For the longest time, ChatGPT felt useful but kind of flat to me. I’d ask it something, get a clean answer back, and then realize a week later that anyone could have typed the same question and gotten the same response. Nothing wrong with the answer. It just wasn’t worth much.
The thing nobody tells you is that ChatGPT is built to give you the safe, middle of the road response by default. It has read a mountain of text and it hands you the average of all of it. That’s perfect when you’re learning something. It’s a problem when you’re trying to make a real decision or do something better than the next person.
So I spent a while figuring out how to get it off that default. Not with magic words or some 500 word mega prompt, but with small tweaks to how I ask. These are the ones I actually keep coming back to.
The One I Use Most: Make It Give Me Two Answers
This is the simplest trick and probably the one I lean on every single day. Instead of taking the first thing ChatGPT says, I ask it for two answers side by side.
The first is the common answer most people would land on. The second is what someone who actually knows the field would say, the part most people miss. Just seeing those two next to each other usually shows me something I would have walked right past.
The trick is being picky about that second answer. I don’t want a vague “expert perspective.” I want something I can do something with.
Give me two answers to this. First, the most common answer most people would land on. Then the answer someone with real expertise would give that most people miss. Make the second one specific and something I can actually act on, not general advice.
I paste my real question under that and I’m already ahead of where I’d be with a plain prompt.
When I Actually Know the Topic
Sometimes I know enough to call out a weak answer when I see one. Say I’m hiring someone for my blog work and I’ve done plenty of hiring before. I don’t need ChatGPT to teach me hiring. I need a sharp second opinion on this one decision.
The lazy version is “should I hire this person based on the job description and interview notes?” You get a pros and cons list. It’s fine and it’s also exactly what everyone else gets. Here’s what I do instead, and these stack on top of each other.
I get it to name the factors first and rank them for my situation. Not a flat list, but what a strong hiring manager actually weighs, including the stuff people skip over.
I just interviewed someone for this role. The notes and job description are below. List the six things a strong hiring manager weighs here, and include the ones most people overlook. Then tell me which single one matters most for my specific situation.
Then, if it’s a bigger call, I make ChatGPT interview me before it answers. This is the one that surprised me the most. It pulls more context out of me, and more context is what makes the answer actually fit.
Before you give me an answer, ask me one question at a time. Keep going until you understand the role, my team, and what a great hire looks like in this exact situation. After six questions, give me a recommendation based only on what I told you.
It asks, I answer, we go back and forth, and the final call is built on my real situation instead of a generic template. You can paste this right under the previous prompt and run them together.
And when I really want to stress test a decision, I have it look at five angles. This isn’t just for hiring. I use the same five points for spending money, dropping a project, changing direction on a blog, basically any decision where I want it to poke holes.
Walk me through this decision across five points: the real trade off, what happens if it goes wrong, the second order effects, the mistakes most people make here, and what would have to be true for this to actually work.
Those five hold up across almost everything, which is why I saved them as a reusable prompt.
When I Have No Idea What I’m Looking At
This is the tricky situation, and it’s the one where I’m most careful. I want an expert opinion but I can’t grade the answer because I don’t know enough to spot when ChatGPT is being lazy or just wrong.
I start by asking it to flag what a beginner would miss.
Tell me what someone who really knows this field would flag for a beginner that they’d otherwise miss. Stick to non obvious things and the stuff people commonly get wrong.
Then I pair that with one more line so the answer is something I can use. Without this, ChatGPT either gives me a headline so thin I can’t act on it, or so much detail that I drown. I want the version in between.
Explain it at a level I can act on, and use one simple example to make it real.
The last piece is my safety net, and I never skip it on anything that matters. When I don’t know the topic, a confident wrong answer is dangerous because I won’t catch it. This works really well when I’ve handed ChatGPT a document to look at, like a contract or an agreement.
For every main point you give me, label it backed or inferred. Backed means you can point to a specific line or source from what I gave you. Inferred means you’re filling a gap with your own reasoning. For anything inferred, tell me what it is and one thing I should double check before I trust it.
That last bit is the whole point. Once ChatGPT tells me what it inferred, I get to decide whether to believe it. I stay in the driver’s seat instead of handing the decision over.
For decisions that really count, I do one more thing. I run the exact same question, with the exact same context, through ChatGPT twice in separate chats and compare them. If both answers line up, I trust it a little more. If they clash, I’ve found the exact spot where I need to stop and use my own head.
What I’d Tell You to Try First
If you only take one thing from this, make it the two answer prompt. It costs you nothing and it instantly bumps you out of the flat, average response that everyone else is getting.
From there it depends on the situation. If you know the topic, make ChatGPT name and rank the factors so you can judge them yourself. If you don’t, have it catch what beginners miss and label everything backed or inferred so you know what’s solid. None of this is complicated. It’s just the difference between asking ChatGPT a question and actually getting your money’s worth out of it.