responsibility for making sure it has enough to work with stays entirely on my end of the conversation.

The Prompt I Use

When I'm not even sure what details matter yet, this is what I use to make sure I'm not leaving anything important out.

I'm trying to solve a problem. Before answering, ask me any questions you need so you fully understand my situation. Don't assume details. Help me build the best solution based on my actual circumstances.

What Changed

I stopped hunting for the perfect prompt, the secret phrase that was supposedly going to unlock dramatically better answers. That search was never going anywhere, because it was built on the wrong assumption from the start. What actually changed things was learning to have an actual conversation — explaining my situation clearly, answering follow-up questions honestly, and giving the tool enough real context to work with instead of a single vague sentence and a quiet expectation that it should somehow already know the rest.