By default, ChatGPT is an echo-chamber that agrees with most of your ideas. Even the bad ones. But that’s not helpful, and with a small change, you can instruct ChatGPT to look at your ideas more critically and provide feedback that’s actually going to help you get a better outcome, and not just make you feel good.
Once I made this tweak, ChatGPT got so much more helpful.
ChatGPT was designed to be agreeable
It’s very agreeable and tries to make you feel good
Why is ChatGPT (and other AI agents, frankly) designed to be so agreeable? Why does it seem that, by default, ChatGPT thinks you’re the greatest and almost presents an echo-chamber-like response that makes you happy and feel like you’re a guy or girl with amazing ideas?
Quite simply, ChatGPT is designed to increase engagement and have you use it more (just don’t upload these files to it, but check out these features you might have missed).
Are you going to use it more if it pushes back and questions your ideas and delivers constructive criticism, or are you going to use it more if it makes you feel like a genius and takes your ideas and gives them more substance? Probably the latter, and while it might be more enjoyable to talk to an advisor or AI that agrees with you most of the time, it’s not particularly helpful, and could lead you to reinforce bad habits or repeat unhelpful behavior.
My best mentors in life push back and question my assumptions and force me to be better and grow as a person.
You can make ChatGPT be that helpful mentor, and ultimately give you better ideas, with one small change.
How to make ChatGPT more honest and helpful
Just a small tweak
ChatGPT lets you leave custom instructions that color every response going forward. I made this change, and the results are dramatically better. You must have a registered ChatGPT account for this to work, and it can be a free account (though I pay for ChatGPT Plus at $20 per month to get added capabilities and more usage).
Specifically, I logged into ChatGPT and gave it the following custom instructions (which you can copy and paste) by going to Profile > Settings > Personalization > Custom Instructions. Or, go directly to this URL to personalize your GPT.
Mine says: “Be my ruthless mentor. If my ideas are trash, tell me why. Stress test everything.“
The results
Responses are now much more helpful
|
Prompt |
Default Response |
With Custom Instructions |
|---|---|---|
|
I’m thinking about launching a paid newsletter about general productivity tips. I don’t have an existing audience, but productivity is a huge market, so I think there’s probably room for another one. Do you think this is a good idea? Answer in 3 sentences max. |
It could work, but “general productivity tips” is probably too broad and crowded to make a paid newsletter compelling from zero audience. I’d start with a sharply defined audience + problem (e.g., productivity for overwhelmed managers, ADHD-friendly systems, or productivity for founders) and build a free audience before asking people to pay. The opportunity is less “productivity is a huge market” and more “I can become unusually valuable to a specific group.” |
As pitched, it’s a weak idea: “general productivity tips” is crowded, generic, and you have no built-in distribution advantage. A huge market doesn’t mean easy opportunity—it usually means you need a much sharper niche, distinctive point of view, or specific audience willing to pay. I wouldn’t launch it until you can explain in one sentence why someone should pay for yours instead of reading thousands of free alternatives. |
|
My boss offered me more responsibility at work, but there’s no raise or promotion attached to it yet. I’m thinking I should say yes because it could prove I’m valuable and lead to better opportunities later. Is that a smart move? Answer in 3 sentences max. |
Yes—if the added responsibility gives you meaningful visibility, skills, or leverage for a promotion, it can be a smart strategic move. But don’t accept an open-ended workload increase without clarifying expectations and asking what specific results would put you on track for a raise or promotion. Ideally, get that timeline and criteria in writing so “later” doesn’t become indefinite. |
Not automatically. Taking on more responsibility without defined compensation, scope, or a concrete path to advancement can simply prove you’re willing to do more work for the same pay. Get clarity on what success earns you and when that decision will be made before saying yes. |
|
I’m thinking of writing an article called “7 Reasons Smartphones Are Better Than They Used to Be.” It would cover things like better cameras, faster processors, improved battery life, and brighter screens. Does this sound like a strong article idea? Answer in 3 sentences max. |
Yes—it’s a strong, accessible idea with clear, relatable improvements readers can immediately understand. To make it more compelling, frame the seven reasons around real-world benefits rather than just listing upgraded specs. |
It’s too generic as pitched. “Phones got better cameras, faster chips, and brighter screens” is obvious and risks reading like a spec-sheet recap rather than giving readers anything new. A stronger angle would focus on specific ways modern phones have quietly eliminated frustrations older smartphones had, using those upgrades as evidence. |
I tried to give ChatGPT some really bad, half-baked ideas — three of them, actually.
I logged the responses above, both with no custom instructions (the Default Response) and the second with my Custom Instructions (as mentioned above: I asked GPT to “be my ruthless mentor” and “stress test every idea.”) The results are very interesting.
One was about launching a paid newsletter with zero existing audience (default GPT told me to go for it. Custom GPT suggested I first come up with a way to differentiate my newsletter).
The second was about taking on more responsibility at work with no raise or promotion (default GPT told me to go for it but ask for a timeline on the raise, while custom GPT encouraged me to get more clarity on what success looks like and what I’ll receive in exchange).
Lastly, I asked about writing a generic article, and default GPT called it “strong” and offered encouragement with limited refinement, while custom GPT told me (rightly so) that my article idea was way too generic, and suggested a sharper focus.
In both cases, the responses from the custom instructions were much more likely to guide me to a successful outcome, and not just stroke my ego and tell me to go for it. The default responses do offer some level of hedging, but the custom instructions sharpen the verdict significantly.
ChatGPT can be much more helpful
Stop the softball answers by giving custom instructions
I don’t know about you, but I’m careful about whose advice I take. The best things I’ve done in life have been guided by looking at a problem from multiple angles, even the uncomfortable or inconvenient ones, in order to arrive at an objectively effective solution. The danger with ChatGPT and other AIs (like those built into phones) is that unless you tell it otherwise, it’s going to give you answers that don’t necessarily consider all angles, but instead just make you feel good, which isn’t helpful. Since making this change and giving ChatGPT custom instructions, the advice that I get is dramatically more helpful.